Few-Shot Incremental Learning with Continually Evolved Classifiers

By Chi Zhang, Nan Song, Guosheng Lin, Yun Zheng, Pan Pan, Yinghui Xu. 2021

428 citations recorded by the source corpus.

Few-shot class-incremental learning (FSCIL) aims to design machine learning algorithms that can continually learn new concepts from a few data points, without forgetting knowledge of old classes. The difficulty lies in that limited data from new classes not only lead to significant overfitting issues but also exacerbate the notorious catastrophic forgetting problems. Moreover, as training data come in sequence in FSCIL, the learned classifier can only provide discriminative information in individual sessions, while FSCIL requires all classes to be involved for evaluation. In this paper, we address the FSCIL problem from two aspects. First, we adopt a simple but effective decoupled learning strategy of representations and classifiers that only the classifiers are updated in each incremental session, which avoids knowledge forgetting in the representations. By doing so, we demonstrate that a pre-trained backbone plus a non-parametric class mean classifier can beat state-of-the-art methods. Second, to make the classifiers learned on individual sessions applicable to all classes, we propose a Continually Evolved Classifier (CEC) that employs a graph model to propagate context information between classifiers for adaptation. To enable the learning of CEC, we design a pseudo incremental learning paradigm that episodically constructs a pseudo incremental learning task to optimize the graph parameters by sampling data from the base dataset. Experiments on three popular benchmark datasets, including CIFAR100, miniImageNet, and Caltech-USCD Birds-200-2011 (CUB200), show that our method significantly outperforms the baselines and sets new state-of-the-art results with remarkable advantages.

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, background): “[490]), developing a tree-based hierarchical classifier with Gaussian processes (GP-Tree [6]), performing hyperdimen-”
S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning (2023, methodology): “For this, to mitigate catastrophic forgetting of the base classes, their class prototypes are stored as is the common practice [35][44].”
Towards Label-Efficient Incremental Learning: A Survey (2023, methodology): “We group the learners according to their machinery to learn base-novel class association, as in graph [ Tao et al. , 2020; Zhang et al. , 2021 ] , clustering [ Chen and Lee, 2020; Cheraghian et al. , 2021; Aky ¨ urek et al. , 2022; Zhou et al. , 2022 ] or the architecture [ Mazumder et al. , 2021;…”
Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, methodology): “Our implementation is based on the code released by CEC [44] under the MIT license.”
Few-Shot Class-Incremental Learning from an Open-Set Perspective (2022, positive): “Also, on both datasets, our ALICE method shows much higher harmonic accuracy on all sessions compared to the SOTA CEC [23] method.”
Incremental Few-Shot Semantic Segmentation via Embedding Adaptive-Update and Hyper-class Representation (2022, methodology): “Most current incremental methods [7, 11, 24, 49] use parameters or embedding vectors to represent category knowledge from a cognitive-inspired perspective, which addresses catastrophic forgetting by knowledge representation with new-class updating scheme.”
Few-Shot Class-Incremental Learning via Entropy-Regularized Data-Free Replay (2022, methodology): “We outperform the existing state-of-the-art method (CEC [31]) by 1.”
Learning to Imagine: Diversify Memory for Incremental Learning using Unlabeled Data (2022, methodology): “Popular structure-driven methods [2,20,21,25,32,33,35] modify the network structure or expand the network for training new tasks.”
Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks (2022, methodology): “of-the-art method is CEC [22], which decouples the training”
Constrained Few-shot Class-incremental Learning (2022, methodology): “Session (s) 1 2 3 4 5 6 7 8 9 No. of classes |C̃(s)| 60 65 70 75 80 85 90 95 100 Mode/ Work Feature extractor AL-MML [48] ResNet-18 64.10 55.88 47.07 45.16 40.11 36.38 33.96 31.55 29.37 Semantic KD∗ [6] ResNet-18 <64 <57 <51 <46 <43 <41 <39 <37 <35 VAE∗ [7] ResNet-18 <62 <58 <57 <52 <51 <49 <46 <45 ”
Doodle It Yourself: Class Incremental Learning by Drawing a Few Sketches (2022, methodology): “Very few methods are present to tackle the FSCIL problem like, pseudo incremental learning [70], knowledge distillation [9, 13], neural-gas network [54].”
Forward Compatible Few-Shot Class-Incremental Learning (2022, methodology): “We report the results of compared methods from [9] and [12].”
Representation Compensation Networks for Continual Semantic Segmentation (2022, background): “Some work propose to learn better representations by combining selfsupervised learning for feature extractor [9, 87] and solving class imbalance [39, 46, 54, 99, 102].”
Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (2021, background): “We notice that a similar observation is made in a newly released paper [59].”
Subspace Regularizers for Few-Shot Class Incremental Learning (2021, background): “Others (Gidaris & Komodakis, 2018; Yoon et al., 2020; Zhang et al., 2021) Our approach is related to Ren et al. (2019), who proposes a nested optimization framework to learn auxiliary parameters for every base and novel class to inﬂuence the novel weights via regularization; we show that these…”
Co-Transport for Class-Incremental Learning (2021, background): “Class-incremental learning (CIL) is now a popular topic in the machine learning community [6, 9, 15, 25, 39, 47, 62, 68, 70].”