Research map / Continual learning / Domain and prompt-based
Decomposing the Neurons: Activation Sparsity via Mixture of Experts for Continual Test Time Adaptation
Rongyu Zhang, Aosong Cheng, Yulin Luo, Gaole Dai, Huanrui Yang, Jiaming Liu, Ran Xu, Li Du, Dan Wang, Yuan Du
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add plain 4-expert MoE (ablation) | CoTTA baseline · Cityscapes-to-ACDC | mean mIoU: 58.6 -> 57.9 | Hurt |
| add Spatial Differentiable Dropout (SDD) to experts (ablation) | MoE + CoTTA · Cityscapes-to-ACDC | mean mIoU: 57.9 -> 61.1 | Improved |
| add Domain-Aware Router (DAR) (ablation) | MoE + SDD · Cityscapes-to-ACDC | mean mIoU: 61.1 -> 61.5 | Improved |
| add Activation Sparsity Gate (ASG) (ablation) | MoE + SDD + DAR · Cityscapes-to-ACDC | mean mIoU: 61.5 -> 62.0 | Improved |
| add Homeostatic-Proximal (HP) loss (ablation) | MoE + SDD + DAR + ASG · Cityscapes-to-ACDC | mean mIoU: 62.0 -> 62.2 | Improved |
| use 4 experts instead of 2/8/16 (ablation) | MoASE · Cityscapes-to-ACDC round 1 | mean mIoU: E=2 61.5, E=4 61.8, E=8 60.5, E=16 61.2 | Mixed |
| MoASE vs ViDA on ImageNet-C | ViDA · ImageNet-to-ImageNet-C | mean error (%): 43.4 -> 40.5 | Improved |
| MoASE vs ViDA on CIFAR10-C | ViDA · CIFAR10-to-CIFAR10-C | mean error (%): 20.7 -> 16.8 | Improved |
| MoASE vs ViDA on Cityscapes-to-ACDC | ViDA · Cityscapes-to-ACDC (Segformer-B5) | mean mIoU: 61.9 -> 62.2 | Improved |
| apply MoASE to Segformer-B0 and SAM-SETR | ViDA · Cityscapes-to-ACDC | mean mIoU: 47.6 -> 48.2 (B0); 65.6 -> 66.0 (SAM-SETR) | Improved |
| MoASE on domain generalization to 5 unseen ImageNet-C domains | ViDA · ImageNet-C DG | mean error (%): 42.3 -> 37.5 | Improved |
| repeat target domain sequence for 3 rounds | MoASE · Cityscapes-to-ACDC | mean mIoU: 61.8 -> 62.3 -> 62.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViDA — ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (ICLR 2024) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion (NeurIPS 2024 · similarity 0.89)
- SURGEON: Memory-Adaptive Fully Test-Time Adaptation via Dynamic Activation Sparsity (CVPR 2025 · similarity 0.88)
- CoreMatching: A Co-adaptive Sparse Inference Framework with Token and Neuron Pruning for Comprehensive Acceleration of Vision-Language Models (ICML 2025 · similarity 0.87)
- Exploring the Benefit of Activation Sparsity in Pre-training (ICML 2024 · similarity 0.86)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.86)
- PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
