Forward Compatible Few-Shot Class-Incremental Learning

By Da-Wei Zhou, Fu-Yun Wang, Han-Jia Ye, Liang Ma, Shiliang Pu, De-chuan Zhan. 2022

288 citations recorded by the source corpus.

Novel classes frequently arise in our dynamically changing world, e.g., new users in the authentication system, and a machine learning model should recognize new classes without forgetting old ones. This scenario becomes more challenging when new class instances are insufficient, which is called few-shot class-incremental learning (FSCIL). Cur-rent methods handle incremental learning retrospectively by making the updated model similar to the old one. By contrast, we suggest learning prospectively to prepare for future updates, and propose ForwArd Compatible Training (FACT) for FSCIL. Forward compatibility requires future new classes to be easily incorporated into the current model based on the current stage data, and we seek to realize it by reserving embedding space for future new classes. In detail, we assign virtual prototypes to squeeze the embedding of known classes and reserve for new ones. Besides, we forecast possible new classes and prepare for the updating process. The virtual prototypes allow the model to accept possible updates in the future, which act as proxies scattered among embedding space to build a stronger classifier during inference. Fact efficiently incorporates new classes with forward compatibility and meanwhile resists for-getting of old ones. Extensive experiments validate FACT's state-of-the-art performance. Code is available at: https://github.com/zhoudw-zdw/CVPR22-Fact

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, 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

In Computer Vision and Pattern Recognition

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “To promote learning plasticity on the basis of memory stability, emergent strategies include renormalization of the old and new task solutions [163], [255], [373], [429], balanced exploitation of the old and new training samples [11], [164], [171], [440], [457], [500], reserving space for subsequent”
First Session Adaptation: A Strong Replay-Free Baseline for Class-Incremental Learning (2023, methodology): “Asterisk (*) indicates that the reported results are from [45] and † indicates results reported in [21].”
Towards Label-Efficient Incremental Learning: A Survey (2023, background): “Forward-Compatible Training [ Zhou et al. , 2022 ] is the current state-of-the-art in few-shot incremental learning.”
Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, methodology): “Few-shot class-incremental learning (FSCIL) can be roughly divided into finetune-based methods [16, 26, 4, 31] and metric-based methods [40, 44].”
FOSTER: Feature Boosting and Compression for Class-Incremental Learning (2022, background): “Therefore, it is necessary to enable the model to continuously learn new categories, namely class-incremental learning [34,44,49].”
Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks (2022, background): “Few-shot class-incremental learning is recently proposed to tackle few-shot inputs in the class-incremental setting [66], [67].”
Steering Prototypes with Prompt-tuning for Rehearsal-free Continual Learning (2023, background): “Prevailing alrothims in continual learning adopt class mean embeddings as prototypes [58,61,62].”
Identifying Ambiguous Similarity Conditions via Semantic Matching (2022, background): “representations) from data benefits machine learning and visual recognition systems [1,6,11,25,27,39,42,43,49].”
LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning (2023, background): “[74] Da-Wei Zhou, Fu-Yun Wang, Han-Jia Ye, Liang Ma, Shiliang Pu, and De-Chuan Zhan.”
Class-Incremental Learning: A Survey (2023, methodology): “…[107], [108], [109] Logit Rectify [83], [89], [110], [111] Weight Rectify [112], [113], [114] § 3.7: Template-Based Classification [37], [82], [104], [115], [116], [117], [118], [119], [120], [121], [122], [123] Note that the classification rules of these categories are based on the ‘key…”
Compositional Few-Shot Class-Incremental Learning (2024, background): “While various approaches, including metric-based ones (Zou et al., 2022; Zhang et al., 2021), adaptation-based ones (Zhou et al., 2022; Yibo Yang, 2023), etc. have been explored for this task, FSCIL remains a challenge due to the scarcity of training data and the risk of catastrophic forgetting.”
Learning Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification (2024, methodology): “Existing CT methods can be categorized into two branches: backward CT [16, 24, 39] and forward CT methods [21, 30, 45].”
Neural Collapse Inspired Feature-Classifier Alignment for Few-Shot Class Incremental Learning (2023, background): “A recent study proposes to reserve feature space for novel classes to circumvent their conﬂict with old classes (Zhou et al., 2022a), but an optimal feature-classiﬁer alignment is hard to be guaranteed with learnable classiﬁer (Pernici et al., 2021).”
A Survey on Few-Shot Class-Incremental Learning (2023, methodology): “In order to enable the model to handle new classes, Zhou et al. (2022a) proposed forward compatible training (FACT), which allocates multiple virtual prototypes as a reserved space in the feature space to make the model scalable.”
Boundary-aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval (2023, background): “Following this topic, compatible learning methods have developed into three branches: backward-compatible learning, forwardcompatible learning [15, 34], and cross-model compatible learning (CMC) [1,14,19,23].”
Deep Class-Incremental Learning: A Survey (2023, methodology): “However, FACT [106] addresses the forward compatibility in CIL, where the model should be prepared for future updates, acting like the pre-assigned interface.”
Pre-trained Vision and Language Transformers are Few-Shot Incremental Learners (2024, methodology): “Feature and feature space-based methods [2, 12, 23, 30, 41, 42] enable the model to adapt to new classes better and improve the generalization ability of feature extractors for new classes.”
A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning (2022, background): “There are other methods that do not fall into these two groups [24, 29, 23, 46, 61, 66], and we refer the readers to [10, 36, 62] for a holistic review.”
Warping the Space: Weight Space Rotation for Class-Incremental Few-Shot Learning (2023, methodology): “Additionally, we further compare our method with more recent works (Zhou et al., 2022b; Chi et al., 2022; Hersche et al., 2022; Liu et al., 2022) on mini ImageNet dataset.”