Questions tagged [self-supervised-learning]
For questions related to self-supervised learning (SSL), which typically refers to techniques that automatically generate the supervisory learning signal. SSL can be used for representation learning, so it can be useful for transfer learning too. Some people consider SSL a sub-field of unsupervised learning given that many (if not all) SSL techniques do not require a human to manually annotate the inputs.
29
questions
0
votes
0
answers
13
views
Self-supervised learning weights initialization "after" projection head
For most Self-supervised learning algorithms: SimCLR, MoCo, BYOL, SimSiam, SwAV, etc., its common to have a projection head after the base encoder (which in most cases is a vanilla ResNet-50 CNN). An ...
- 235
0
votes
0
answers
24
views
How are conditional models different from supervised models?
I'm wondering what the difference between conditional learning and supervised learning is - especially in diffusion models? Am I correct to assume that diffusion models are supervised because in ...
- 1
1
vote
2
answers
62
views
How do Energy Based Models solve Multiple possible outputs given one input
I've been looking into Energy Based Models recently which Yann LeCun has been strongly advocating for. One problem that he lists with probabilistic based models is that in the case when there are ...
0
votes
1
answer
98
views
How can I combine unsupervised learning with supervised learning?
I am currently using an isolation forest (from sklearn library) to detect anomalies in a data frame (basically it's a dynamic data frame more of a kind of time series I am. But I have certain criteria ...
- 111
2
votes
2
answers
290
views
Definition of negatives in NT-Xent loss
I'm trying to understand few details about NT-Xent loss defined in SimCLR paper(link). The loss is defined as
$$\mathcal{l}_{i,j} = -\log\frac{\exp(sim(z_i,z_j)/\tau)}{\sum_{k=1}^{2N}\mathbb{1}_{[k\...
- 307
0
votes
2
answers
240
views
BYOL: Why is there a prediction network in the online network but not in the target network?
In the BYOL paper, the following architecture is presented:
Why is a prediction network added to the online network, which is not present in the target network?
How are the online and target compared ...
- 2,620
17
votes
4
answers
12k
views
What is the difference between self-supervised and unsupervised learning?
What is the difference between self-supervised and unsupervised learning? The terms logically overlap (and maybe self-supervised learning is a subset of unsupervised learning?), but I cannot pinpoint ...
- 2,620
1
vote
0
answers
25
views
Should we consider the prototypical forecasting task as self-supervised learning?
In NLP, the task of "predicting the next word" is an example of self-supervised learning. An essential part is that the label can be computed programmaticaly and does not require explicit ...
- 21
0
votes
1
answer
38
views
Self Supervised Learning Application of trained model... A bit confused
I am trying to apply a self supervised task as stated in this github repo.The Self-Supervised Sketch Recognition
In this work, authors are using 345.000 image samples to train the model and the ...
- 1
1
vote
0
answers
24
views
How can (pretrained) language models actively seek additional training data - possibly reference request?
I am reading the paper "Large Language Models Can Self-Improve" https://arxiv.org/abs/2210.11610 in which the authors consider that LLM can generate Chain-of-Thoughts sequences and even ...
- 863
1
vote
1
answer
75
views
Deep Clustering Approach for Unsupervised Video Anomaly Detection
I'm working on Unsupervised Video Anomaly Detection, and I've tried implementing the Generative Cooperative Learning method, with the help of this paper.
The method uses a fixed backbone (ResNext-101) ...
- 141
0
votes
1
answer
638
views
1D Sequence Classification with self-supervised learning
I am working on a multi-class classification task on long one-dimensional sequences. The sequence length may vary in the range $[512, 30720]$, and there is one feature associated each time-step in the ...
- 101
0
votes
2
answers
247
views
How do multimodal models establish connections between different modes?
I am specifically interested in data2vec, Meta's new model that can convert image, text, and sound data into a unified neural network representation. To my understanding, they did this through self-...
- 101
2
votes
1
answer
359
views
Perform clustering on high dimensional data
Recently I trained a BYOL model on a set of images to learn an embedding space where similar vectors are close by. The performance was fantastic when I performed approximate K-nearest neighbours ...
- 31
1
vote
0
answers
57
views
How do the scale of an embedding affects a downstream task?
I am currently training a neural network in a self-supervised fashion, using Contrastive Loss and I want to use that network then to fine-tune it in a classification task with a small fraction of the ...
- 11
2
votes
2
answers
437
views
Is it realistic to train a transformer-based model (e.g. GPT) in a self-supervised way directly on the Mel spectrogram?
In music information retrieval, one usually converts an audio signal into some kind "sequence of frequency-vectors", such as STFT or Mel-spectrogram.
I'm wondering if it is a good idea to ...
- 452
2
votes
0
answers
192
views
Does Yann LeCun consider k-means self-supervised learning?
I was discussing the topic of self-supervised learning with a colleague. After a while we realized we were using different definitions. That's never helpful.
Both of us were introduced to self-...
- 121
2
votes
1
answer
109
views
What are some most promising ways to approximate common sense and background knowledge?
I learned from this blog post Self-Supervised Learning: The Dark Matter of Intelligence that
We believe that self-supervised learning (SSL) is one of the most promising ways to build such background ...
- 947
2
votes
1
answer
140
views
Is it possible to use self-supervised learning on different images for the pretext and downstream tasks?
I have just come across the idea of self-supervised learning. It seems that it is possible to get higher accuracies on downstream tasks when the network is trained on pretext tasks.
Suppose that I ...
- 1,291
2
votes
1
answer
316
views
Is it possible to pre-train a CNN in a self-supervised way so that it can later be used to solve an instance segmentation task?
I would like to use self-supervised learning (SSL) to learn features from images (the dataset consists of similar images with small differences), then use the resulting trained model to bootstrap an ...
- 21
3
votes
1
answer
2k
views
How to generate labels for self-supervised training?
I've been reading a lot lately about self-supervised learning and I didn't understand very well how to generate the desired label for a given image.
Let's say that I have an image classification task, ...
- 161
1
vote
1
answer
65
views
Is there a way to get landmark features automatically learned by a neural network?
Is there a way to get landmark features automatically learned by a neural network without having to manually pre-label them in the images that are being fed into the network?
- 111
2
votes
1
answer
1k
views
What is the difference between distant supervision and self-supervision?
Weak supervision is supervised learning, with uncertainty in the labeling, e.g. due to automatic labeling or because non-experts labelled the data [1].
Distant supervision [2, 3] is a type of weak ...
- 163
3
votes
1
answer
350
views
Does self-supervised learning require auxiliary tasks?
Self-supervised learning algorithms provide labels automatically. But, it is not clear what else is required for an algorithm to fall under the category "self-supervised":
Some say, self-...
- 163
6
votes
2
answers
1k
views
How to understand the concept of self-supervised learning in AI?
I am new to self-supervised learning and it all seems a little magical at the moment.
The only way I can get an intuitive understanding is to assume that, for real-world problems, features are still ...
- 61
13
votes
3
answers
14k
views
What is the relation between semi-supervised and self-supervised visual representation learning?
What's the differences between semi-supervised learning and self-supervised visual representation learning, and how they are connected?
- 281
97
votes
3
answers
86k
views
What is self-supervised learning in machine learning?
What is self-supervised learning in machine learning? How is it different from supervised learning?
- 41.1k
0
votes
1
answer
1k
views
What is the purpose of the GAN?
The Generative Adversarial Network (GAN) is composed of a generator $G$ and a discriminator $D$. How do these two components interact? What is the intuition behind the GAN, its purpose, and how it is ...
13
votes
2
answers
2k
views
How are generative adversarial networks trained?
I am reading about generative adversarial networks (GANs) and I have some doubts regarding it. So far, I understand that in a GAN there are two different types of neural networks: one is generative ($...
- 1,076