Questions tagged [catastrophic-forgetting]
For questions related to the concept of catastrophic forgetting (or, also called, catastrophic interference), which is the problem of forgetting previously acquired information (or the ability to solve certain tasks) while learning new information (in an online or incremental fashion) that certain machine learning models (in particular, neural networks) face.
12 questions
3
votes
1
answer
4k
views
Should I be layer freezing when fine-tuning an LLM?
I've had it in my head that generally speaking, it's better to freeze layers when fine-tuning an LLM, as per this quote from HuggingFace's article:
PEFT approaches only fine-tune a small number of (...
1
vote
1
answer
135
views
Do other online/incremental algorithms not suffer from catastrophic forgetting?
All the literature I read seems to indicate catastrophic forgetting affects only neural networks. Do other online/incremental algorithms not suffer from catastrophic forgetting (for example, ...
- 13
1
vote
0
answers
281
views
Gradual decrease in performance of a DDPG agent
I'm trying to solve the OpenAI's CarRacing-v0 environment with the DDPG algorithm. I've observed that after a period of learning, the agent's performance starts to deteriorate slowly. For some ...
- 11
1
vote
0
answers
201
views
If REINFORCE agent suddenly drops, how do I verify if it's due to catastrophic forgetting?
I am using the default implementations of REINFORCE, DQN and c51 available from the tf.agents repo (links). As you can see, DQN manages to improve performance while REINFORCE seems to suffer from ...
- 185
5
votes
1
answer
874
views
Why do DQNs tend to forget?
Why do DQNs tend to forget? Is it because when you feed highly correlated samples, your model (function approximation) doesn't give a general solution?
For example:
I use level 1 experiences, my ...
- 369
3
votes
1
answer
1k
views
How is transfer learning used to mitigate catastrophic forgetting in neural networks?
How can transfer learning be used to mitigate catastrophic forgetting. Could someone elaborate on this?
- 709
6
votes
3
answers
1k
views
What are the state-of-the-art approaches for continual learning with neural networks?
There seems to be a lot of literature and research on the problems of stochastic gradient descent and catastrophic forgetting, but I can't find much on solutions to perform continual learning with ...
- 366
7
votes
2
answers
2k
views
Was the corruption of Microsoft's "Tay" chatbot an example of catastrophic forgetting?
Tay was a chatbot, who learned from Twitter users.
Microsoft's AI fam from the internet that's got zero chill. The more you talk the smarter Tay gets. — Twitter tagline.
Microsoft trained the AI ...
- 225
4
votes
1
answer
2k
views
How can I incrementally train a Yolo model without catastrophic forgetting?
I have successfully trained a Yolo model to recognize k classes. Now I want to train by adding k+1 class to the pre-trained weights (k classes) without forgetting previous k classes. Ideally, I want ...
- 83
66
votes
4
answers
17k
views
Are neural networks prone to catastrophic forgetting?
Imagine you show a neural network a picture of a lion 100 times and label it with "dangerous", so it learns that lions are dangerous.
Now imagine that previously you have shown it millions ...
- 2,246
4
votes
1
answer
127
views
Could new training pictures destroy the trained weights of the neural network?
Let's say an image has 28*28 pixels, which leads to 784 input nodes in a feed-forward neural network. If an image can be classified into 1 of 10 numbers (e.g. MNIST), there are 10 output nodes.
We ...
- 41
31
votes
5
answers
14k
views
Is it possible to train a neural network as new classes are given?
I would like to train a neural network (NN) where the output classes are not (all) defined from the start. More and more classes will be introduced later based on incoming data. This means that, every ...
- 457