Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Enhancing Pretrained Model-based Continual Representation Learning via Guided Random Projection

About

Recent paradigms in Random Projection Layer (RPL)-based continual representation learning have demonstrated superior performance when building upon a pre-trained model (PTM). These methods insert a randomly initialized RPL after a PTM to enhance feature representation in the initial stage. Subsequently, a linear classification head is used for analytic updates in the continual learning stage. However, under severe domain gaps between pre-trained representations and target domains, a randomly initialized RPL exhibits limited expressivity under large domain shifts. While largely scaling up the RPL dimension can improve expressivity, it also induces an ill-conditioned feature matrix, thereby destabilizing the recursive analytic updates of the linear head. To this end, we propose the Stochastic Continual Learner with MemoryGuard Supervisory Mechanism (SCL-MGSM). Unlike random initialization, MGSM constructs the projection layer via a principled, data-guided mechanism that progressively selects target-aligned random bases to adapt the PTM representation to downstream tasks. This facilitates the construction of a compact yet expressive RPL while improving the numerical stability of analytic updates. Extensive experiments on multiple exemplar-free Class Incremental Learning (CIL) benchmarks demonstrate that SCL-MGSM achieves superior performance compared to state-of-the-art methods.

Ruilin Li, Heming Zou, Xiufeng Yan, Zheming Liang, Jie Yang, Chenliang Li, Xue Yang• 2026

Related benchmarks

TaskDatasetResultRank
Class-incremental learningImageNet-R B0 Inc20
Last Accuracy77.12
79
Class-incremental learningImageNet-A (B0 Inc20)
Last Accuracy56.05
26
Class-incremental learningImageNet-R Inc5 (test)
Average Accuracy6.42
13
Class-incremental learningImageNet-R Inc-10 B-0 (test)
Favg5.31
7
Class-incremental learningImageNet-A Inc-5 B-0 (test)
Average Score9.5
7
Class-incremental learningImageNet-A Inc-10 B-0 (test)
Favg9.9
7
Class-incremental learningObjectNet Inc-5 B-0 (test)
Favg9.19
7
Class-incremental learningObjectNet Inc-10 B-0 (test)
Favg8.23
7
Class-incremental learningOmniBenchmark Inc-5 B-0 (test)
Favg8.01
7
Class-incremental learningOmniBenchmark Inc-10 B-0 (test)
Average F-Score7.6
7
Showing 10 of 12 rows

Other info

Follow for update