Supervised Contrastive Learning

Поделиться
HTML-код
  • Опубликовано: 3 ноя 2024

Комментарии • 73

  • @ghostlv4030
    @ghostlv4030 4 года назад +7

    The paper delivers the main idea clearly and effectively, it is that they are rich ! ! !

  • @srivatsabhargavajagarlapud2274
    @srivatsabhargavajagarlapud2274 4 года назад +6

    It would have been great to see if this (pre-training) method could achieve(as a by-product) representations that honor semantic similarity based inter-class representation distance amongst classes. By this I mean, for example, cats are more similar in a semantic sense to dogs, than are cars/trucks to dogs so, after pre-training here, though you haven't explicitly sought for this in your loss(both in this supervised-contrastive other losses such as triplet losses more commonly used in siamese nets), do you by any chance see d(cat,dog)

    • @YannicKilcher
      @YannicKilcher  4 года назад +2

      there is a hierarchy in imagenet, so this would actually be feasible (and I'm sure people have done this) :D

  • @bradhatch8302
    @bradhatch8302 4 года назад +1

    Listening at 1.75 speed it’s like I read and understood this paper in about 18 mins. Mucho thanks!

    • @louislouis7388
      @louislouis7388 3 года назад +1

      The paper tried to make it complicated. Not interesting direction, it is not advantage to self-supervised learning at all. Just wasting my time to read that paper.

    • @Metalwrath2
      @Metalwrath2 3 года назад +1

      @@louislouis7388 Lots of papers do that. One of the reasons why I don't like academy

  • @ruitao2099
    @ruitao2099 3 года назад +3

    Nice talk! But I still confused about the motivation of supervised contrastive learning. What were the differences between it with normal supervised learning. We could get the embedded space by training a deep supervised model and take the feature layers out and put them into different work. Thanks for your replying!

    • @shaikrasool1316
      @shaikrasool1316 3 года назад

      Contrastive supervised learning is used to compare two images, example:- siamese network

  • @aliawad2244
    @aliawad2244 Год назад +1

    What an elegant explanation. Huge thanks!

  • @mhadnanali
    @mhadnanali 2 года назад

    Thank you very much. I was stuck on some problem in my contrastive learning paper implementation. your explanation helped me understand better.

  • @herp_derpingson
    @herp_derpingson 4 года назад +19

    This doesnt sound very novel to me. I swear I saw something similar in an introductory ML course.
    Regardless, I wonder how much of that 1% is from this algorithm and how much is from raw GPU power.

    • @YannicKilcher
      @YannicKilcher  4 года назад

      Yes, I agree. This will have to be replicated before I believe it.

    • @kaushikroy4041
      @kaushikroy4041 4 года назад +5

      Herp Derpingson This sounds like supervised metric learning to me. Then take last but one layer. Done before to my mind.

    • @markdaoust4598
      @markdaoust4598 4 года назад +2

      Yes. Isn’t this “center loss”: ydwen.github.io/papers/WenECCV16.pdf

    • @delikatus
      @delikatus 4 года назад +1

      ​@@markdaoust4598 As said at 26:08, isn't it also pretty much the same thing as "siamese networks" / "triplet loss"? arxiv.org/pdf/1503.03832.pdf
      Also see: yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf and probably there's some Schmidhuber stuff that's exactly the same, too? :D
      Also relevant:
      arxiv.org/abs/1907.13625
      and
      arxiv.org/pdf/2003.08505.pdf

  • @philippeisen1910
    @philippeisen1910 4 года назад +12

    Nice level of detail for going over papers - really appreciate your work!
    Im curious, what is your setup to create those nice visualizations?

  • @生活空間
    @生活空間 2 года назад +1

    In supervised contrastive loss, the augmented view of images seem not necessary.
    But without the two-crop-transform augmentation, the accuracy of CIFAR-10, CIFAR-100, tinyImageNet will drop down 3% ~ 5% depend on the tasks.

  • @msfasha
    @msfasha Год назад

    Elegant and appreciated, thanx for the effort

  • @soufianekun11
    @soufianekun11 4 года назад +11

    I wonder why they didn't use the triplet loss of a siamese network ??!

    • @grb321
      @grb321 2 года назад

      The claim in the paper is that supervised contrastive loss is a lot more robust than triplet loss, which usually requires some form of negative example mining to work well. The authors also claim that supervised contrastive loss makes hyperparameter tuning easier, as classification performance is less sensitive on hyperparameter settings.

  • @abirnaskar3458
    @abirnaskar3458 10 месяцев назад

    Nice, I was wondering how it will work in text, I mean if I replace transformers with this.
    Is there any paper which use transformer based model along with contrastive learning?

  • @kalastasurepe
    @kalastasurepe 2 года назад +1

    Thanks a lot! I liked it so much! You explained it in a very simple way even though all these are very complex.

  • @patrickjdarrow
    @patrickjdarrow 4 года назад +2

    Couldn't you use a standard training epoch as a proxy for mining hard negatives? Before each next epoch, take the top n lossy samples to use for contrastive learning.

    • @iloos7457
      @iloos7457 2 года назад

      probably, a good extention to the triplet loss.. but perhaps unnessecary for supcon. I feel like Supcon tries to solve the hard-negative with contrastive learning

  • @XecutionStyle
    @XecutionStyle 3 года назад +1

    But that's the point right. ImageNet percentages had saturated regardless of hardware. This answers can we be more efficient just as much as can we incorporate more compute.

  • @thongnguyen1292
    @thongnguyen1292 4 года назад +2

    1:30 "Supervised learning is the only thing right now in deep learning that works"
    Woaah who is making the big claim here :D

  • @SparshGarg-n8e
    @SparshGarg-n8e 7 месяцев назад

    Thank you soo much!

  • @loveislulu264
    @loveislulu264 9 месяцев назад

    can you provide a vid for implementation of supervised contrastive learning

  • @robertchamoun7914
    @robertchamoun7914 3 года назад +1

    Great explanation thank you!!
    Can Someone please explain to me what would be the benefit of contrastive pre-training compared to Autoencoder Pretraining for CNN ?

    • @abedog90210
      @abedog90210 3 года назад +2

      I think maybe because here (with contrastive loss) you're explicitly training your model to cluster the same images together,
      whereas in autoencoder pretraining you're training the encoder to extract useful features for reconstruction of the same image, hoping that images from the same class will have similar features in that latent space, but you're not explicitly telling it to do so.

    • @robertchamoun7914
      @robertchamoun7914 3 года назад

      thanks for the explanation.

  • @AKSINHA00
    @AKSINHA00 3 года назад +1

    hey.. Does contrastive loss on self-supervised learning require the presence of minimal positive samples in the denominator of loss function? would this make it harder to deploy this in live unlabelled data or random samples?

    • @YannicKilcher
      @YannicKilcher  3 года назад +1

      The numerator is always included in the denominator, so you have some positive samples by construction

  • @johnkrafnik5414
    @johnkrafnik5414 4 года назад

    Great stuff. Im impressed how many videos you have put up.

  • @shreejaltrivedi9731
    @shreejaltrivedi9731 3 года назад +1

    Great video Yannic.
    I was curious about one thing. Here in Contrastive Pretraining whether it is supervised/unsupervised, they do the different augmentations and then do the pretraining. What if we do the same augmentations for every image in my labeled dataset that Unsupervised Contrastaive Pretraining uses and train the network on this new augmented dataset in the simple supervised fashion accompanying the cross-entropy loss? . At the end of the day supervision and mass of the data matters in DL is the best path to achieve commendable results.
    What are your thoughts?

    • @YannicKilcher
      @YannicKilcher  3 года назад

      I don't know, but it's a good idea, maybe worth a try

  • @igorl01
    @igorl01 Год назад

    Amazing explanations!

  • @mattiasfagerlund
    @mattiasfagerlund 2 года назад

    You can still use unlabeled data for the negative samples, because the odds of them being in the same class is miniscule?

  • @aday7475
    @aday7475 2 года назад

    Thank you for the clear, great explanation!

  • @davidcato6192
    @davidcato6192 4 года назад +1

    Excellent explanation, thank you!

  • @DistortedV12
    @DistortedV12 4 года назад +3

    Yannic are you going to ICLR 2020?

    • @YannicKilcher
      @YannicKilcher  4 года назад +4

      If you mean whether I'll be sitting on my couch and on the internet, then yes :D I'll probably follow the interesting bits, panels and such

  • @amirafsharmoshtaghpour8895
    @amirafsharmoshtaghpour8895 3 года назад +2

    Another excellent paper explanation. Around 23:00, I wonder why a hard positive amounts to = 0, not = -1.

    • @YannicKilcher
      @YannicKilcher  3 года назад +1

      -1 would be as much aligned as +1

    • @GradientDude
      @GradientDude 3 года назад

      @@YannicKilcher I also noticed that. I don't agree. The sign does matter here. And, if = -1 , then the loss will be pretty high for such a pair, because exp (z_i • z_p /τ ) = exp(-1) which is much much smaller than exp(1), and in this case denominator on Eq.4 will prevail.
      Actually all the derivations in the supplementary break apart (maybe there is a mistake somewhere) if you consider hard positive = -1 and hard negative = 1.
      I'm very surprised that nobody noticed such a flaw in the paper.

    • @vzoryan1769
      @vzoryan1769 3 года назад

      @@GradientDude they kinda leave this out, but in a high-dimensional space the probability of two random vectors being orthogonal is close to 1. Therefore, it's improbable that a positive example will face the opposite direction and you don't need to account for that. You can do a little numerical simulation and see for yourself.

  • @jonatan01i
    @jonatan01i 4 года назад

    So, pre-train on a HUGE image dataset with self-supervised contrastive learning and then start with this network to pre-train on your dataset with supervised contrastive and then can come softmax.

  • @ai_station_fa
    @ai_station_fa 2 года назад

    Thanks! I just wanted to ask if you could make more videos that you actually code in them. I learned a lot from them.

  • @yataoabian7094
    @yataoabian7094 4 года назад +1

    nice talk, Yannic!!

  • @waterflarz
    @waterflarz 4 года назад +2

    Great paper review! What software do you use for pdf annotation and recording?

  • @LouisChiaki
    @LouisChiaki 3 года назад +1

    I was about to try this on the Kaggle competition until I saw their batch size...

  • @GradientDude
    @GradientDude 4 года назад +2

    Hey! Thanks for the review. Which software do you use to annotate and draw on pdfs ?

  • @theBatchNorm
    @theBatchNorm 4 года назад

    Thank you for the enjoyable explanation

  • @eduarddurech5188
    @eduarddurech5188 4 года назад +2

    Yannic what did you mean by, "Supervised learning is the only thing right now in deep learning that works"? ;) Thank you for the videos btw!

  • @Shujaat-Khan
    @Shujaat-Khan 2 года назад

    Nice explanation 👌

  • @JapiSandhu
    @JapiSandhu 2 года назад

    Good explanation thanks

  • @florianhonicke5448
    @florianhonicke5448 4 года назад +1

    very informative

  • @songmeishu5445
    @songmeishu5445 3 года назад

    super great video!!!

  • @lucazzo1990
    @lucazzo1990 3 года назад

    Thank you, very helpful!

  • @reginaldanderson7218
    @reginaldanderson7218 4 года назад +1

    Cool content

  • @jinusbordbar1264
    @jinusbordbar1264 3 года назад

    TnX

  • @wyalexlee8578
    @wyalexlee8578 4 года назад

    Thank you for explaining this!

  • @Manu-em6ed
    @Manu-em6ed 3 года назад +1

    isn't that just normal supervised learning with extra steps ? :-P

  • @yosefricardochmulek2822
    @yosefricardochmulek2822 4 года назад +1

    You gotta be a time traveler.