Learnable Distribution Calibration for Few-Shot Class-Incremental Learning

TL;DR

This paper introduces a learnable distribution calibration method for few-shot class-incremental learning that effectively addresses distribution estimation and memorization challenges, improving performance without prior class similarity assumptions.

Contribution

The study proposes a unified, memory-efficient framework with a parameterized calibration unit that dynamically calibrates class distributions during training and incremental learning.

Findings

LDC outperforms state-of-the-art methods on multiple datasets.

LDC effectively mitigates forgetting and overfitting in FSCIL.

The approach is validated on few-shot learning scenarios.

Abstract

Few-shot class-incremental learning (FSCIL) faces challenges of memorizing old class distributions and estimating new class distributions given few training samples. In this study, we propose a learnable distribution calibration (LDC) approach, with the aim to systematically solve these two challenges using a unified framework. LDC is built upon a parameterized calibration unit (PCU), which initializes biased distributions for all classes based on classifier vectors (memory-free) and a single covariance matrix. The covariance matrix is shared by all classes, so that the memory costs are fixed. During base training, PCU is endowed with the ability to calibrate biased distributions by recurrently updating sampled features under the supervision of real distributions. During incremental learning, PCU recovers distributions for old classes to avoid `forgetting', as well as estimating distributions and augmenting samples for new classes to alleviate `over-fitting' caused by the biased distributions of few-shot samples. LDC is theoretically plausible by formatting a variational inference procedure. It improves FSCIL's flexibility as the training procedure requires no class similarity priori. Experiments on CUB200, CIFAR100, and mini-ImageNet datasets show that LDC outperforms the state-of-the-arts by 4.64%, 1.98%, and 3.97%, respectively. LDC's effectiveness is also validated on few-shot learning scenarios.