Scalable Analytic Classifiers with Associative Drift Compensation for Class-Incremental Learning of Vision Transformers

TL;DR

This paper introduces LR-RGDA, a scalable classifier for class-incremental learning with Vision Transformers, combining analytic Gaussian Discriminant Analysis with low-rank approximation and a Hopfield-based mechanism to handle drift efficiently.

Contribution

It proposes LR-RGDA, a low-rank factorized analytic classifier, and HopDC, a drift compensation method, enabling scalable and accurate class-incremental learning with ViTs.

Findings

Achieves state-of-the-art performance on CIL benchmarks.

Reduces inference complexity from quadratic to linear in class number.

Effectively mitigates representation drift with HopDC.

Abstract

Class-incremental learning (CIL) with Vision Transformers (ViTs) faces a major computational bottleneck during the classifier reconstruction phase, where most existing methods rely on costly iterative stochastic gradient descent (SGD). We observe that analytic Regularized Gaussian Discriminant Analysis (RGDA) provides a Bayes-optimal alternative with accuracy comparable to SGD-based classifiers; however, its quadratic inference complexity limits its use in large-scale CIL scenarios. To overcome this, we propose Low-Rank Factorized RGDA (LR-RGDA), a scalable classifier that combines RGDA's expressivity with the efficiency of linear classifiers. By exploiting the low-rank structure of the covariance via the Woodbury matrix identity, LR-RGDA decomposes the discriminant function into a global affine term refined by a low-rank quadratic perturbation, reducing the inference complexity from $\mathcal{O}(Cd^2)$ to $\mathcal{O}(d^2 + Crd^2)$, where $C$ is the class number, $d$ the feature dimension, and $r \ll d$ the subspace rank. To mitigate representation drift caused by backbone updates, we further introduce Hopfield-based Distribution Compensator (HopDC), a training-free mechanism that uses modern continuous Hopfield Networks to recalibrate historical class statistics through associative memory dynamics on unlabeled anchors, accompanied by a theoretical bound on the estimation error. Extensive experiments on diverse CIL benchmarks demonstrate that our framework achieves state-of-the-art performance, providing a scalable solution for large-scale class-incremental learning with ViTs. Code: https://github.com/raoxuan98-hash/lr_rgda_hopdc.