Home

Awesome

iCarl2.0

This is an on-going pytorch implementation of iCarl[1].

Interface to run experiments

runExperiment.py [-h] [--batch-size N] [--test-batch-size N]
                        [--epochs N] [--lr LR]
                        [--schedule SCHEDULE [SCHEDULE ...]]
                        [--gammas GAMMAS [GAMMAS ...]] [--momentum M]
                        [--no-cuda] [--no-distill] [--no-random]
                        [--no-herding] [--oversampling] [--seed S]
                        [--log-interval N] [--model-type MODEL_TYPE]
                        [--name NAME] [--sortby SORTBY] [--decay DECAY]
                        [--step-size STEP_SIZE]
                        [--memory-budget MEMORY_BUDGET]
                        [--epochs-class EPOCHS_CLASS] [--classes CLASSES]
                        [--depth DEPTH] [--dataset DATASET]

Dependencies

  1. Pytorch 0.3.0.post4
  2. Python 3.6
  3. torchnet (https://github.com/pytorch/tnt)

Ideas to implement and test

  1. Auto-encoder to preserve information of a class
  2. Generative models for data seen in the past to remove dependency on examplers. =======

Results

alt text

References

[1] https://arxiv.org/abs/1611.07725