Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, Christoph H. Lampert
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.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Continual Learning | cifar100 | 10-stage average accuracy | 63.24 | iCaRL |
| Incremental Learning | CIFAR-100 - 50 classes + 10 steps of 5 classes | Average Incremental Accuracy | 52.57 | iCaRL* |
| Incremental Learning | CIFAR-100 - 50 classes + 5 steps of 10 classes | Average Incremental Accuracy | 57.17 | iCaRL* |
| Incremental Learning | CIFAR-100-B0(5steps of 20 classes) | Average Incremental Accuracy | 71.14 | iCaRL |
| Incremental Learning | ImageNet - 10 steps | # M Params | 11.68 | iCaRL |
| Incremental Learning | ImageNet - 10 steps | Average Incremental Accuracy | 38.4 | iCaRL |
| Incremental Learning | ImageNet - 10 steps | Average Incremental Accuracy Top-5 | 63.7 | iCaRL |
| Incremental Learning | ImageNet - 10 steps | Final Accuracy | 22.7 | iCaRL |
| Incremental Learning | ImageNet - 10 steps | Final Accuracy Top-5 | 44 | iCaRL |
| Incremental Learning | ImageNet100 - 10 steps | # M Params | 11.22 | iCaRL |
| Incremental Learning | ImageNet100 - 10 steps | Average Incremental Accuracy Top-5 | 83.6 | iCaRL |
| Incremental Learning | ImageNet100 - 10 steps | Final Accuracy Top-5 | 63.8 | iCaRL |
| Incremental Learning | CIFAR-100 - 50 classes + 2 steps of 25 classes | Average Incremental Accuracy | 71.33 | iCaRL |
| Incremental Learning | ImageNet-100 - 50 classes + 5 steps of 10 classes | Average Incremental Accuracy | 65.56 | iCaRL* |
| Class Incremental Learning | cifar100 | 10-stage average accuracy | 63.24 | iCaRL |