title
stringlengths 12
112
| published
stringlengths 19
23
| url
stringlengths 28
28
| video_id
stringlengths 11
11
| channel_id
stringclasses 5
values | id
stringlengths 16
31
| text
stringlengths 0
596
| start
float64 0
37.8k
| end
float64 2.18
37.8k
|
|---|---|---|---|---|---|---|---|---|
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t35.52
|
perceptron and neural networks, a lot of that was researched and discovered
| 35.52 | 46.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t41.68
|
back in the 50s and 60s and 70s, but we
| 41.68 | 50.76 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t47.160000000000004
|
didn't see that really applied in industry or
| 47.16 | 55.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t51.56
|
anywhere really until the past decade and
| 51.56 | 59.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t55.16
|
there are two main reasons for this.
| 55.16 | 64.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t59.72
|
So the first is that we didn't have enough compute power back in the
| 59.72 | 71.48 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t64.8
|
50s, 60s, 70s to train the models that we needed to train and we also didn't have the data to actually
| 64.8 | 73.84 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t71.84
|
train those models. Now
| 71.84 | 76.28 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t74.28
|
compute power is
| 74.28 | 83.64 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t76.32
|
not really a problem anymore. We sort of look at this graph, it depends on what model you're training
| 76.32 | 88.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t83.64
|
of course if you are open AI and you're training GPT-4 or 5 or whatever
| 83.64 | 96.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t89.96000000000001
|
yeah, maybe compute power is pretty relevant, but for most of us we can get access to
| 89.96 | 102.08 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t97.96000000000001
|
cloud machinery, personal machines and we can wait
| 97.96 | 108.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t102.68
|
a few hours or a couple of days and fine tune or pre-train a transform model
| 102.68 | 111.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t109.6
|
that is
| 109.6 | 113.6 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t111.4
|
good performance for what we need.
| 111.4 | 120.12 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t115.44000000000001
|
Now that obviously wasn't always the case until very recently back in
| 115.44 | 124.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t120.88000000000001
|
1960s, you see on this graph here we have the IBM
| 120.88 | 127.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t125.16000000000001
|
704 and
| 125.16 | 133.64 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t127.72
|
you can see under the Y axes we have floating point operations per second
| 127.72 | 136.56 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t134.56
|
and that's a logarithmic scale.
| 134.56 | 142.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t136.56
|
So linear scale just basically looks like a straight line until a few years ago and then shoots up.
| 136.56 | 144.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t143.36
|
It's
| 143.36 | 149.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t144.68
|
pretty impressive how much progress is made in terms of
| 144.68 | 152.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t150.52
|
computing power. Now
| 150.52 | 156.88 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t153.08
|
like I said, that's not really an issue for us anymore.
| 153.08 | 161.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t156.88
|
We have the compute in most cases to do what we need to do and
| 156.88 | 167.04 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t161.16
|
and data is not as much of a problem anymore, but we'll talk about that in a moment.
| 161.16 | 170.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t168.04
|
So data again, we have a very
| 168.04 | 179 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t171.64
|
big increase in data, not quite as big as the computing power and this graph here doesn't go quite as far back.
| 171.64 | 181 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t179.0
|
It's only 2010
| 179 | 186.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t181.56
|
where I believe it was at 2 zettabytes and now
| 181.56 | 189.2 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t187.2
|
71 or
| 187.2 | 196.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t189.2
|
so in 2021. So there's a fairly big increase, not quite as much as computing power
| 189.2 | 201.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t197.28
|
over time, but still pretty massive. Now the
| 197.28 | 204.44 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t202.44
|
thing with data is
| 202.44 | 210.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t204.76
|
yes, there's a lot of data out there, but is there that much data out there for
| 204.76 | 214.08 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t211.12
|
what we need to train models to do and
| 211.12 | 216.44 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t214.44
|
in a lot of cases, yes, there is.
| 214.44 | 219.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t216.44
|
But it really depends on what you're doing. If you are
| 216.44 | 223.12 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t221.12
|
focusing on
| 221.12 | 230.48 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t223.12
|
a more niche domain. So what I have here on the left over here are a couple of niche domains.
| 223.12 | 234.2 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t231.2
|
There's not that much data out there on
| 231.2 | 240.84 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t235.0
|
sentence pairs for climate evidence and claims, for example. So where you have a
| 235 | 250.08 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t240.84
|
piece of evidence and a claim and whether the claim supports evidence or not, there is a very small data set called climate fever data set, but it's not big.
| 240.84 | 256.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t251.76
|
For agriculture, I assume within that industry, there's not that much data, although I
| 251.76 | 259.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t256.96
|
have never worked in that industry. So I
| 256.96 | 263.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t260.28000000000003
|
am not fully aware. I just assume there's probably not that much.
| 260.28 | 267.24 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t264.36
|
And then also niche finance, which I do
| 264.36 | 273.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t267.24
|
at least have a bit more experience with and I imagine this is probably something that a lot of you will find useful as well.
| 267.24 | 276.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t274.72
|
Because
| 274.72 | 280.96 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t276.16
|
finance is a big industry. There's a lot of finance data out there, but there's a lot of niche
| 276.16 | 286.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t282.0
|
projects and problems in finance where you find much less data.
| 282 | 294.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t287.76
|
So yes, we have a lot more data nowadays, but we don't have enough for a lot of the data that we need.
| 287.76 | 299.32 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t294.4
|
On the right here, we have a couple of examples of low resource data sets. So we have
| 294.4 | 306.32 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t299.79999999999995
|
Adave from the Maldives and also the Navajo languages as well. So with these
| 299.8 | 311.44 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t307.35999999999996
|
we kind of need to find a different approach. Now we can
| 307.36 | 319.28 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t312.0
|
investigate, depending on your use case, unsupervised learning, TSEA, which we have covered in a previous video article and
| 312 | 324.64 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t319.28
|
that does work when you're trying to build a model that recognizes
| 319.28 | 328.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t325.28
|
generic similarity. It works very well as well.
| 325.28 | 334.32 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t328.71999999999997
|
But for example with the climate claims data, we are not necessarily trying to
| 328.72 | 338.88 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t334.96
|
match sentence A and B based on their semantic similarity.
| 334.96 | 345.76 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t338.88
|
But we're trying to match sentence A, which is a claim, to sentence B, which is a claim.
| 338.88 | 352.8 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t345.76
|
As to whether that evidence supports the claim or not.
| 345.76 | 359.2 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t353.52
|
So in that case, unsupervised approach like TSEA doesn't really work.
| 353.52 | 366.4 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t360.4
|
So what we have is very little data and there aren't really
| 360.4 | 372.72 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t367.2
|
any alternative training approaches that we can use. So basically what we need to do is
| 367.2 | 377.2 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t372.72
|
create more data. Now data orientation is
| 372.72 | 385.04 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t378.08000000000004
|
difficult, particularly for language. So data orientation is not specific to NLP.
| 378.08 | 391.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t385.04
|
It's used across ML and it's more established in the field of computer vision.
| 385.04 | 400.96 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t392.24
|
And that makes sense because computer vision, say you have an image, you can modify that image using a few different approaches.
| 392.24 | 406.08 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t400.96
|
And a person can still look at that image and think, OK, that is the same image.
| 400.96 | 411.2 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t406.08
|
It's just maybe it's rotated a little bit. We've changed the color grading, the brightness,
| 406.08 | 418.48 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t412.32
|
or something along those lines. We just modified it slightly. But it's still in essence the same image.
| 412.32 | 428.48 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t420.08
|
Now for language it's a bit difficult because language is very abstract and nuanced.
| 420.08 | 435.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t428.48
|
So if you start randomly changing certain words, the chances are you're going to produce something
| 428.48 | 441.12 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t435.68
|
that doesn't make any sense. And we, when we're augmenting our data, we don't want to just throw
| 435.68 | 450.64 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t441.12
|
rubbish into our model. We want something that makes sense. So there are some data augmentation
| 441.12 | 459.44 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t450.64
|
techniques. And we'll look at a couple of the simpler ones now. So there is a library called NLP.org
| 450.64 | 465.6 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t459.44
|
which I think is very good for this sort of thing. It's essentially a library that allows us to do
| 459.44 | 478.48 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t465.59999999999997
|
data augmentation for NLP. And what you can see here is two methods using word2vec vectors and
| 465.6 | 486.24 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t478.48
|
similarity. And what we're doing is taking this original sentence. So the quick brown fox jumps
| 478.48 | 493.36 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t486.24
|
over the lazy dog. And we're just inserting some words using word2vec. So we're trying to find
| 486.24 | 499.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t493.36
|
what words word2vec thinks could go in here, which words are the most similar to the surrounding
| 493.36 | 508.32 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t499.52000000000004
|
words. And we have this al-Ziari, which I don't know. I think it seems like a name to me. But I
| 499.52 | 516 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t508.32
|
am not sure. That I don't think really fits there. So it's not great. It's not perfect.
| 508.32 | 524.24 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t517.6
|
Lazy superintendents dog. That does kind of make sense. I feel like a lazy superintendents dog is
| 517.6 | 531.92 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t525.52
|
maybe a stereotype or I'm sure it's been in The Simpsons or something before. So,
| 525.52 | 537.76 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t532.56
|
okay, fair enough. I can see how that can fit in there. Which again, it's a bit weird. It's not
| 532.56 | 545.76 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t537.76
|
great. Substitution for me seems to work better. So rather than the quick brown fox, we have the
| 537.76 | 551.52 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t545.76
|
easy brown fox. And rather than jumping over the lazy dog, jumps around the lazy dog. Which
| 545.76 | 559.68 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t551.52
|
changes the meaning slightly. Easy is a bit weird there to be fair. But we still have a sentence
| 551.52 | 568.16 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t559.68
|
that kind of makes sense. So that's good, I think. Now we don't have to use words to vet. We can also
| 559.68 | 575.76 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t568.16
|
use contextual word embeddings like with Bert. And for me, I think the results look better. So
| 568.16 | 583.04 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t575.76
|
for insertion, we get even the quick brown fox usually jumps over the lazy dog. So we're adding
| 575.76 | 589.92 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t583.04
|
some words there. It makes sense. That's I think good for substitution. And we're only doing one
| 583.04 | 596.96 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t589.92
|
word here. And we're changing that to a little quick brown fox instead of just quick brown fox.
| 589.92 | 604.32 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t597.68
|
So I think that makes sense. And this is a good way of augmenting your data and bring more data
| 597.68 | 615.6 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t604.32
|
from less. But for us, because we are using sentence pairs, we can basically just take
| 604.32 | 624.8 |
Making The Most of Data: Augmented SBERT
|
2021-12-17 14:24:40 UTC
|
https://youtu.be/3IPCEeh4xTg
|
3IPCEeh4xTg
|
UCv83tO5cePwHMt1952IVVHw
|
3IPCEeh4xTg-t616.8000000000001
|
all of the data from say we have A and B over here. Imagine this is a data frame.
| 616.8 | 635.04 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.