iCaRL: Incremental Classifier and Representation Learning

A major open problem on the road to artificial intelligence is the development of incrementally learning systems that learn about more and more concepts over time from a stream of data. In this work, we introduce a new training strategy, iCaRL, that allows learning in such a class-incremental way: only the training data for a small number of classes has to be present at the same time and new classes can be added progressively. iCaRL learns strong classifiers and a data representation simultaneously. This distinguishes it from earlier works that were fundamentally limited to fixed data representations and therefore incompatible with deep learning architectures. We show by experiments on CIFAR-100 and ImageNet ILSVRC 2012 data that iCaRL can learn many classes incrementally over a long period of time where other strategies quickly fail.

PDF Abstract CVPR 2017 PDF CVPR 2017 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
class-incremental learning cifar100 iCaRL 10-stage average accuracy 63.24 # 4
Incremental Learning ImageNet100 - 10 steps iCaRL Average Incremental Accuracy Top-5 83.60 # 8
Final Accuracy Top-5 63.80 # 8
# M Params 11.22 # 2
Incremental Learning ImageNet-100 - 50 classes + 5 steps of 10 classes iCaRL* Average Incremental Accuracy 65.56 # 5
Incremental Learning ImageNet - 10 steps iCaRL Average Incremental Accuracy 38.40 # 7
Final Accuracy 22.70 # 5
Average Incremental Accuracy Top-5 63.70 # 7
Final Accuracy Top-5 44.00 # 7
# M Params 11.68 # 2

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Incremental Learning CIFAR-100 - 50 classes + 10 steps of 5 classes iCaRL* Average Incremental Accuracy 52.57 # 10
Incremental Learning CIFAR-100 - 50 classes + 5 steps of 10 classes iCaRL* Average Incremental Accuracy 57.17 # 9

Methods


No methods listed for this paper. Add relevant methods here