iCaRL: Incremental Classifier and Representation Learning
By Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, G. Sperl, Christoph H. Lampert. 2016
4,930 citations recorded by the source corpus.
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.
Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Meta & Continual Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Dacheng Tao
- Sergey Levine
- Mohit Bansal
- Yoshua Bengio
- Bernt Schiele
- Trevor Darrell
- Jianfeng Gao
- Bing Liu
In Representation & Self-Supervised Learning
- Han-Jia Ye
- Dacheng Tao
- Trevor Darrell
- De-chuan Zhan
- Andrea Vedaldi
- Tinne Tuytelaars
- Bernt Schiele
- Joost van de Weijer
In Meta & Continual Learning
Citations in the corpus
- Re-evaluating Continual Learning Scenarios: A Categorization and Case for Strong Baselines (2018, methodology): “This management is similar to iCaRL [11], except that we randomly pick the images for staying in the buffer.”
- AutoML: A Survey of the State-of-the-Art (2019, methodology): “In addition, iCaRL [168] is a new training strategy based on”
- A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “iCaRL [356] and EEIL [53] are two early works that perform”
- DisCoRL: Continual Reinforcement Learning via Policy Distillation (2019, methodology): “[Rebuffi et al., 2017] Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. (2017”
- Dynamic Few-Shot Visual Learning Without Forgetting (2018, methodology): “The proposed modification in the architecture of a classification ConvNet allows to unify the recognition of base and novel categories without significantly altering the classification pipeline for the recognition of base categories (in contrast to [11, 17]).”
- Continuous Learning in Single-Incremental-Task Scenarios (2018, methodology): “Permuted MNIST [12][20] Rotated MNIST [20] MNIST Split [31] [24] CIFAR-10/100 Split [31] iCIFAR-100 [20] [27] ILSVRC2012 [27] CUB-200 [24] CORe50 [17] In [20], the authors decided to further split the iCIFAR-100 benchmark in 20 different “tasks” containing 5 different classes each.”
- Overcoming catastrophic forgetting with hard attention to the task (2018, background): “The use of memory modules to alleviate catastrophic forgetting has been a subject of research until today (Rebuffi et al., 2017; Lopez-Paz & Ranzato, 2017).”
- FearNet: Brain-Inspired Model for Incremental Learning (2017, background): “Gepp-Net+STM is better at retaining base-knowledge since it only trains during its consolidation phase, but the STM-free version learns new data better because it is updates the model on every novel labeled input. iCaRL (Rebuffi et al., 2017) is an incremental class learning framework.”
- Deep Generative Dual Memory Network for Continual Learning (2017, methodology): “in iCaRL (Rebuffi et al., 2017) or indirectly used to modify future gradient updates to the system e.”
- Gradient Episodic Memory for Continual Learning (2017, methodology): “Most of the literature about learning over a sequence of tasks [Rusu et al., 2016, Fernando et al., 2017, Kirkpatrick et al., 2017, Rebuffi et al., 2017] describes a setting where i) the number of tasks is small, ii) the number of examples per task is large, iii) the learner performs several passes…”
- Closed-Loop GAN for continual Learning (2018, methodology): “Other works have built around using a buffer of real data to approximate past distributions [Rebuffi et al., 2017, Lopez-Paz et al., 2017].”
- Memory Efficient Experience Replay for Streaming Learning (2018, methodology): “Streaming classification is distinct from incremental batch learning, which has recently received much attention [41], [42], [49], [54], [65], [74], and it has unique challenges.”
- Progress & Compress: A scalable framework for continual learning (2018, methodology): “Examples making use of this idea are (Rebuffi et al., 2016; Schmidhuber, 2013; Thrun, 1996).”
- Incorporating neuro-inspired adaptability for continual learning in artificial intelligence (2023, background): “Representative strategies include selectively stabilizing parameters [7–11], recovering old data distributions [12–16] and allocating dedicated parameter subspaces [17, 18], etc.”
- S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning (2023, methodology): “LwF [23] proposed to use knowledge distillation [15] to alleviate catastrophic forgetting. iCaRL [31] showed that nearest classifier mean (NCM) using old class exemplars can generate robust classifiers for CIL.”
- Learnability and Algorithm for Continual Learning (2023, methodology): “For the replay-based methods, we follow the budget sizes of (Rebuffi et al., 2017; Buzzega et al., 2020).”