Giter Club home page Giter Club logo

birt's Introduction

BiRT

This is the official repository of the ICML, 2023 paper "BiRT: Bio-inspired Replay in Vision Transformers for Continual Learning" by Kishaan Jeeveswaran, Prashant Bhat, Bahram Zonooz and Elahe Arani.

TLDR: A novel representation rehearsal-based continual learning approach that, by incorporating constructive noises at various stages of the vision transformer and regularization, enables effective and memory-efficient Continual Learning.

Schematic of MTSL:

image info

Setup:

OUTPUT_DIR: Directory to save output contents.
DATA_DIR: Directory containing the datasets.

Datasets supported:

  • CIFAR-100
  • ImageNet-100
  • Tiny ImageNet

BiRT Training Script:

To train BiRT on CIFAR-100 dataset 10 tasks setting with buffer size 500:

python main.py --seed 42 --options options/data/cifar100_10-10.yaml options/data/cifar100_order1.yaml options/model/cifar_birt.yaml --data-path <DATA_DIR> --output-basedir <OUTPUT_DIR> --base-epochs 500 --batch_mixup --batch_logitnoise --ema_alpha 0.001 --ema_frequency 0.003 --distill_version l2 --distill_weight 0.05 --distill_weight_buffer 0.001 --rep_noise_weight 1.0 --repnoise_prob 0.5 --finetune_weight 2 --representation_replay --replay_from 1 --sep_memory --num_workers 8 --csv_filename results.csv --memory-size 500 --tensorboard --epochs 500

Hyperparameters for other settings:

Dataset Num of Tasks Buffer Size ema_alpha ema_frequency distill_weight distill_weight_buffer
CIFAR-100 5 200 0.0005 0.001 0.05 0.01
500 0.005 0.003 0.05 0.01
10 200 0.001 0.003 0.05 0.001
500 0.001 0.003 0.05 0.001
1000 0.0005 0.0008 0.05 0.01
2000 0.0002 0.0015 0.05 0.01
20 200 0.005 0.001 0.05 0.08
500 0.0005 0.003 0.05 0.1
TINYIMAGENET 10 500 0.001 0.003 0.05 0.01
1000 0.01 0.0008 0.01 0.001
2000 0.0001 0.008 0.01 0.0008
IMAGENET- 100 10 500 0.0001 0.003 0.05 0.001
1000 0.0001 0.003 0.05 0.001
2000 0.01 0.005 0.01 0.001

Cite Our Work:

If you find the code useful in your research please consider citing our paper:

@article{jeeveswaran2023birt,
  title={BiRT: Bio-inspired Replay in Vision Transformers for Continual Learning},
  author={Jeeveswaran, Kishaan and Bhat, Prashant and Zonooz, Bahram and Arani, Elahe},
  journal={arXiv preprint arXiv:2305.04769},
  year={2023}
}

birt's People

Contributors

elahearani avatar kishaan avatar

Stargazers

 avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar

Watchers

 avatar  avatar

Forkers

mudit004

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.