ml-research / bowll

BOWLL: A Deceptively Simple Open World Lifelong Learner

This repository contains the code for the axXiv paper BOWLL: A deceptively simple Open World Lifelong Learner.

Abstract: The quest to improve scalar performance numbers on predetermined benchmarks seems to be deeply engraved in deep learning. However, the real world is seldom carefully curated and applications are seldom limited to excelling on test sets. A practical system is generally required to recognize novel concepts, refrain from actively including uninformative data, and retain previously acquired knowledge throughout its lifetime. Despite these key elements being rigorously researched individually, the study of their conjunction, open world lifelong learning, is only a recent trend. To accelerate this multifaceted field’s exploration, we introduce its first monolithic and much-needed baseline. Leveraging the ubiquitous use of batch normalization across deep neural networks, we propose a deceptively simple yet highly effective way to repurpose standard models for open world lifelong learning. Through extensive empirical evaluation, we highlight why our approach should serve as a future standard for models that are able to effectively maintain their knowledge, selectively focus on informative data, and accelerate future learning.

Credits

• We implemented DeepInversion using the official repository DeepInversion
• We adapted Active Query pool mechanism from DDU

Environmental Setup

You can install packages from requirements.txt after creating your own environment with python 3.7.x.

$ pip install --upgrade pip
$ pip install -r requirements.txt

Please download the DeepInversion package from the github repository and place it in the root directory. This generates pseudo-images for BOWLL.

Run Experiments

You can reproduce the experiments in the paper by running the following command:

python ../bowll_mnist.py \
		--training_batch_size 256 \
        --acquisition_batch_size 256 \
		--test_batch_size 256 \
		--ood_batch_size 4 \
	 	--n_domains 4 \
		--n_epochs 1 \
		--n_repeats 5 \
		--buffer_size 5000 \
        --arch alexnet \
        --path_to_weights ...


python ../bowll_cifar10.py \
		--training_batch_size 256 \
        --acquisition_batch_size 256 \
		--test_batch_size 256 \
		--ood_batch_size 8 \
	 	--n_timestpes 5 \
		--n_epochs 3 \
		--n_repeats 5 \
		--buffer_size 5000 \
        --arch resnet18 \
        --path_to_weights ... \