主动学习简介

foreign

[Music]

amounts of data thanks to the

superpowers of large models including

the famous chatgpt but also Vision

models and all other types you may be

working with indeed the secrets behind

those models is not only the large

amount of data they are being trained on

but also the quality of that data but

what does this mean it means we need

lots of very good balance and varied

data and as data scientists we all know

how complicated and painful it can be to

build such a good data set fast and at

large scale and maybe with a limited

budget what if we could have helped

build that or even have automated help

well that is where Active Learning comes

in in one sentence the goal of active

learning is to use the least amount of

training data to optimize The annotation

of your whole data set and train the

best possible model it's a supervised

learning approach that will go back and

forth between your model's predictions

and your data what I mean here is that

you may start with a small batch of

curated annotated data and train your

model with it you don't have to wait for

your whole millions of images that are

set to be ready just push it out there

then using Active Learning you can use

your model on your unseen data and get

human annotators to label it but that is

not only it we can also evaluate how

accurate the predictions are and using a

variety of acquisition functions which

are functions used to select the next

unseen data to annotate we can quantify

the impact of labeling a larger data set

volume or improving the accuracy of the

labels generated to improve the model's

performance thanks to how you train the

models you can analyze the confidence

they have in their predictions

predictions with low confidence will

automatically request additional images

of this type to be labeled and

predictions with high confidence won't

need additional data so you will

basically save a lot of time and money

by having to annotate fewer images in

the end and have the most optimized

model possible how cool is that Active

Learning is one of the most promising

approach to working with large-scale

data sets and there are a few important

key Notions to remember with active

learning the most important is that it

uses humans which you can clearly see

here in the middle of this great

presentation of active learning it will

still require humans to annotate data

which has the plus side to give you full

control over the quality of your model's

prediction it's not a complete Black Box

that trained with millions of images

anymore you iteratively follow its

development and help it get better when

it fails of course it does have the

downside of increasing costs versus

unsupervised approaches where you don't

need anyone but it allows you to limit

those costs by only training where the

models need it instead of feeding it as

much data as possible and hoping for the

best moreover the reduction in time

taken to train the model and put it into

production often outweighs these costs

and you can use some automatic

annotation tools and manually correct it

after again reducing the costs then

obviously you will have your labeled

data set the labeled set of data is what

your current model is being trained on

and the unlabeled set is the data you

could put in usually used but hasn't

been annotated yet another key notion is

actually the answer to the most

important question you may already have

in mind how do you find the bad data to

annotate and add to the training set

the solution here is called query

strategies and they are essential to any

Active Learning algorithm deciding which

data to label and which not to there are

multiple possible approaches to finding

the most informative subsets in our

large pool of unlabeled data that will

most help our model by being annotated

like uncertainty sampling where you test

your current model on your unlabeled

data and draw the least confident

classified examples to annotate another

technique shown here is the query by

committee or QBC approach here we have

multiple models our committee models

they will all be trained on a different

subset of our label data and thus have a

different understanding of our problem

these models will each have a hypothesis

on the classification of our unlabeled

data that should be somewhat similar but

still different because they basically

see the world differently just like us

that have different live experience and

have seen different animals in our lives

but still have the same concepts of a

cat and a dog then it's easy the data to

be annotated is simply the ones our

models most disagree on which means it

is complicated to understand and we

start over by feeding the selected data

to our experts for annotation this is of

course a basic explanation of active

learning with only one example of a

query strategy let me know if you'd like

more videos on other machine learning

strategies like this here A clear

example of the active learning process

is when you answer captchas on Google it

helps you identify complex images and

build data sets using you and many other

people as a committee jury for

annotation

building cheap and great data sets while

entering you are a human serving two

purposes so next time you are annoyed by

a captcha just think that you are

helping AI models progress but we have

enough theory for now I thought it would

be great to partner with some friends

from encord a great company I have known

for a while now to Showcase a real

example of active learning since we are

in this team it's for sure the best

platform I have seen yet for active

learning and the team is amazing before

diving into a short practical example I

just wanted to mention that I will be at

cvpr in person this year and so will

Encore if you are attending in person 2

let me know and go check out their Booth

it's Booth 1310. here's a quick demo we

put together for exploring one of

encore's products that perfectly fits

this episode and chord active it is

basically an active learning platform

where you can perform everything we

talked about in this video without any

coding with a great visual interface

here's what you would see in a classic

visual task like segmentation once you

open up your project you directly have

relevant information and statistics

about your data you'll see all the

outlier characteristics of your data

which will help you figure out what

causes the issues in your test for

example here we see that blur is one of

those outliers that has been

automatically identified if we check out

the worst images for that category we

can easily find some problematic images

and tag them for review like here where

the image is super saturated you can

also visualize groups of data thanks to

their embeddings just like clip

embeddings that you might have heard a

lot these days and those embeddings can

easily be compared together and grouped

when similar helping you find

problematic groups all at once instead

of going through your data one by one

then once you are satisfied with your

identified images to review you can

simply export it to their encode

platform where you can do your

annotation directly when you have your

annotations and you get back on the

encode active platform you can now

visualize what it looks like with labels

you can see how the embedding plots have

changed now with the different classes

attached here again you can look at

different subgroups of data to find

problematic ones for example you can

look at images containing school buses

this can be done using natural language

to look for any information in images

metadata or classes something quite

necessary these days if you want to say

that you are working in AI when you

cannot find any more problems easily

with your data you train your model and

come back to the platform to analyze its

performance once again you have access

to a ton of valuable information about

how well your model is performing for

example if we take a look at the object

area where we see that small images seem

problematic we can easily filter them

out and create a new sub data set using

only our problematic small object images

the project is created in your Encore

active dashboard with all the same

statistics you had but for only this set

of data if you want to have a closer

look or run experiments with this more

complicated part of the data like using

it for training one of your committee

models and you repeat this Loop over and

over on the annotating problematic data

and improving your model as efficiently

as possible it will both reduce the need

for paying experts annotators especially

if you work with medical applications as

I do or other applications where experts

are quite expensive and maximize the

results of your model I hope you can now

see how valuable Active Learning can be

and maybe even try it out with your own

application and it can all be done with

a single product if you want to let me

know if you do so

but before ending this video I just

wanted to thank ankord for sponsoring

this week's episode with a great example

of active learning and an amazing

product I also wanted to point out that

they had a webinar on June 14th on how

to build a semantic search for visual

data using chatgpt and clip that is

housed on encode active with a recording

available if you want to check it out

it's definitely worthwhile and super

interesting I hope you enjoyed this

episode format as much as I enjoyed

making it thank you for watching