Abstract:Few-shot class-incremental learning is crucial for developing scalable and adaptive intelligent systems, as it enables models to acquire new classes with minimal annotated data while safeguarding the previously accumulated knowledge. Nonetheless, existing methods deal with continuous data streams in a centralized manner, limiting their applicability in scenarios that prioritize data privacy and security. To this end, this paper introduces federated few-shot class-incremental learning, a decentralized machine learning paradigm tailored to progressively learn new classes from scarce data distributed across multiple clients. In this learning paradigm, clients locally update their models with new classes while preserving data privacy, and then transmit the model updates to a central server where they are aggregated globally. However, this paradigm faces several issues, such as difficulties in few-shot learning, catastrophic forgetting, and data heterogeneity. To address these challenges, we present a synthetic data-driven framework that leverages replay buffer data to maintain existing knowledge and facilitate the acquisition of new knowledge. Within this framework, a noise-aware generative replay module is developed to fine-tune local models with a balance of new and replay data, while generating synthetic data of new classes to further expand the replay buffer for future tasks. Furthermore, a class-specific weighted aggregation strategy is designed to tackle data heterogeneity by adaptively aggregating class-specific parameters based on local models performance on synthetic data. This enables effective global model optimization without direct access to client data. Comprehensive experiments across three widely-used datasets underscore the effectiveness and preeminence of the introduced framework.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the problem of few - shot class - incremental learning (FSCIL) in a distributed data environment. Specifically, the paper focuses on how to enable the model to continuously learn new classes on multiple clients with limited labeled data while retaining the knowledge of previously learned classes. ### Background and challenges 1. **Few - shot learning**: The amount of data on each client is very limited, which makes it difficult for the model to effectively learn new classes. 2. **Catastrophic forgetting**: As the model is continuously updated to incorporate new classes, it may forget the previously learned classes, resulting in a performance decline. 3. **Data heterogeneity**: The data distributions of different clients may be non - independent and identically distributed (Non - IID), which further exacerbates the difficulty of learning. ### Solutions To solve the above problems, the paper proposes a federated few - shot class - incremental learning (F2SCIL) framework and introduces the following key techniques: 1. **Synthetic data - driven framework (SDD)**: - **Noise - aware generative replay module (NAGR)**: Generate synthetic data to retain the knowledge of old classes and use a noise - robust loss function to mitigate the impact of noisy labels in the synthetic data on the model's learning. - **Conditional generator**: Train a conditional generator based on the integration of local models to generate new class data, which is added to the replay buffer for future incremental learning tasks. 2. **Class - specific weighted aggregation strategy (CSWA)**: - To deal with data heterogeneity, a class - specific weighted aggregation strategy is designed. This strategy adaptively aggregates class - specific parameters according to the performance of local models on synthetic data, thereby optimizing the global model. ### Main contributions 1. **Propose the F2SCIL paradigm**: In a distributed data environment, enable the global model to continuously learn new classes from the limited data of multiple clients. 2. **Introduce the SDD framework**: Use synthetic data to retain old knowledge and improve the model integration on the server side. 3. **Develop the NAGR module**: Mitigate catastrophic forgetting through a noise - robust loss function without relying on real data. 4. **Design the CSWA strategy**: Effectively deal with data heterogeneity and improve the performance of the global model. ### Experimental verification The paper conducted comprehensive experiments on three widely - used datasets to verify the effectiveness and superiority of the proposed framework. ### Summary By proposing the F2SCIL framework and its key techniques, this paper successfully solves multiple challenges in few - shot class - incremental learning in a distributed data environment and provides an effective solution for continuous learning in practical applications.

Few-Shot Class-Incremental Learning with Non-IID Decentralized Data

Few-Shot Class-Incremental Learning via Entropy-Regularized Data-Free Replay

Few-shot Class Incremental Learning Via Prompt Transfer and Knowledge Distillation

Few-Shot Incremental Learning for Label-to-Image Translation

Rethinking Few-Shot Class-Incremental Learning: A Lazy Learning Baseline

Flexible Few-Shot Class-Incremental Learning with Prototype Container

Few-Shot Incremental Learning with Continually Evolved Classifiers

A survey on few-shot class-incremental learning

Class-Incremental Few-Shot Object Detection

On Distilling the Displacement Knowledge for Few-Shot Class-Incremental Learning

Uncertainty-Guided Semi-Supervised Few-Shot Class-Incremental Learning With Knowledge Distillation

Class similarity weighted knowledge distillation for few shot incremental learning

Few-Shot Class-Incremental Learning Via Class-Aware Bilateral Distillation

Few Shot Class Incremental Learning via Grassmann Manifold and Information Entropy

A Cognition-Driven Framework for Few-Shot Class-Incremental Learning

Few-Shot Federated Learning: A Federated Learning Model for Small-Sample Scenarios

Pseudo Initialization Based Few-Shot Class Incremental Learning

Overcoming Spatial-Temporal Catastrophic Forgetting for Federated Class-Incremental Learning

Secure Collaborative Few-Shot Learning

RRA-FFSCIL: Inter-intra classes representation and relationship augmentation federated few-shot incremental learning

Deep Class Incremental Learning from Decentralized Data