Dynamic LoRA-Experts and Prototype-Ensemble Matching for Class-Incremental Learning
Not on the map. Why?
Abstract (arXiv)
Class-Incremental Learning (CIL) aims to continuously learn new classes without forgetting previously acquired knowledge. Parameter-efficient fine-tuning with pre-trained models reduces parameter overhead but can suffer from cumulative interference and suboptimal alignment between inference samples and specialized modules. We propose Dynamic LoRA-Experts and Prototype-Ensemble Matching (DLEPEM), a two-stage rehearsal-free framework. DLEPEM allocates a task-specific LoRA-Expert for each incremental task to reduce cross-task interference, then combines frozen pre-trained-model prototypes with task-adaptive LoRA-Expert prototypes for reliable task-level discrimination. Experiments on standard CIL and Few-Shot CIL benchmarks demonstrate strong performance under the evaluated protocols.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 30 of its 40 references that are in the map. First posted on arXiv 30 Sep 2026.
Builds on
30 of 40 references.
References · 30
- 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
- 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
- 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,893
- 2015Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classificationcited by 3,590
- 2016iCaRL: Incremental Classifier and Representation Learningcited by 2,313
- 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,842
- 2021Learning to Prompt for Continual Learningcited by 938
- 2021Scaling Vision with Sparse Mixture of Expertscited by 740
- 2022DualPrompt: Complementary Prompting for Rehearsal-free Continual Learningcited by 635
- 2022CODA-Prompt: COntinual Decomposed Attention-based Prompting for Rehearsal-Free Continual Learningcited by 446
- 2019A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmarkcited by 419
- 2021Pre-Trained Models: Past, Present and Futurecited by 319
18 more
- 2021When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentationscited by 259
- 2020Few-Shot Class-Incremental Learningcited by 249
- 2022S-Prompts Learning with Pre-trained Transformers: An Occam's Razor for Domain Incremental Learningcited by 246
- 2020Semantic Drift Compensation for Class-Incremental Learningcited by 241
- 2022FOSTER: Feature Boosting and Compression for Class-Incremental Learningcited by 228
- 2018Task-Free Continual Learningcited by 219
- 2024Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapterscited by 199
- 2024InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learningcited by 189
- 2023Revisiting Class-Incremental Learning with Pre-Trained Models: Generalizability and Adaptivity are All You Needcited by 183
- 2023Hierarchical Decomposition of Prompt-Based Continual Learning: Rethinking Obscured Sub-optimalitycited by 121
- 2024Expandable Subspace Ensemble for Pre-Trained Model-Based Class-Incremental Learningcited by 115
- 2023Mixture of Cluster-conditional LoRA Experts for Vision-language Instruction Tuningcited by 106
- 2025SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learningcited by 60
- 2023On the Stability-Plasticity Dilemma of Class-Incremental Learningcited by 55
- 2024MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Expertscited by 55
- 2022Benchmarking Omni-Vision Representation through the Lens of Visual Realmscited by 38
- 2024Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learnerscited by 35
- 2024Few-Shot Class Incremental Learning with Attention-Aware Self-Adaptive Promptcited by 15
Built on by
Not cited yet.