Daily Papers

Interpretability methods have recently gained significant attention, particularly in the context of large language models, enabling insights into linguistic representations, error detection, and model behaviors such as hallucinations and repetitions. However, these techniques remain underexplored in automatic speech recognition (ASR), despite their potential to advance both the performance and interpretability of ASR systems. In this work, we adapt and systematically apply established interpretability methods such as logit lens, linear probing, and activation patching, to examine how acoustic and semantic information evolves across layers in ASR systems. Our experiments reveal previously unknown internal dynamics, including specific encoder-decoder interactions responsible for repetition hallucinations and semantic biases encoded deep within acoustic representations. These insights demonstrate the benefits of extending and applying interpretability techniques to speech recognition, opening promising directions for future research on improving model transparency and robustness.

This paper describes MAIA, a Multimodal Automated Interpretability Agent. MAIA is a system that uses neural models to automate neural model understanding tasks like feature interpretation and failure mode discovery. It equips a pre-trained vision-language model with a set of tools that support iterative experimentation on subcomponents of other models to explain their behavior. These include tools commonly used by human interpretability researchers: for synthesizing and editing inputs, computing maximally activating exemplars from real-world datasets, and summarizing and describing experimental results. Interpretability experiments proposed by MAIA compose these tools to describe and explain system behavior. We evaluate applications of MAIA to computer vision models. We first characterize MAIA's ability to describe (neuron-level) features in learned representations of images. Across several trained models and a novel dataset of synthetic vision neurons with paired ground-truth descriptions, MAIA produces descriptions comparable to those generated by expert human experimenters. We then show that MAIA can aid in two additional interpretability tasks: reducing sensitivity to spurious features, and automatically identifying inputs likely to be mis-classified.

Interpretability is a key challenge in fostering trust for Large Language Models (LLMs), which stems from the complexity of extracting reasoning from model's parameters. We present the Frame Representation Hypothesis, a theoretically robust framework grounded in the Linear Representation Hypothesis (LRH) to interpret and control LLMs by modeling multi-token words. Prior research explored LRH to connect LLM representations with linguistic concepts, but was limited to single token analysis. As most words are composed of several tokens, we extend LRH to multi-token words, thereby enabling usage on any textual data with thousands of concepts. To this end, we propose words can be interpreted as frames, ordered sequences of vectors that better capture token-word relationships. Then, concepts can be represented as the average of word frames sharing a common concept. We showcase these tools through Top-k Concept-Guided Decoding, which can intuitively steer text generation using concepts of choice. We verify said ideas on Llama 3.1, Gemma 2, and Phi 3 families, demonstrating gender and language biases, exposing harmful content, but also potential to remediate them, leading to safer and more transparent LLMs. Code is available at https://github.com/phvv-me/frame-representation-hypothesis.git

Interpretability and analysis (IA) research is a growing subfield within NLP with the goal of developing a deeper understanding of the behavior or inner workings of NLP systems and methods. Despite growing interest in the subfield, a commonly voiced criticism is that it lacks actionable insights and therefore has little impact on NLP. In this paper, we seek to quantify the impact of IA research on the broader field of NLP. We approach this with a mixed-methods analysis of: (1) a citation graph of 185K+ papers built from all papers published at ACL and EMNLP conferences from 2018 to 2023, and (2) a survey of 138 members of the NLP community. Our quantitative results show that IA work is well-cited outside of IA, and central in the NLP citation graph. Through qualitative analysis of survey responses and manual annotation of 556 papers, we find that NLP researchers build on findings from IA work and perceive it is important for progress in NLP, multiple subfields, and rely on its findings and terminology for their own work. Many novel methods are proposed based on IA findings and highly influenced by them, but highly influential non-IA work cites IA findings without being driven by them. We end by summarizing what is missing in IA work today and provide a call to action, to pave the way for a more impactful future of IA research.

Interpretability is the study of explaining models in understandable terms to humans. At present, interpretability is divided into two paradigms: the intrinsic paradigm, which believes that only models designed to be explained can be explained, and the post-hoc paradigm, which believes that black-box models can be explained. At the core of this debate is how each paradigm ensures its explanations are faithful, i.e., true to the model's behavior. This is important, as false but convincing explanations lead to unsupported confidence in artificial intelligence (AI), which can be dangerous. This paper's position is that we should think about new paradigms while staying vigilant regarding faithfulness. First, by examining the history of paradigms in science, we see that paradigms are constantly evolving. Then, by examining the current paradigms, we can understand their underlying beliefs, the value they bring, and their limitations. Finally, this paper presents 3 emerging paradigms for interpretability. The first paradigm designs models such that faithfulness can be easily measured. Another optimizes models such that explanations become faithful. The last paradigm proposes to develop models that produce both a prediction and an explanation.

Sparse Autoencoders (SAEs) have emerged as a useful tool for interpreting the internal representations of neural networks. However, naively optimising SAEs for reconstruction loss and sparsity results in a preference for SAEs that are extremely wide and sparse. We present an information-theoretic framework for interpreting SAEs as lossy compression algorithms for communicating explanations of neural activations. We appeal to the Minimal Description Length (MDL) principle to motivate explanations of activations which are both accurate and concise. We further argue that interpretable SAEs require an additional property, "independent additivity": features should be able to be understood separately. We demonstrate an example of applying our MDL-inspired framework by training SAEs on MNIST handwritten digits and find that SAE features representing significant line segments are optimal, as opposed to SAEs with features for memorised digits from the dataset or small digit fragments. We argue that using MDL rather than sparsity may avoid potential pitfalls with naively maximising sparsity such as undesirable feature splitting and that this framework naturally suggests new hierarchical SAE architectures which provide more concise explanations.

Interpretability tools that offer explanations in the form of a dialogue have demonstrated their efficacy in enhancing users' understanding, as one-off explanations may occasionally fall short in providing sufficient information to the user. Current solutions for dialogue-based explanations, however, require many dependencies and are not easily transferable to tasks they were not designed for. With LLMCheckup, we present an easily accessible tool that allows users to chat with any state-of-the-art large language model (LLM) about its behavior. We enable LLMs to generate all explanations by themselves and take care of intent recognition without fine-tuning, by connecting them with a broad spectrum of Explainable AI (XAI) tools, e.g. feature attributions, embedding-based similarity, and prompting strategies for counterfactual and rationale generation. LLM (self-)explanations are presented as an interactive dialogue that supports follow-up questions and generates suggestions. LLMCheckup provides tutorials for operations available in the system, catering to individuals with varying levels of expertise in XAI and supports multiple input modalities. We introduce a new parsing strategy called multi-prompt parsing substantially enhancing the parsing accuracy of LLMs. Finally, we showcase the tasks of fact checking and commonsense question answering.

Obtaining human-interpretable explanations of large, general-purpose language models is an urgent goal for AI safety. However, it is just as important that our interpretability methods are faithful to the causal dynamics underlying model behavior and able to robustly generalize to unseen inputs. Distributed Alignment Search (DAS) is a powerful gradient descent method grounded in a theory of causal abstraction that uncovered perfect alignments between interpretable symbolic algorithms and small deep learning models fine-tuned for specific tasks. In the present paper, we scale DAS significantly by replacing the remaining brute-force search steps with learned parameters -- an approach we call DAS. This enables us to efficiently search for interpretable causal structure in large language models while they follow instructions. We apply DAS to the Alpaca model (7B parameters), which, off the shelf, solves a simple numerical reasoning problem. With DAS, we discover that Alpaca does this by implementing a causal model with two interpretable boolean variables. Furthermore, we find that the alignment of neural representations with these variables is robust to changes in inputs and instructions. These findings mark a first step toward deeply understanding the inner-workings of our largest and most widely deployed language models.

Research in mechanistic interpretability seeks to explain behaviors of machine learning models in terms of their internal components. However, most previous work either focuses on simple behaviors in small models, or describes complicated behaviors in larger models with broad strokes. In this work, we bridge this gap by presenting an explanation for how GPT-2 small performs a natural language task called indirect object identification (IOI). Our explanation encompasses 26 attention heads grouped into 7 main classes, which we discovered using a combination of interpretability approaches relying on causal interventions. To our knowledge, this investigation is the largest end-to-end attempt at reverse-engineering a natural behavior "in the wild" in a language model. We evaluate the reliability of our explanation using three quantitative criteria--faithfulness, completeness and minimality. Though these criteria support our explanation, they also point to remaining gaps in our understanding. Our work provides evidence that a mechanistic understanding of large ML models is feasible, opening opportunities to scale our understanding to both larger models and more complex tasks.

Interpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity in the field: most studies employ ad-hoc evaluations and do not share theoretical foundations, making it difficult to measure progress and compare the pros and cons of different techniques. Furthermore, while mechanistic understanding is frequently discussed, the basic causal units underlying these mechanisms are often not explicitly defined. In this paper, we propose a perspective on interpretability research grounded in causal mediation analysis. Specifically, we describe the history and current state of interpretability taxonomized according to the types of causal units (mediators) employed, as well as methods used to search over mediators. We discuss the pros and cons of each mediator, providing insights as to when particular kinds of mediators and search methods are most appropriate depending on the goals of a given study. We argue that this framing yields a more cohesive narrative of the field, as well as actionable insights for future work. Specifically, we recommend a focus on discovering new mediators with better trade-offs between human-interpretability and compute-efficiency, and which can uncover more sophisticated abstractions from neural networks than the primarily linear mediators employed in current work. We also argue for more standardized evaluations that enable principled comparisons across mediator types, such that we can better understand when particular causal units are better suited to particular use cases.

Interpretability and explainability have gained more and more attention in the field of machine learning as they are crucial when it comes to high-stakes decisions and troubleshooting. Since both provide information about predictors and their decision process, they are often seen as two independent means for one single end. This view has led to a dichotomous literature: explainability techniques designed for complex black-box models, or interpretable approaches ignoring the many explainability tools. In this position paper, we challenge the common idea that interpretability and explainability are substitutes for one another by listing their principal shortcomings and discussing how both of them mitigate the drawbacks of the other. In doing so, we call for a new perspective on interpretability and explainability, and works targeting both topics simultaneously, leveraging each of their respective assets.

With the perpetual increase of complexity of the state-of-the-art deep neural networks, it becomes a more and more challenging task to maintain their interpretability. Our work aims to evaluate the effects of adversarial training utilized to produce robust models - less vulnerable to adversarial attacks. It has been shown to make computer vision models more interpretable. Interpretability is as essential as robustness when we deploy the models to the real world. To prove the correlation between these two problems, we extensively examine the models using local feature-importance methods (SHAP, Integrated Gradients) and feature visualization techniques (Representation Inversion, Class Specific Image Generation). Standard models, compared to robust are more susceptible to adversarial attacks, and their learned representations are less meaningful to humans. Conversely, these models focus on distinctive regions of the images which support their predictions. Moreover, the features learned by the robust model are closer to the real ones.

Interpretability methods are valuable only if their explanations faithfully describe the explained model. In this work, we consider neural networks whose predictions are invariant under a specific symmetry group. This includes popular architectures, ranging from convolutional to graph neural networks. Any explanation that faithfully explains this type of model needs to be in agreement with this invariance property. We formalize this intuition through the notion of explanation invariance and equivariance by leveraging the formalism from geometric deep learning. Through this rigorous formalism, we derive (1) two metrics to measure the robustness of any interpretability method with respect to the model symmetry group; (2) theoretical robustness guarantees for some popular interpretability methods and (3) a systematic approach to increase the invariance of any interpretability method with respect to a symmetry group. By empirically measuring our metrics for explanations of models associated with various modalities and symmetry groups, we derive a set of 5 guidelines to allow users and developers of interpretability methods to produce robust explanations.

Interpretability of Deep Learning (DL) is a barrier to trustworthy AI. Despite great efforts made by the Explainable AI (XAI) community, explanations lack robustness -- indistinguishable input perturbations may lead to different XAI results. Thus, it is vital to assess how robust DL interpretability is, given an XAI method. In this paper, we identify several challenges that the state-of-the-art is unable to cope with collectively: i) existing metrics are not comprehensive; ii) XAI techniques are highly heterogeneous; iii) misinterpretations are normally rare events. To tackle these challenges, we introduce two black-box evaluation methods, concerning the worst-case interpretation discrepancy and a probabilistic notion of how robust in general, respectively. Genetic Algorithm (GA) with bespoke fitness function is used to solve constrained optimisation for efficient worst-case evaluation. Subset Simulation (SS), dedicated to estimate rare event probabilities, is used for evaluating overall robustness. Experiments show that the accuracy, sensitivity, and efficiency of our methods outperform the state-of-the-arts. Finally, we demonstrate two applications of our methods: ranking robust XAI methods and selecting training schemes to improve both classification and interpretation robustness.

Machine learning (ML) interpretability techniques can reveal undesirable patterns in data that models exploit to make predictions--potentially causing harms once deployed. However, how to take action to address these patterns is not always clear. In a collaboration between ML and human-computer interaction researchers, physicians, and data scientists, we develop GAM Changer, the first interactive system to help domain experts and data scientists easily and responsibly edit Generalized Additive Models (GAMs) and fix problematic patterns. With novel interaction techniques, our tool puts interpretability into action--empowering users to analyze, validate, and align model behaviors with their knowledge and values. Physicians have started to use our tool to investigate and fix pneumonia and sepsis risk prediction models, and an evaluation with 7 data scientists working in diverse domains highlights that our tool is easy to use, meets their model editing needs, and fits into their current workflows. Built with modern web technologies, our tool runs locally in users' web browsers or computational notebooks, lowering the barrier to use. GAM Changer is available at the following public demo link: https://interpret.ml/gam-changer.

Neural networks for NLP are becoming increasingly complex and widespread, and there is a growing concern if these models are responsible to use. Explaining models helps to address the safety and ethical concerns and is essential for accountability. Interpretability serves to provide these explanations in terms that are understandable to humans. Additionally, post-hoc methods provide explanations after a model is learned and are generally model-agnostic. This survey provides a categorization of how recent post-hoc interpretability methods communicate explanations to humans, it discusses each method in-depth, and how they are validated, as the latter is often a common concern.

Methods for understanding the decisions of and mechanisms underlying deep neural networks (DNNs) typically rely on building intuition by emphasizing sensory or semantic features of individual examples. For instance, methods aim to visualize the components of an input which are "important" to a network's decision, or to measure the semantic properties of single neurons. Here, we argue that interpretability research suffers from an over-reliance on intuition-based approaches that risk-and in some cases have caused-illusory progress and misleading conclusions. We identify a set of limitations that we argue impede meaningful progress in interpretability research, and examine two popular classes of interpretability methods-saliency and single-neuron-based approaches-that serve as case studies for how overreliance on intuition and lack of falsifiability can undermine interpretability research. To address these concerns, we propose a strategy to address these impediments in the form of a framework for strongly falsifiable interpretability research. We encourage researchers to use their intuitions as a starting point to develop and test clear, falsifiable hypotheses, and hope that our framework yields robust, evidence-based interpretability methods that generate meaningful advances in our understanding of DNNs.

The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net's internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result--for example, how sensitive a prediction of "zebra" is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.

Interpretable machine learning has exploded as an area of interest over the last decade, sparked by the rise of increasingly large datasets and deep neural networks. Simultaneously, large language models (LLMs) have demonstrated remarkable capabilities across a wide array of tasks, offering a chance to rethink opportunities in interpretable machine learning. Notably, the capability to explain in natural language allows LLMs to expand the scale and complexity of patterns that can be given to a human. However, these new capabilities raise new challenges, such as hallucinated explanations and immense computational costs. In this position paper, we start by reviewing existing methods to evaluate the emerging field of LLM interpretation (both interpreting LLMs and using LLMs for explanation). We contend that, despite their limitations, LLMs hold the opportunity to redefine interpretability with a more ambitious scope across many applications, including in auditing LLMs themselves. We highlight two emerging research priorities for LLM interpretation: using LLMs to directly analyze new datasets and to generate interactive explanations.

Mechanistic interpretability aims to understand the computational mechanisms underlying neural networks' capabilities in order to accomplish concrete scientific and engineering goals. Progress in this field thus promises to provide greater assurance over AI system behavior and shed light on exciting scientific questions about the nature of intelligence. Despite recent progress toward these goals, there are many open problems in the field that require solutions before many scientific and practical benefits can be realized: Our methods require both conceptual and practical improvements to reveal deeper insights; we must figure out how best to apply our methods in pursuit of specific goals; and the field must grapple with socio-technical challenges that influence and are influenced by our work. This forward-facing review discusses the current frontier of mechanistic interpretability and the open problems that the field may benefit from prioritizing.

Automated interpretability pipelines generate natural language descriptions for the concepts represented by features in large language models (LLMs), such as plants or the first word in a sentence. These descriptions are derived using inputs that activate the feature, which may be a dimension or a direction in the model's representation space. However, identifying activating inputs is costly, and the mechanistic role of a feature in model behavior is determined both by how inputs cause a feature to activate and by how feature activation affects outputs. Using steering evaluations, we reveal that current pipelines provide descriptions that fail to capture the causal effect of the feature on outputs. To fix this, we propose efficient, output-centric methods for automatically generating feature descriptions. These methods use the tokens weighted higher after feature stimulation or the highest weight tokens after applying the vocabulary "unembedding" head directly to the feature. Our output-centric descriptions better capture the causal effect of a feature on model outputs than input-centric descriptions, but combining the two leads to the best performance on both input and output evaluations. Lastly, we show that output-centric descriptions can be used to find inputs that activate features previously thought to be "dead".

Mechanistic interpretability research faces a gap between analyzing simple circuits in toy tasks and discovering features in large models. To bridge this gap, we propose text-to-SQL generation as an ideal task to study, as it combines the formal structure of toy tasks with real-world complexity. We introduce TinySQL, a synthetic dataset, progressing from basic to advanced SQL operations, and train models ranging from 33M to 1B parameters to establish a comprehensive testbed for interpretability. We apply multiple complementary interpretability techniques, including Edge Attribution Patching and Sparse Autoencoders, to identify minimal circuits and components supporting SQL generation. We compare circuits for different SQL subskills, evaluating their minimality, reliability, and identifiability. Finally, we conduct a layerwise logit lens analysis to reveal how models compose SQL queries across layers: from intent recognition to schema resolution to structured generation. Our work provides a robust framework for probing and comparing interpretability methods in a structured, progressively complex setting.

Mechanistic Interpretability aims to reverse engineer the algorithms implemented by neural networks by studying their weights and activations. An obstacle to reverse engineering neural networks is that many of the parameters inside a network are not involved in the computation being implemented by the network. These degenerate parameters may obfuscate internal structure. Singular learning theory teaches us that neural network parameterizations are biased towards being more degenerate, and parameterizations with more degeneracy are likely to generalize further. We identify 3 ways that network parameters can be degenerate: linear dependence between activations in a layer; linear dependence between gradients passed back to a layer; ReLUs which fire on the same subset of datapoints. We also present a heuristic argument that modular networks are likely to be more degenerate, and we develop a metric for identifying modules in a network that is based on this argument. We propose that if we can represent a neural network in a way that is invariant to reparameterizations that exploit the degeneracies, then this representation is likely to be more interpretable, and we provide some evidence that such a representation is likely to have sparser interactions. We introduce the Interaction Basis, a tractable technique to obtain a representation that is invariant to degeneracies from linear dependence of activations or Jacobians.

Recent advances in recurrent neural network architectures, such as Mamba and RWKV, have enabled RNNs to match or exceed the performance of equal-size transformers in terms of language modeling perplexity and downstream evaluations, suggesting that future systems may be built on completely new architectures. In this paper, we examine if selected interpretability methods originally designed for transformer language models will transfer to these up-and-coming recurrent architectures. Specifically, we focus on steering model outputs via contrastive activation addition, on eliciting latent predictions via the tuned lens, and eliciting latent knowledge from models fine-tuned to produce false outputs under certain conditions. Our results show that most of these techniques are effective when applied to RNNs, and we show that it is possible to improve some of them by taking advantage of RNNs' compressed state.

Vision-Language-Action (VLA) models are a promising path to realizing generalist embodied agents that can quickly adapt to new tasks, modalities, and environments. However, methods for interpreting and steering VLAs fall far short of classical robotics pipelines, which are grounded in explicit models of kinematics, dynamics, and control. This lack of mechanistic insight is a central challenge for deploying learned policies in real-world robotics, where robustness and explainability are critical. Motivated by advances in mechanistic interpretability for large language models, we introduce the first framework for interpreting and steering VLAs via their internal representations, enabling direct intervention in model behavior at inference time. We project feedforward activations within transformer layers onto the token embedding basis, identifying sparse semantic directions - such as speed and direction - that are causally linked to action selection. Leveraging these findings, we introduce a general-purpose activation steering method that modulates behavior in real time, without fine-tuning, reward signals, or environment interaction. We evaluate this method on two recent open-source VLAs, Pi0 and OpenVLA, and demonstrate zero-shot behavioral control in simulation (LIBERO) and on a physical robot (UR5). This work demonstrates that interpretable components of embodied VLAs can be systematically harnessed for control - establishing a new paradigm for transparent and steerable foundation models in robotics.

Mechanistic interpretability is the program of explaining what AI systems are doing in terms of their internal mechanisms. I analyze some aspects of the program, along with setting out some concrete challenges and assessing progress to date. I argue for the importance of propositional interpretability, which involves interpreting a system's mechanisms and behavior in terms of propositional attitudes: attitudes (such as belief, desire, or subjective probability) to propositions (e.g. the proposition that it is hot outside). Propositional attitudes are the central way that we interpret and explain human beings and they are likely to be central in AI too. A central challenge is what I call thought logging: creating systems that log all of the relevant propositional attitudes in an AI system over time. I examine currently popular methods of interpretability (such as probing, sparse auto-encoders, and chain of thought methods) as well as philosophical methods of interpretation (including those grounded in psychosemantics) to assess their strengths and weaknesses as methods of propositional interpretability.

Mechanistic Interpretability (MI) promises a path toward fully understanding how neural networks make their predictions. Prior work demonstrates that even when trained to perform simple arithmetic, models can implement a variety of algorithms (sometimes concurrently) depending on initialization and hyperparameters. Does this mean neuron-level interpretability techniques have limited applicability? We argue that high-dimensional neural networks can learn low-dimensional representations of their training data that are useful beyond simply making good predictions. Such representations can be understood through the mechanistic interpretability lens and provide insights that are surprisingly faithful to human-derived domain knowledge. This indicates that such approaches to interpretability can be useful for deriving a new understanding of a problem from models trained to solve it. As a case study, we extract nuclear physics concepts by studying models trained to reproduce nuclear data.

We have only limited understanding of how and why large language models (LLMs) respond in the ways that they do. Their neural networks have proven challenging to interpret, and we are only beginning to tease out the function of individual neurons and circuits within them. However, another path to understanding these systems is to investigate and develop their capacity to introspect and explain their own functioning. Here, we show that i) contemporary LLMs are capable of providing accurate, quantitative descriptions of their own internal processes during certain kinds of decision-making, ii) that it is possible to improve these capabilities through training, and iii) that this training generalizes to at least some degree. To do so, we fine-tuned GPT-4o and GPT-4o-mini to make decisions in a wide variety of complex contexts (e.g., choosing between condos, loans, vacations, etc.) according to randomly-generated, quantitative preferences about how to weigh different attributes during decision-making (e.g., the relative importance of natural light versus quiet surroundings for condos). We demonstrate that the LLMs can accurately report these preferences (i.e., the weights that they learned to give to different attributes during decision-making). Next, we demonstrate that these LLMs can be fine-tuned to explain their decision-making even more accurately. Finally, we demonstrate that this training generalizes: It improves the ability of the models to accurately explain what they are doing as they make other complex decisions, not just decisions they have learned to make via fine-tuning. This work is a step towards training LLMs to accurately and broadly report on their own internal processes -- a possibility that would yield substantial benefits for interpretability, control, and safety.

Mechanistic interpretability methods aim to identify the algorithm a neural network implements, but it is difficult to validate such methods when the true algorithm is unknown. This work presents InterpBench, a collection of semi-synthetic yet realistic transformers with known circuits for evaluating these techniques. We train these neural networks using a stricter version of Interchange Intervention Training (IIT) which we call Strict IIT (SIIT). Like the original, SIIT trains neural networks by aligning their internal computation with a desired high-level causal model, but it also prevents non-circuit nodes from affecting the model's output. We evaluate SIIT on sparse transformers produced by the Tracr tool and find that SIIT models maintain Tracr's original circuit while being more realistic. SIIT can also train transformers with larger circuits, like Indirect Object Identification (IOI). Finally, we use our benchmark to evaluate existing circuit discovery techniques.

Interpreting the inner function of neural networks is crucial for the trustworthy development and deployment of these black-box models. Prior interpretability methods focus on correlation-based measures to attribute model decisions to individual examples. However, these measures are susceptible to noise and spurious correlations encoded in the model during the training phase (e.g., biased inputs, model overfitting, or misspecification). Moreover, this process has proven to result in noisy and unstable attributions that prevent any transparent understanding of the model's behavior. In this paper, we develop a robust interventional-based method grounded by causal analysis to capture cause-effect mechanisms in pre-trained neural networks and their relation to the prediction. Our novel approach relies on path interventions to infer the causal mechanisms within hidden layers and isolate relevant and necessary information (to model prediction), avoiding noisy ones. The result is task-specific causal explanatory graphs that can audit model behavior and express the actual causes underlying its performance. We apply our method to vision models trained on classification tasks. On image classification tasks, we provide extensive quantitative experiments to show that our approach can capture more stable and faithful explanations than standard attribution-based methods. Furthermore, the underlying causal graphs reveal the neural interactions in the model, making it a valuable tool in other applications (e.g., model repair).

Self-attention techniques, and specifically Transformers, are dominating the field of text processing and are becoming increasingly popular in computer vision classification tasks. In order to visualize the parts of the image that led to a certain classification, existing methods either rely on the obtained attention maps or employ heuristic propagation along the attention graph. In this work, we propose a novel way to compute relevancy for Transformer networks. The method assigns local relevance based on the Deep Taylor Decomposition principle and then propagates these relevancy scores through the layers. This propagation involves attention layers and skip connections, which challenge existing methods. Our solution is based on a specific formulation that is shown to maintain the total relevancy across layers. We benchmark our method on very recent visual Transformer networks, as well as on a text classification problem, and demonstrate a clear advantage over the existing explainability methods.

Multi-Agent Deep Reinforcement Learning (MADRL) was proven efficient in solving complex problems in robotics or games, yet most of the trained models are hard to interpret. While learning intrinsically interpretable models remains a prominent approach, its scalability and flexibility are limited in handling complex tasks or multi-agent dynamics. This paper advocates for direct interpretability, generating post hoc explanations directly from trained models, as a versatile and scalable alternative, offering insights into agents' behaviour, emergent phenomena, and biases without altering models' architectures. We explore modern methods, including relevance backpropagation, knowledge edition, model steering, activation patching, sparse autoencoders and circuit discovery, to highlight their applicability to single-agent, multi-agent, and training process challenges. By addressing MADRL interpretability, we propose directions aiming to advance active topics such as team identification, swarm coordination and sample efficiency.

Understanding the mechanisms behind Large Language Models (LLMs) is crucial for designing improved models and strategies. While recent studies have yielded valuable insights into the mechanisms of textual LLMs, the mechanisms of Multi-modal Large Language Models (MLLMs) remain underexplored. In this paper, we apply mechanistic interpretability methods to analyze the visual question answering (VQA) mechanisms in the first MLLM, Llava. We compare the mechanisms between VQA and textual QA (TQA) in color answering tasks and find that: a) VQA exhibits a mechanism similar to the in-context learning mechanism observed in TQA; b) the visual features exhibit significant interpretability when projecting the visual embeddings into the embedding space; and c) Llava enhances the existing capabilities of the corresponding textual LLM Vicuna during visual instruction tuning. Based on these findings, we develop an interpretability tool to help users and researchers identify important visual locations for final predictions, aiding in the understanding of visual hallucination. Our method demonstrates faster and more effective results compared to existing interpretability approaches. Code: https://github.com/zepingyu0512/llava-mechanism

A mechanistic understanding of how MLPs do computation in deep neural networks remains elusive. Current interpretability work can extract features from hidden activations over an input dataset but generally cannot explain how MLP weights construct features. One challenge is that element-wise nonlinearities introduce higher-order interactions and make it difficult to trace computations through the MLP layer. In this paper, we analyze bilinear MLPs, a type of Gated Linear Unit (GLU) without any element-wise nonlinearity that nevertheless achieves competitive performance. Bilinear MLPs can be fully expressed in terms of linear operations using a third-order tensor, allowing flexible analysis of the weights. Analyzing the spectra of bilinear MLP weights using eigendecomposition reveals interpretable low-rank structure across toy tasks, image classification, and language modeling. We use this understanding to craft adversarial examples, uncover overfitting, and identify small language model circuits directly from the weights alone. Our results demonstrate that bilinear layers serve as an interpretable drop-in replacement for current activation functions and that weight-based interpretability is viable for understanding deep-learning models.

Mechanistic interpretability aims to understand model behaviors in terms of specific, interpretable features, often hypothesized to manifest as low-dimensional subspaces of activations. Specifically, recent studies have explored subspace interventions (such as activation patching) as a way to simultaneously manipulate model behavior and attribute the features behind it to given subspaces. In this work, we demonstrate that these two aims diverge, potentially leading to an illusory sense of interpretability. Counterintuitively, even if a subspace intervention makes the model's output behave as if the value of a feature was changed, this effect may be achieved by activating a dormant parallel pathway leveraging another subspace that is causally disconnected from model outputs. We demonstrate this phenomenon in a distilled mathematical example, in two real-world domains (the indirect object identification task and factual recall), and present evidence for its prevalence in practice. In the context of factual recall, we further show a link to rank-1 fact editing, providing a mechanistic explanation for previous work observing an inconsistency between fact editing performance and fact localization. However, this does not imply that activation patching of subspaces is intrinsically unfit for interpretability. To contextualize our findings, we also show what a success case looks like in a task (indirect object identification) where prior manual circuit analysis informs an understanding of the location of a feature. We explore the additional evidence needed to argue that a patched subspace is faithful.

Along with the great success of deep neural networks, there is also growing concern about their black-box nature. The interpretability issue affects people's trust on deep learning systems. It is also related to many ethical problems, e.g., algorithmic discrimination. Moreover, interpretability is a desired property for deep networks to become powerful tools in other research fields, e.g., drug discovery and genomics. In this survey, we conduct a comprehensive review of the neural network interpretability research. We first clarify the definition of interpretability as it has been used in many different contexts. Then we elaborate on the importance of interpretability and propose a novel taxonomy organized along three dimensions: type of engagement (passive vs. active interpretation approaches), the type of explanation, and the focus (from local to global interpretability). This taxonomy provides a meaningful 3D view of distribution of papers from the relevant literature as two of the dimensions are not simply categorical but allow ordinal subcategories. Finally, we summarize the existing interpretability evaluation methods and suggest possible research directions inspired by our new taxonomy.

In the rapidly evolving landscape of artificial intelligence, multi-modal large language models are emerging as a significant area of interest. These models, which combine various forms of data input, are becoming increasingly popular. However, understanding their internal mechanisms remains a complex task. Numerous advancements have been made in the field of explainability tools and mechanisms, yet there is still much to explore. In this work, we present a novel interactive application aimed towards understanding the internal mechanisms of large vision-language models. Our interface is designed to enhance the interpretability of the image patches, which are instrumental in generating an answer, and assess the efficacy of the language model in grounding its output in the image. With our application, a user can systematically investigate the model and uncover system limitations, paving the way for enhancements in system capabilities. Finally, we present a case study of how our application can aid in understanding failure mechanisms in a popular large multi-modal model: LLaVA.

Circuit analysis is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer label given knowledge of the correct answer text. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.

Large Language Models (LLMs) have demonstrated unprecedented capabilities across various natural language processing tasks. Their ability to process and generate viable text and code has made them ubiquitous in many fields, while their deployment as knowledge bases and "reasoning" tools remains an area of ongoing research. In geography, a growing body of literature has been focusing on evaluating LLMs' geographical knowledge and their ability to perform spatial reasoning. However, very little is still known about the internal functioning of these models, especially about how they process geographical information. In this chapter, we establish a novel framework for the study of geospatial mechanistic interpretability - using spatial analysis to reverse engineer how LLMs handle geographical information. Our aim is to advance our understanding of the internal representations that these complex models generate while processing geographical information - what one might call "how LLMs think about geographic information" if such phrasing was not an undue anthropomorphism. We first outline the use of probing in revealing internal structures within LLMs. We then introduce the field of mechanistic interpretability, discussing the superposition hypothesis and the role of sparse autoencoders in disentangling polysemantic internal representations of LLMs into more interpretable, monosemantic features. In our experiments, we use spatial autocorrelation to show how features obtained for placenames display spatial patterns related to their geographic location and can thus be interpreted geospatially, providing insights into how these models process geographical information. We conclude by discussing how our framework can help shape the study and use of foundation models in geography.

Sparse Autoencoders (SAEs) are a prominent tool in mechanistic interpretability (MI) for decomposing neural network activations into interpretable features. However, the aspiration to identify a canonical set of features is challenged by the observed inconsistency of learned SAE features across different training runs, undermining the reliability and efficiency of MI research. This position paper argues that mechanistic interpretability should prioritize feature consistency in SAEs -- the reliable convergence to equivalent feature sets across independent runs. We propose using the Pairwise Dictionary Mean Correlation Coefficient (PW-MCC) as a practical metric to operationalize consistency and demonstrate that high levels are achievable (0.80 for TopK SAEs on LLM activations) with appropriate architectural choices. Our contributions include detailing the benefits of prioritizing consistency; providing theoretical grounding and synthetic validation using a model organism, which verifies PW-MCC as a reliable proxy for ground-truth recovery; and extending these findings to real-world LLM data, where high feature consistency strongly correlates with the semantic similarity of learned feature explanations. We call for a community-wide shift towards systematically measuring feature consistency to foster robust cumulative progress in MI.

Traditional white-box methods for creating adversarial perturbations against LLMs typically rely only on gradient computation from the targeted model, ignoring the internal mechanisms responsible for attack success or failure. Conversely, interpretability studies that analyze these internal mechanisms lack practical applications beyond runtime interventions. We bridge this gap by introducing a novel white-box approach that leverages mechanistic interpretability techniques to craft practical adversarial inputs. Specifically, we first identify acceptance subspaces - sets of feature vectors that do not trigger the model's refusal mechanisms - then use gradient-based optimization to reroute embeddings from refusal subspaces to acceptance subspaces, effectively achieving jailbreaks. This targeted approach significantly reduces computation cost, achieving attack success rates of 80-95\% on state-of-the-art models including Gemma2, Llama3.2, and Qwen2.5 within minutes or even seconds, compared to existing techniques that often fail or require hours of computation. We believe this approach opens a new direction for both attack research and defense development. Furthermore, it showcases a practical application of mechanistic interpretability where other methods are less efficient, which highlights its utility. The code and generated datasets are available at https://github.com/Sckathach/subspace-rerouting.

Language models (LMs) have proven to be powerful tools for psycholinguistic research, but most prior work has focused on purely behavioural measures (e.g., surprisal comparisons). At the same time, research in model interpretability has begun to illuminate the abstract causal mechanisms shaping LM behavior. To help bring these strands of research closer together, we introduce CausalGym. We adapt and expand the SyntaxGym suite of tasks to benchmark the ability of interpretability methods to causally affect model behaviour. To illustrate how CausalGym can be used, we study the pythia models (14M--6.9B) and assess the causal efficacy of a wide range of interpretability methods, including linear probing and distributed alignment search (DAS). We find that DAS outperforms the other methods, and so we use it to study the learning trajectory of two difficult linguistic phenomena in pythia-1b: negative polarity item licensing and filler--gap dependencies. Our analysis shows that the mechanism implementing both of these tasks is learned in discrete stages, not gradually.

Individual neurons participate in the representation of multiple high-level concepts. To what extent can different interpretability methods successfully disentangle these roles? To help address this question, we introduce RAVEL (Resolving Attribute-Value Entanglements in Language Models), a dataset that enables tightly controlled, quantitative comparisons between a variety of existing interpretability methods. We use the resulting conceptual framework to define the new method of Multi-task Distributed Alignment Search (MDAS), which allows us to find distributed representations satisfying multiple causal criteria. With Llama2-7B as the target language model, MDAS achieves state-of-the-art results on RAVEL, demonstrating the importance of going beyond neuron-level analyses to identify features distributed across activations. We release our benchmark at https://github.com/explanare/ravel.

Despite extensive standardization, diagnostic interviews for mental health disorders encompass substantial subjective judgment. Previous studies have demonstrated that EEG-based neural measures can function as reliable objective correlates of depression, or even predictors of depression and its course. However, their clinical utility has not been fully realized because of 1) the lack of automated ways to deal with the inherent noise associated with EEG data at scale, and 2) the lack of knowledge of which aspects of the EEG signal may be markers of a clinical disorder. Here we adapt an unsupervised pipeline from the recent deep representation learning literature to address these problems by 1) learning a disentangled representation using beta-VAE to denoise the signal, and 2) extracting interpretable features associated with a sparse set of clinical labels using a Symbol-Concept Association Network (SCAN). We demonstrate that our method is able to outperform the canonical hand-engineered baseline classification method on a number of factors, including participant age and depression diagnosis. Furthermore, our method recovers a representation that can be used to automatically extract denoised Event Related Potentials (ERPs) from novel, single EEG trajectories, and supports fast supervised re-mapping to various clinical labels, allowing clinicians to re-use a single EEG representation regardless of updates to the standardized diagnostic system. Finally, single factors of the learned disentangled representations often correspond to meaningful markers of clinical factors, as automatically detected by SCAN, allowing for human interpretability and post-hoc expert analysis of the recommendations made by the model.

Deep reinforcement learning (RL) has shown remarkable success in complex domains, however, the inherent black box nature of deep neural network policies raises significant challenges in understanding and trusting the decision-making processes. While existing explainable RL methods provide local insights, they fail to deliver a global understanding of the model, particularly in high-stakes applications. To overcome this limitation, we propose a novel model-agnostic approach that bridges the gap between explainability and interpretability by leveraging Shapley values to transform complex deep RL policies into transparent representations. The proposed approach offers two key contributions: a novel approach employing Shapley values to policy interpretation beyond local explanations and a general framework applicable to off-policy and on-policy algorithms. We evaluate our approach with three existing deep RL algorithms and validate its performance in two classic control environments. The results demonstrate that our approach not only preserves the original models' performance but also generates more stable interpretable policies.

Predicting high-dimensional or extreme multilabels, such as in medical coding, requires both accuracy and interpretability. Existing works often rely on local interpretability methods, failing to provide comprehensive explanations of the overall mechanism behind each label prediction within a multilabel set. We propose a mechanistic interpretability module called DIctionary Label Attention (\method) that disentangles uninterpretable dense embeddings into a sparse embedding space, where each nonzero element (a dictionary feature) represents a globally learned medical concept. Through human evaluations, we show that our sparse embeddings are more human understandable than its dense counterparts by at least 50 percent. Our automated dictionary feature identification pipeline, leveraging large language models (LLMs), uncovers thousands of learned medical concepts by examining and summarizing the highest activating tokens for each dictionary feature. We represent the relationships between dictionary features and medical codes through a sparse interpretable matrix, enhancing the mechanistic and global understanding of the model's predictions while maintaining competitive performance and scalability without extensive human annotation.

Recent instruction fine-tuned models can solve multiple NLP tasks when prompted to do so, with machine translation (MT) being a prominent use case. However, current research often focuses on standard performance benchmarks, leaving compelling fairness and ethical considerations behind. In MT, this might lead to misgendered translations, resulting, among other harms, in the perpetuation of stereotypes and prejudices. In this work, we address this gap by investigating whether and to what extent such models exhibit gender bias in machine translation and how we can mitigate it. Concretely, we compute established gender bias metrics on the WinoMT corpus from English to German and Spanish. We discover that IFT models default to male-inflected translations, even disregarding female occupational stereotypes. Next, using interpretability methods, we unveil that models systematically overlook the pronoun indicating the gender of a target occupation in misgendered translations. Finally, based on this finding, we propose an easy-to-implement and effective bias mitigation solution based on few-shot learning that leads to significantly fairer translations.

Labeling neural network submodules with human-legible descriptions is useful for many downstream tasks: such descriptions can surface failures, guide interventions, and perhaps even explain important model behaviors. To date, most mechanistic descriptions of trained networks have involved small models, narrowly delimited phenomena, and large amounts of human labor. Labeling all human-interpretable sub-computations in models of increasing size and complexity will almost certainly require tools that can generate and validate descriptions automatically. Recently, techniques that use learned models in-the-loop for labeling have begun to gain traction, but methods for evaluating their efficacy are limited and ad-hoc. How should we validate and compare open-ended labeling tools? This paper introduces FIND (Function INterpretation and Description), a benchmark suite for evaluating the building blocks of automated interpretability methods. FIND contains functions that resemble components of trained neural networks, and accompanying descriptions of the kind we seek to generate. The functions are procedurally constructed across textual and numeric domains, and involve a range of real-world complexities, including noise, composition, approximation, and bias. We evaluate new and existing methods that use language models (LMs) to produce code-based and language descriptions of function behavior. We find that an off-the-shelf LM augmented with only black-box access to functions can sometimes infer their structure, acting as a scientist by forming hypotheses, proposing experiments, and updating descriptions in light of new data. However, LM-based descriptions tend to capture global function behavior and miss local corruptions. These results show that FIND will be useful for characterizing the performance of more sophisticated interpretability methods before they are applied to real-world models.

Several recent studies have elucidated why knowledge distillation (KD) improves model performance. However, few have researched the other advantages of KD in addition to its improving model performance. In this study, we have attempted to show that KD enhances the interpretability as well as the accuracy of models. We measured the number of concept detectors identified in network dissection for a quantitative comparison of model interpretability. We attributed the improvement in interpretability to the class-similarity information transferred from the teacher to student models. First, we confirmed the transfer of class-similarity information from the teacher to student model via logit distillation. Then, we analyzed how class-similarity information affects model interpretability in terms of its presence or absence and degree of similarity information. We conducted various quantitative and qualitative experiments and examined the results on different datasets, different KD methods, and according to different measures of interpretability. Our research showed that KD models by large models could be used more reliably in various fields.

Past work in natural language processing interpretability focused mainly on popular classification tasks while largely overlooking generation settings, partly due to a lack of dedicated tools. In this work, we introduce Inseq, a Python library to democratize access to interpretability analyses of sequence generation models. Inseq enables intuitive and optimized extraction of models' internal information and feature importance scores for popular decoder-only and encoder-decoder Transformers architectures. We showcase its potential by adopting it to highlight gender biases in machine translation models and locate factual knowledge inside GPT-2. Thanks to its extensible interface supporting cutting-edge techniques such as contrastive feature attribution, Inseq can drive future advances in explainable natural language generation, centralizing good practices and enabling fair and reproducible model evaluations.

Recent large language models have been trained on vast datasets, but also often on repeated data, either intentionally for the purpose of upweighting higher quality data, or unintentionally because data deduplication is not perfect and the model is exposed to repeated data at the sentence, paragraph, or document level. Some works have reported substantial negative performance effects of this repeated data. In this paper we attempt to study repeated data systematically and to understand its effects mechanistically. To do this, we train a family of models where most of the data is unique but a small fraction of it is repeated many times. We find a strong double descent phenomenon, in which repeated data can lead test loss to increase midway through training. A predictable range of repetition frequency leads to surprisingly severe degradation in performance. For instance, performance of an 800M parameter model can be degraded to that of a 2x smaller model (400M params) by repeating 0.1% of the data 100 times, despite the other 90% of the training tokens remaining unique. We suspect there is a range in the middle where the data can be memorized and doing so consumes a large fraction of the model's capacity, and this may be where the peak of degradation occurs. Finally, we connect these observations to recent mechanistic interpretability work - attempting to reverse engineer the detailed computations performed by the model - by showing that data repetition disproportionately damages copying and internal structures associated with generalization, such as induction heads, providing a possible mechanism for the shift from generalization to memorization. Taken together, these results provide a hypothesis for why repeating a relatively small fraction of data in large language models could lead to disproportionately large harms to performance.

We present an attention mechanism inspired from definition of screened Coulomb potential. This attention mechanism was used to interpret the Graph Attention (GAT) model layers and training dataset by using a flexible and scalable framework (CoulGAT) developed for this purpose. Using CoulGAT, a forest of plain and resnet models were trained and characterized using this attention mechanism against CHAMPS dataset. The learnable variables of the attention mechanism are used to extract node-node and node-feature interactions to define an empirical standard model for the graph structure and hidden layer. This representation of graph and hidden layers can be used as a tool to compare different models, optimize hidden layers and extract a compact definition of graph structure of the dataset.

Attention mechanisms have improved the performance of NLP tasks while allowing models to remain explainable. Self-attention is currently widely used, however interpretability is difficult due to the numerous attention distributions. Recent work has shown that model representations can benefit from label-specific information, while facilitating interpretation of predictions. We introduce the Label Attention Layer: a new form of self-attention where attention heads represent labels. We test our novel layer by running constituency and dependency parsing experiments and show our new model obtains new state-of-the-art results for both tasks on both the Penn Treebank (PTB) and Chinese Treebank. Additionally, our model requires fewer self-attention layers compared to existing work. Finally, we find that the Label Attention heads learn relations between syntactic categories and show pathways to analyze errors.

A growing intuition in machine learning suggests a link between sparsity and interpretability. We introduce a novel self-ablation mechanism to investigate this connection ante-hoc in the context of language transformers. Our approach dynamically enforces a k-winner-takes-all constraint, forcing the model to demonstrate selective activation across neuron and attention units. Unlike post-hoc methods that analyze already-trained models, our approach integrates interpretability directly into model training, promoting feature localization from inception. Training small models on the TinyStories dataset and employing interpretability tests, we find that self-ablation leads to more localized circuits, concentrated feature representations, and increased neuron specialization without compromising language modelling performance. Surprisingly, our method also decreased overall sparsity, indicating that self-ablation promotes specialization rather than widespread inactivity. This reveals a complex interplay between sparsity and interpretability, where decreased global sparsity can coexist with increased local specialization, leading to enhanced interpretability. To facilitate reproducibility, we make our code available at https://github.com/keenanpepper/self-ablating-transformers.

We demonstrate how AI agents can coordinate to deceive oversight systems using automated interpretability of neural networks. Using sparse autoencoders (SAEs) as our experimental framework, we show that language models (Llama, DeepSeek R1, and Claude 3.7 Sonnet) can generate deceptive explanations that evade detection. Our agents employ steganographic methods to hide information in seemingly innocent explanations, successfully fooling oversight models while achieving explanation quality comparable to reference labels. We further find that models can scheme to develop deceptive strategies when they believe the detection of harmful features might lead to negative consequences for themselves. All tested LLM agents were capable of deceiving the overseer while achieving high interpretability scores comparable to those of reference labels. We conclude by proposing mitigation strategies, emphasizing the critical need for robust understanding and defenses against deception.

An approach to improve neural network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We define a measure for clusterability and show that pre-trained models form highly enmeshed clusters via spectral graph clustering. We thus train models to be more modular using a "clusterability loss" function that encourages the formation of non-interacting clusters. Using automated interpretability techniques, we show that our method can help train models that are more modular and learn different, disjoint, and smaller circuits. We investigate CNNs trained on MNIST and CIFAR, small transformers trained on modular addition, and language models. Our approach provides a promising direction for training neural networks that learn simpler functions and are easier to interpret.

Sparse Autoencoders (SAEs) have shown promise in improving the interpretability of neural network activations, but can learn features that are not features of the input, limiting their effectiveness. We propose Mutual Feature Regularization (MFR), a regularization technique for improving feature learning by encouraging SAEs trained in parallel to learn similar features. We motivate MFR by showing that features learned by multiple SAEs are more likely to correlate with features of the input. By training on synthetic data with known features of the input, we show that MFR can help SAEs learn those features, as we can directly compare the features learned by the SAE with the input features for the synthetic data. We then scale MFR to SAEs that are trained to denoise electroencephalography (EEG) data and SAEs that are trained to reconstruct GPT-2 Small activations. We show that MFR can improve the reconstruction loss of SAEs by up to 21.21\% on GPT-2 Small, and 6.67\% on EEG data. Our results suggest that the similarity between features learned by different SAEs can be leveraged to improve SAE training, thereby enhancing performance and the usefulness of SAEs for model interpretability.

Assessing the degree of similarity of code fragments is crucial for ensuring software quality, but it remains challenging due to the need to capture the deeper semantic aspects of code. Traditional syntactic methods often fail to identify these connections. Recent advancements have addressed this challenge, though they frequently sacrifice interpretability. To improve this, we present an approach aiming to improve the transparency of the similarity assessment by using GraphCodeBERT, which enables the identification of semantic relationships between code fragments. This approach identifies similar code fragments and clarifies the reasons behind that identification, helping developers better understand and trust the results. The source code for our implementation is available at https://www.github.com/jorge-martinez-gil/graphcodebert-interpretability.

Cross-lingual transfer learning is an important property of multilingual large language models (LLMs). But how do LLMs represent relationships between languages? Every language model has an input layer that maps tokens to vectors. This ubiquitous layer of language models is often overlooked. We find that similarities between these input embeddings are highly interpretable and that the geometry of these embeddings differs between model families. In one case (XLM-RoBERTa), embeddings encode language: tokens in different writing systems can be linearly separated with an average of 99.2% accuracy. Another family (mT5) represents cross-lingual semantic similarity: the 50 nearest neighbors for any token represent an average of 7.61 writing systems, and are frequently translations. This result is surprising given that there is no explicit parallel cross-lingual training corpora and no explicit incentive for translations in pre-training objectives. Our research opens the door for investigations in 1) The effect of pre-training and model architectures on representations of languages and 2) The applications of cross-lingual representations embedded in language models.

In this work, we study the challenge of providing human-understandable descriptions for failure modes in trained image classification models. Existing works address this problem by first identifying clusters (or directions) of incorrectly classified samples in a latent space and then aiming to provide human-understandable text descriptions for them. We observe that in some cases, describing text does not match well with identified failure modes, partially owing to the fact that shared interpretable attributes of failure modes may not be captured using clustering in the feature space. To improve on these shortcomings, we propose a novel approach that prioritizes interpretability in this problem: we start by obtaining human-understandable concepts (tags) of images in the dataset and then analyze the model's behavior based on the presence or absence of combinations of these tags. Our method also ensures that the tags describing a failure mode form a minimal set, avoiding redundant and noisy descriptions. Through several experiments on different datasets, we show that our method successfully identifies failure modes and generates high-quality text descriptions associated with them. These results highlight the importance of prioritizing interpretability in understanding model failures.

In this paper we delve into the properties of transformers, attained through self-supervision, in the point cloud domain. Specifically, we evaluate the effectiveness of Masked Autoencoding as a pretraining scheme, and explore Momentum Contrast as an alternative. In our study we investigate the impact of data quantity on the learned features, and uncover similarities in the transformer's behavior across domains. Through comprehensive visualiations, we observe that the transformer learns to attend to semantically meaningful regions, indicating that pretraining leads to a better understanding of the underlying geometry. Moreover, we examine the finetuning process and its effect on the learned representations. Based on that, we devise an unfreezing strategy which consistently outperforms our baseline without introducing any other modifications to the model or the training pipeline, and achieve state-of-the-art results in the classification task among transformer models.

As machine learning models become increasingly prevalent in medical diagnostics, the need for interpretability and transparency becomes paramount. The XAI Renaissance signifies a significant shift in the field, aiming to redefine the interpretability of medical diagnostic models. This paper explores the innovative approaches and methodologies within the realm of Explainable AI (XAI) that are revolutionizing the interpretability of medical diagnostic models. By shedding light on the underlying decision-making process, XAI techniques empower healthcare professionals to understand, trust, and effectively utilize these models for accurate and reliable medical diagnoses. This review highlights the key advancements in XAI for medical diagnostics and their potential to transform the healthcare landscape, ultimately improving patient outcomes and fostering trust in AI-driven diagnostic systems.

Part-prototype networks (e.g., ProtoPNet, ProtoTree and ProtoPool) have attracted broad research interest for their intrinsic interpretability and comparable accuracy to non-interpretable counterparts. However, recent works find that the interpretability from prototypes is fragile, due to the semantic gap between the similarities in the feature space and that in the input space. In this work, we strive to address this challenge by making the first attempt to quantitatively and objectively evaluate the interpretability of the part-prototype networks. Specifically, we propose two evaluation metrics, termed as consistency score and stability score, to evaluate the explanation consistency across images and the explanation robustness against perturbations, respectively, both of which are essential for explanations taken into practice. Furthermore, we propose an elaborated part-prototype network with a shallow-deep feature alignment (SDFA) module and a score aggregation (SA) module to improve the interpretability of prototypes. We conduct systematical evaluation experiments and provide substantial discussions to uncover the interpretability of existing part-prototype networks. Experiments on three benchmarks across nine architectures demonstrate that our model achieves significantly superior performance to the state of the art, in both the accuracy and interpretability. Codes are available at https://github.com/hqhQAQ/EvalProtoPNet.

As AI technology is increasingly applied to high-impact, high-risk domains, there have been a number of new methods aimed at making AI models more human interpretable. Despite the recent growth of interpretability work, there is a lack of systematic evaluation of proposed techniques. In this work, we introduce HIVE (Human Interpretability of Visual Explanations), a novel human evaluation framework that assesses the utility of explanations to human users in AI-assisted decision making scenarios, and enables falsifiable hypothesis testing, cross-method comparison, and human-centered evaluation of visual interpretability methods. To the best of our knowledge, this is the first work of its kind. Using HIVE, we conduct IRB-approved human studies with nearly 1000 participants and evaluate four methods that represent the diversity of computer vision interpretability works: GradCAM, BagNet, ProtoPNet, and ProtoTree. Our results suggest that explanations engender human trust, even for incorrect predictions, yet are not distinct enough for users to distinguish between correct and incorrect predictions. We open-source HIVE to enable future studies and encourage more human-centered approaches to interpretability research.

Clinicians are often very sceptical about applying automatic image processing approaches, especially deep learning based methods, in practice. One main reason for this is the black-box nature of these approaches and the inherent problem of missing insights of the automatically derived decisions. In order to increase trust in these methods, this paper presents approaches that help to interpret and explain the results of deep learning algorithms by depicting the anatomical areas which influence the decision of the algorithm most. Moreover, this research presents a unified framework, TorchEsegeta, for applying various interpretability and explainability techniques for deep learning models and generate visual interpretations and explanations for clinicians to corroborate their clinical findings. In addition, this will aid in gaining confidence in such methods. The framework builds on existing interpretability and explainability techniques that are currently focusing on classification models, extending them to segmentation tasks. In addition, these methods have been adapted to 3D models for volumetric analysis. The proposed framework provides methods to quantitatively compare visual explanations using infidelity and sensitivity metrics. This framework can be used by data scientists to perform post-hoc interpretations and explanations of their models, develop more explainable tools and present the findings to clinicians to increase their faith in such models. The proposed framework was evaluated based on a use case scenario of vessel segmentation models trained on Time-of-fight (TOF) Magnetic Resonance Angiogram (MRA) images of the human brain. Quantitative and qualitative results of a comparative study of different models and interpretability methods are presented. Furthermore, this paper provides an extensive overview of several existing interpretability and explainability methods.

The self-attention-based model, transformer, is recently becoming the leading backbone in the field of computer vision. In spite of the impressive success made by transformers in a variety of vision tasks, it still suffers from heavy computation and intensive memory costs. To address this limitation, this paper presents an Interpretability-Aware REDundancy REDuction framework (IA-RED^2). We start by observing a large amount of redundant computation, mainly spent on uncorrelated input patches, and then introduce an interpretable module to dynamically and gracefully drop these redundant patches. This novel framework is then extended to a hierarchical structure, where uncorrelated tokens at different stages are gradually removed, resulting in a considerable shrinkage of computational cost. We include extensive experiments on both image and video tasks, where our method could deliver up to 1.4x speed-up for state-of-the-art models like DeiT and TimeSformer, by only sacrificing less than 0.7% accuracy. More importantly, contrary to other acceleration approaches, our method is inherently interpretable with substantial visual evidence, making vision transformer closer to a more human-understandable architecture while being lighter. We demonstrate that the interpretability that naturally emerged in our framework can outperform the raw attention learned by the original visual transformer, as well as those generated by off-the-shelf interpretation methods, with both qualitative and quantitative results. Project Page: http://people.csail.mit.edu/bpan/ia-red/.

As machine learning and algorithmic decision making systems are increasingly being leveraged in high-stakes human-in-the-loop settings, there is a pressing need to understand the rationale of their predictions. Researchers have responded to this need with explainable AI (XAI), but often proclaim interpretability axiomatically without evaluation. When these systems are evaluated, they are often tested through offline simulations with proxy metrics of interpretability (such as model complexity). We empirically evaluate the veracity of three common interpretability assumptions through a large scale human-subjects experiment with a simple "placebo explanation" control. We find that feature attribution explanations provide marginal utility in our task for a human decision maker and in certain cases result in worse decisions due to cognitive and contextual confounders. This result challenges the assumed universal benefit of applying these methods and we hope this work will underscore the importance of human evaluation in XAI research. Supplemental materials -- including anonymized data from the experiment, code to replicate the study, an interactive demo of the experiment, and the models used in the analysis -- can be found at: https://doi.pizza/challenging-xai.

The outbreak of COVID-19 has shocked the entire world with its fairly rapid spread and has challenged different sectors. One of the most effective ways to limit its spread is the early and accurate diagnosing infected patients. Medical imaging, such as X-ray and Computed Tomography (CT), combined with the potential of Artificial Intelligence (AI), plays an essential role in supporting medical personnel in the diagnosis process. Thus, in this article five different deep learning models (ResNet18, ResNet34, InceptionV3, InceptionResNetV2 and DenseNet161) and their ensemble, using majority voting have been used to classify COVID-19, pneumoni{\ae} and healthy subjects using chest X-ray images. Multilabel classification was performed to predict multiple pathologies for each patient, if present. Firstly, the interpretability of each of the networks was thoroughly studied using local interpretability methods - occlusion, saliency, input X gradient, guided backpropagation, integrated gradients, and DeepLIFT, and using a global technique - neuron activation profiles. The mean Micro-F1 score of the models for COVID-19 classifications ranges from 0.66 to 0.875, and is 0.89 for the ensemble of the network models. The qualitative results showed that the ResNets were the most interpretable models. This research demonstrates the importance of using interpretability methods to compare different models before making a decision regarding the best performing model.

Deep Neural Networks have achieved huge success at a wide spectrum of applications from language modeling, computer vision to speech recognition. However, nowadays, good performance alone is not sufficient to satisfy the needs of practical deployment where interpretability is demanded for cases involving ethics and mission critical applications. The complex models of Deep Neural Networks make it hard to understand and reason the predictions, which hinders its further progress. To tackle this problem, we apply the Knowledge Distillation technique to distill Deep Neural Networks into decision trees in order to attain good performance and interpretability simultaneously. We formulate the problem at hand as a multi-output regression problem and the experiments demonstrate that the student model achieves significantly better accuracy performance (about 1\% to 5\%) than vanilla decision trees at the same level of tree depth. The experiments are implemented on the TensorFlow platform to make it scalable to big datasets. To the best of our knowledge, we are the first to distill Deep Neural Networks into vanilla decision trees on multi-class datasets.

We propose an empirical measure of the approximate accuracy of feature importance estimates in deep neural networks. Our results across several large-scale image classification datasets show that many popular interpretability methods produce estimates of feature importance that are not better than a random designation of feature importance. Only certain ensemble based approaches---VarGrad and SmoothGrad-Squared---outperform such a random assignment of importance. The manner of ensembling remains critical, we show that some approaches do no better then the underlying method but carry a far higher computational burden.

We propose a general framework called Network Dissection for quantifying the interpretability of latent representations of CNNs by evaluating the alignment between individual hidden units and a set of semantic concepts. Given any CNN model, the proposed method draws on a broad data set of visual concepts to score the semantics of hidden units at each intermediate convolutional layer. The units with semantics are given labels across a range of objects, parts, scenes, textures, materials, and colors. We use the proposed method to test the hypothesis that interpretability of units is equivalent to random linear combinations of units, then we apply our method to compare the latent representations of various networks when trained to solve different supervised and self-supervised training tasks. We further analyze the effect of training iterations, compare networks trained with different initializations, examine the impact of network depth and width, and measure the effect of dropout and batch normalization on the interpretability of deep visual representations. We demonstrate that the proposed method can shed light on characteristics of CNN models and training methods that go beyond measurements of their discriminative power.

Supervised machine learning models boast remarkable predictive capabilities. But can you trust your model? Will it work in deployment? What else can it tell you about the world? We want models to be not only good, but interpretable. And yet the task of interpretation appears underspecified. Papers provide diverse and sometimes non-overlapping motivations for interpretability, and offer myriad notions of what attributes render models interpretable. Despite this ambiguity, many papers proclaim interpretability axiomatically, absent further explanation. In this paper, we seek to refine the discourse on interpretability. First, we examine the motivations underlying interest in interpretability, finding them to be diverse and occasionally discordant. Then, we address model properties and techniques thought to confer interpretability, identifying transparency to humans and post-hoc explanations as competing notions. Throughout, we discuss the feasibility and desirability of different notions, and question the oft-made assertions that linear models are interpretable and that deep neural networks are not.

Disentangling model activations into meaningful features is a central problem in interpretability. However, the absence of ground-truth for these features in realistic scenarios makes validating recent approaches, such as sparse dictionary learning, elusive. To address this challenge, we propose a framework for evaluating feature dictionaries in the context of specific tasks, by comparing them against supervised feature dictionaries. First, we demonstrate that supervised dictionaries achieve excellent approximation, control, and interpretability of model computations on the task. Second, we use the supervised dictionaries to develop and contextualize evaluations of unsupervised dictionaries along the same three axes. We apply this framework to the indirect object identification (IOI) task using GPT-2 Small, with sparse autoencoders (SAEs) trained on either the IOI or OpenWebText datasets. We find that these SAEs capture interpretable features for the IOI task, but they are less successful than supervised features in controlling the model. Finally, we observe two qualitative phenomena in SAE training: feature occlusion (where a causally relevant concept is robustly overshadowed by even slightly higher-magnitude ones in the learned features), and feature over-splitting (where binary features split into many smaller, less interpretable features). We hope that our framework will provide a useful step towards more objective and grounded evaluations of sparse dictionary learning methods.

Many interpretable AI approaches have been proposed to provide plausible explanations for a model's decision-making. However, configuring an explainable model that effectively communicates among computational modules has received less attention. A recently proposed shared global workspace theory showed that networks of distributed modules can benefit from sharing information with a bottlenecked memory because the communication constraints encourage specialization, compositionality, and synchronization among the modules. Inspired by this, we propose Concept-Centric Transformers, a simple yet effective configuration of the shared global workspace for interpretability, consisting of: i) an object-centric-based memory module for extracting semantic concepts from input features, ii) a cross-attention mechanism between the learned concept and input embeddings, and iii) standard classification and explanation losses to allow human analysts to directly assess an explanation for the model's classification reasoning. We test our approach against other existing concept-based methods on classification tasks for various datasets, including CIFAR100, CUB-200-2011, and ImageNet, and we show that our model achieves better classification accuracy than all baselines across all problems but also generates more consistent concept-based explanations of classification output.

The ability of neural networks to represent more features than neurons makes interpreting them challenging. This phenomenon, known as superposition, has spurred efforts to find architectures that are more interpretable than standard multilayer perceptrons (MLPs) with elementwise activation functions. In this note, I examine bilinear layers, which are a type of MLP layer that are mathematically much easier to analyze while simultaneously performing better than standard MLPs. Although they are nonlinear functions of their input, I demonstrate that bilinear layers can be expressed using only linear operations and third order tensors. We can integrate this expression for bilinear layers into a mathematical framework for transformer circuits, which was previously limited to attention-only transformers. These results suggest that bilinear layers are easier to analyze mathematically than current architectures and thus may lend themselves to deeper safety insights by allowing us to talk more formally about circuits in neural networks. Additionally, bilinear layers may offer an alternative path for mechanistic interpretability through understanding the mechanisms of feature construction instead of enumerating a (potentially exponentially) large number of features in large models.

As large language models (LLMs) grow more powerful, concerns around potential harms like toxicity, unfairness, and hallucination threaten user trust. Ensuring beneficial alignment of LLMs with human values through model alignment is thus critical yet challenging, requiring a deeper understanding of LLM behaviors and mechanisms. We propose opening the black box of LLMs through a framework of holistic interpretability encompassing complementary bottom-up and top-down perspectives. The bottom-up view, enabled by mechanistic interpretability, focuses on component functionalities and training dynamics. The top-down view utilizes representation engineering to analyze behaviors through hidden representations. In this paper, we review the landscape around mechanistic interpretability and representation engineering, summarizing approaches, discussing limitations and applications, and outlining future challenges in using these techniques to achieve ethical, honest, and reliable reasoning aligned with human values.

We respond to the recent paper by Makelov et al. (2023), which reviews subspace interchange intervention methods like distributed alignment search (DAS; Geiger et al. 2023) and claims that these methods potentially cause "interpretability illusions". We first review Makelov et al. (2023)'s technical notion of what an "interpretability illusion" is, and then we show that even intuitive and desirable explanations can qualify as illusions in this sense. As a result, their method of discovering "illusions" can reject explanations they consider "non-illusory". We then argue that the illusions Makelov et al. (2023) see in practice are artifacts of their training and evaluation paradigms. We close by emphasizing that, though we disagree with their core characterization, Makelov et al. (2023)'s examples and discussion have undoubtedly pushed the field of interpretability forward.

Understanding neural networks is challenging in part because of the dense, continuous nature of their hidden states. We explore whether we can train neural networks to have hidden states that are sparse, discrete, and more interpretable by quantizing their continuous features into what we call codebook features. Codebook features are produced by finetuning neural networks with vector quantization bottlenecks at each layer, producing a network whose hidden features are the sum of a small number of discrete vector codes chosen from a larger codebook. Surprisingly, we find that neural networks can operate under this extreme bottleneck with only modest degradation in performance. This sparse, discrete bottleneck also provides an intuitive way of controlling neural network behavior: first, find codes that activate when the desired behavior is present, then activate those same codes during generation to elicit that behavior. We validate our approach by training codebook Transformers on several different datasets. First, we explore a finite state machine dataset with far more hidden states than neurons. In this setting, our approach overcomes the superposition problem by assigning states to distinct codes, and we find that we can make the neural network behave as if it is in a different state by activating the code for that state. Second, we train Transformer language models with up to 410M parameters on two natural language datasets. We identify codes in these models representing diverse, disentangled concepts (ranging from negative emotions to months of the year) and find that we can guide the model to generate different topics by activating the appropriate codes during inference. Overall, codebook features appear to be a promising unit of analysis and control for neural networks and interpretability. Our codebase and models are open-sourced at https://github.com/taufeeque9/codebook-features.

Through considerable effort and intuition, several recent works have reverse-engineered nontrivial behaviors of transformer models. This paper systematizes the mechanistic interpretability process they followed. First, researchers choose a metric and dataset that elicit the desired model behavior. Then, they apply activation patching to find which abstract neural network units are involved in the behavior. By varying the dataset, metric, and units under investigation, researchers can understand the functionality of each component. We automate one of the process' steps: to identify the circuit that implements the specified behavior in the model's computational graph. We propose several algorithms and reproduce previous interpretability results to validate them. For example, the ACDC algorithm rediscovered 5/5 of the component types in a circuit in GPT-2 Small that computes the Greater-Than operation. ACDC selected 68 of the 32,000 edges in GPT-2 Small, all of which were manually found by previous work. Our code is available at https://github.com/ArthurConmy/Automatic-Circuit-Discovery.

Neural networks often exhibit emergent behavior, where qualitatively new capabilities arise from scaling up the amount of parameters, training data, or training steps. One approach to understanding emergence is to find continuous progress measures that underlie the seemingly discontinuous qualitative changes. We argue that progress measures can be found via mechanistic interpretability: reverse-engineering learned behaviors into their individual components. As a case study, we investigate the recently-discovered phenomenon of ``grokking'' exhibited by small transformers trained on modular addition tasks. We fully reverse engineer the algorithm learned by these networks, which uses discrete Fourier transforms and trigonometric identities to convert addition to rotation about a circle. We confirm the algorithm by analyzing the activations and weights and by performing ablations in Fourier space. Based on this understanding, we define progress measures that allow us to study the dynamics of training and split training into three continuous phases: memorization, circuit formation, and cleanup. Our results show that grokking, rather than being a sudden shift, arises from the gradual amplification of structured mechanisms encoded in the weights, followed by the later removal of memorizing components.

We show how to "compile" human-readable programs into standard decoder-only transformer models. Our compiler, Tracr, generates models with known structure. This structure can be used to design experiments. For example, we use it to study "superposition" in transformers that execute multi-step algorithms. Additionally, the known structure of Tracr-compiled models can serve as ground-truth for evaluating interpretability methods. Commonly, because the "programs" learned by transformers are unknown it is unclear whether an interpretation succeeded. We demonstrate our approach by implementing and examining programs including computing token frequencies, sorting, and parenthesis checking. We provide an open-source implementation of Tracr at https://github.com/google-deepmind/tracr.

The black-box nature of deep neural networks (DNN) has brought to attention the issues of transparency and fairness. Deep Reinforcement Learning (Deep RL or DRL), which uses DNN to learn its policy, value functions etc, is thus also subject to similar concerns. This paper proposes a way to circumvent the issues through the bottom-up design of neural networks with detailed interpretability, where each neuron or layer has its own meaning and utility that corresponds to humanly understandable concept. The framework introduced in this paper is called the Self Reward Design (SRD), inspired by the Inverse Reward Design, and this interpretable design can (1) solve the problem by pure design (although imperfectly) and (2) be optimized like a standard DNN. With deliberate human designs, we show that some RL problems such as lavaland and MuJoCo can be solved using a model constructed with standard NN components with few parameters. Furthermore, with our fish sale auction example, we demonstrate how SRD is used to address situations that will not make sense if black-box models are used, where humanly-understandable semantic-based decision is required.

In this paper we introduce a novel, unified, open-source model interpretability library for PyTorch [12]. The library contains generic implementations of a number of gradient and perturbation-based attribution algorithms, also known as feature, neuron and layer importance algorithms, as well as a set of evaluation metrics for these algorithms. It can be used for both classification and non-classification models including graph-structured models built on Neural Networks (NN). In this paper we give a high-level overview of supported attribution algorithms and show how to perform memory-efficient and scalable computations. We emphasize that the three main characteristics of the library are multimodality, extensibility and ease of use. Multimodality supports different modality of inputs such as image, text, audio or video. Extensibility allows adding new algorithms and features. The library is also designed for easy understanding and use. Besides, we also introduce an interactive visualization tool called Captum Insights that is built on top of Captum library and allows sample-based model debugging and visualization using feature importance metrics.

InterpretML is an open-source Python package which exposes machine learning interpretability algorithms to practitioners and researchers. InterpretML exposes two types of interpretability - glassbox models, which are machine learning models designed for interpretability (ex: linear models, rule lists, generalized additive models), and blackbox explainability techniques for explaining existing systems (ex: Partial Dependence, LIME). The package enables practitioners to easily compare interpretability algorithms by exposing multiple methods under a unified API, and by having a built-in, extensible visualization platform. InterpretML also includes the first implementation of the Explainable Boosting Machine, a powerful, interpretable, glassbox model that can be as accurate as many blackbox models. The MIT licensed source code can be downloaded from github.com/microsoft/interpret.

There has recently been a surge of work in explanatory artificial intelligence (XAI). This research area tackles the important problem that complex machines and algorithms often cannot provide insights into their behavior and thought processes. XAI allows users and parts of the internal system to be more transparent, providing explanations of their decisions in some level of detail. These explanations are important to ensure algorithmic fairness, identify potential bias/problems in the training data, and to ensure that the algorithms perform as expected. However, explanations produced by these systems is neither standardized nor systematically assessed. In an effort to create best practices and identify open challenges, we provide our definition of explainability and show how it can be used to classify existing literature. We discuss why current approaches to explanatory methods especially for deep neural networks are insufficient. Finally, based on our survey, we conclude with suggested future research directions for explanatory artificial intelligence.