neurai-lab / cls-er Goto Github PK
View Code? Open in Web Editor NEWThe official PyTorch code for ICLR'22 Paper "Learning Fast, Learning Slow: A General Continual Learning Method based on Complementary Learning System""
License: MIT License
The official PyTorch code for ICLR'22 Paper "Learning Fast, Learning Slow: A General Continual Learning Method based on Complementary Learning System""
License: MIT License
Hi! Thanks for the awesome work! I am trying to reproduce your results but I have some questions about the details. Thanks for your answers in advance.
I found that, within each task, you utilize the model of the last epoch rather than the best one. Are there any special considerations?
I reproduced your work on a single RTX 3090 GPU, with the default seed 1993, but I am not sure whether I chose the right value as the final accuracy. I chose the accuracy of the last task, e.g., in seq-mnist experiment, the value of task 5's accuracy, which is also the final accuracy shown in the terminal. Am I right?
Could you please let us know the exact seeds you were using? So that we might be able to get closer results.
We found that, during the training of most continual learning approaches, your code will have an accuracy first before starting training, as the following image. I guess it is pre-train, but I did not find the pre-trained setup. Could you please indicate how these values are obtained?
Thank you very much again!
thanks for sharing!
But when I was debugging the model "icarl", I found that the function forward in class ICarL is not used at all.
I wonder whether it is normal or there's sth wrong with it.
Hello, your paper has greatly inspired me, and I have some questions about it. In general class-incremental learning scenarios, reservoir sampling is performed after training each batch. So, during the training of the first task, the data within the task will also be replayed during the training process?
Hello, I try to reproduce your work in scope of 'ML Reproducibility Challenge 2022' but I faced with few problem about code base.
- There is a problem about compressed SEQ-TINYIMG file, unzip tools returns 'tiny-imagenet-processed.zip not an archive' error. I tried different OS but unable to export files in zip file.
- For gcil-cifar100 dataset given --weight_dist parameter is not recognized by main.py so it is not reproducible
Could you help me about this issues ?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.