Research map / Architectures and efficiency / Mixture of experts
ACO-MoE-LoRA: Evolving-while-Training for Adapting Segment Anything Model 2 to Specialized Domains
Kaiyi Luo, Bangjun Wang, Li Zhang, Fanzhang Li, Fei Zhu, Jiaqing Fan
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 encoder adaptation (LoRA) to frozen SAM2 with U-Net decoder (ablation) | Baseline · Polyp avg | mDice: 0.641 -> 0.858 | Improved |
| add latent MoE with random gating (ablation) | Encoder LoRA model · Polyp avg | mDice: 0.858 -> 0.853 | No clear effect |
| use gradient gate instead of random gate in latent MoE (ablation) | Latent MoE · Polyp avg | mDice: 0.853 -> 0.857 | Improved |
| use auxiliary-loss-free gate instead of random gate (ablation) | Latent MoE · Polyp avg | mDice: 0.853 -> 0.858 | Improved |
| add pheromone-guided gating (ablation) | Latent MoE · Polyp avg | mDice: 0.853 -> 0.865 | Improved |
| add ACO-driven elastic rank (ablation) | ACO-MoE-LoRA · Polyp avg | mDice: 0.865 -> 0.871 | Improved |
| use dynamic evolved rank instead of best fixed rank (r=16) (ablation) | ACO-MoE-LoRA decoder · Polyp avg | mDice: 0.865 -> 0.871 | Improved |
| use fixed rank r=4 instead of r=16 (ablation) | ACO-MoE-LoRA decoder · Polyp avg | mDice: 0.865 -> 0.852 | Hurt |
| use fixed rank r=32 instead of r=16 (ablation) | ACO-MoE-LoRA decoder · Polyp avg | mDice: 0.865 -> 0.859 | Hurt |
| use ACO instead of GA as meta-controller (ablation) | ACO-MoE-LoRA · Polyp | mDice: 0.862 -> 0.871 | Improved |
| use ACO instead of PSO as meta-controller (ablation) | ACO-MoE-LoRA · Polyp | mDice: 0.868 -> 0.871 | Improved |
| turn on pheromone at inference (ablation) | ACO-MoE-LoRA · Polyp test | mDice: 0.865 -> 0.865 | No clear effect |
Closest papers by meaning
- Model Swarms: Collaborative Search to Adapt LLM Experts via Swarm Intelligence (ICML 2025 · similarity 0.85)
- RefineEvo: Planning-Guided Heuristic Evolution with Bidirectional Experience (ICML 2026 · similarity 0.84)
- ConfigX: Modular Configuration for Evolutionary Algorithms via Multitask Reinforcement Learning (AAAI 2025 · similarity 0.83)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.82)
- HSEvo: Elevating Automatic Heuristic Design with Diversity-Driven Harmony Search and Genetic Algorithm Using LLMs (AAAI 2025 · similarity 0.82)
- SE-Agent: Self-Evolution Trajectory Optimization in Multi-Step Reasoning with LLM-Based Agents (NeurIPS 2025 · similarity 0.82)
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.
