Forward Compatible Few-Shot Class-Incremental Learning

Zhou, Dawei; Wang, Fuyun; Ye, Han-Jia; Li, Ma; Pu, Shiliang; Zhan, De-Chuan

doi:10.48550/arxiv.2203.06953

Cited by 4 publications

(11 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare our proposed method with 6 other methods: iCaRL [30], TOPIC [38], IDLVQ-C [4], F2M [34], CEC [45], C-FSCIL [14] and FACT [49]. Overall, our proposed M-FSCIL outperforms all state-of-the-art methods on these three benchmark datasets.Because our model's backbone is pre-trained by various image-text pairs, it performs better on the object classification dataset, such as miniImageNet.…”

Section: Experimental Results Analysismentioning

confidence: 92%

Memory-Based Label-Text Tuning for Few-Shot Class-Incremental Learning

Li¹,

Yan²,

Lou³

et al. 2022

Preprint

View full text Add to dashboard Cite

Few-shot class-incremental learning(FSCIL) focuses on designing learning algorithms that can continually learn a sequence of new tasks from a few samples without forgetting old ones. The difficulties are that training on a sequence of limited data from new tasks leads to severe overfitting issues and causes the wellknown catastrophic forgetting problem. Existing researches mainly utilize the image information, such as storing the image knowledge of previous tasks or limiting classifiers updating. However, they ignore analyzing the informative and less noisy text information of class labels. In this work, we propose leveraging the label-text information by adopting the memory prompt. The memory prompt can learn new data sequentially, and meanwhile store the previous knowledge. Furthermore, to optimize the memory prompt without undermining the stored knowledge, we propose a stimulation-based training strategy. It optimizes the memory prompt depending on the image embedding stimulation, which is the distribution of the image embedding elements. Experiments show that our proposed method outperforms all prior state-of-the-art approaches, significantly mitigating the catastrophic forgetting and overfitting problems.Preprint. Under review.

show abstract

Section: Experimental Results Analysismentioning

confidence: 92%

Memory-Based Label-Text Tuning for Few-Shot Class-Incremental Learning

Li¹,

Yan²,

Lou³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Few-shot class-incremental learning is recently proposed to tackle few-shot inputs in the class-incremental setting [68], [69].…”

Section: Few-shot Class-incremental Learningmentioning

confidence: 99%

Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks

Zhou¹,

Ye²,

Zhan³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

New classes arise frequently in our ever-changing world, e.g., emerging topics in social media and new types of products in e-commerce. A model should recognize new classes and meanwhile maintain discriminability over old classes. Under severe circumstances, only limited novel instances are available to incrementally update the model. The task of recognizing few-shot new classes without forgetting old classes is called few-shot class-incremental learning (FSCIL). In this work, we propose a new paradigm for FSCIL based on meta-learning by LearnIng Multi-phase Incremental Tasks (LIMIT), which synthesizes fake FSCIL tasks from the base dataset. The data format of fake tasks is consistent with the 'real' incremental tasks, and we can build a generalizable feature space for the unseen tasks through meta-learning. Besides, LIMIT also constructs a calibration module based on transformer, which calibrates the old class classifiers and new class prototypes into the same scale and fills in the semantic gap. The calibration module also adaptively contextualizes the instance-specific embedding with a set-to-set function. LIMIT efficiently adapts to new classes and meanwhile resists forgetting over old classes. Experiments on three benchmark datasets (CIFAR100, miniImageNet, and CUB200) and large-scale dataset, i.e., ImageNet ILSVRC2012 validate that LIMIT achieves state-of-the-art performance.

show abstract

“…The former group seeks to rehearse former knowledge when learning new, and the latter group saves extra model components to assist incremental learning. 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. Exemplar-Based Methods: Exemplars are representative instances from former classes [51], and CIL models can selectively save a relatively small amount of exemplars for rehearsal during updating [22].…”

Section: Related Workmentioning

confidence: 99%

A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning

Zhou¹,

Wang²,

Ye³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Real-world applications require the classification model to adapt to new classes without forgetting old ones. Correspondingly, Class-Incremental Learning (CIL) aims to train a model with limited memory size to meet this requirement. Typical CIL methods tend to save representative exemplars from former classes to resist forgetting, while recent works find that storing models from history can substantially boost the performance. However, the stored models are not counted into the memory budget, which implicitly results in unfair comparisons. We find that when counting the model size into the total budget and comparing methods with aligned memory size, saving models do not consistently work, especially for the case with limited memory budgets. As a result, we need to holistically evaluate different CIL methods at different memory scales and simultaneously consider accuracy and memory size for measurement. On the other hand, we dive deeply into the construction of the memory buffer for memory efficiency. By analyzing the effect of different layers in the network, we find that shallow and deep layers have different characteristics in CIL. Motivated by this, we propose a simple yet effective baseline, denoted as MEMO for Memory-efficient Expandable MOdel. MEMO extends specialized layers based on the shared generalized representations, efficiently extracting diverse representations with modest cost and maintaining representative exemplars. Extensive experiments on benchmark datasets validate MEMO's competitive performance.

show abstract

Forward Compatible Few-Shot Class-Incremental Learning

Cited by 4 publications

References 41 publications

Memory-Based Label-Text Tuning for Few-Shot Class-Incremental Learning

Memory-Based Label-Text Tuning for Few-Shot Class-Incremental Learning

Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks

A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning

Contact Info

Product

Resources

About