PASs-MoE: Mitigating Misaligned Co-drift among Router and Experts via Pathway Activation Subspaces for Continual Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing LoRA-based Mixture-of-Experts (MoE) methods often jointly update the router and experts in an indiscriminate way, causing the router’s preferences to co-drift with experts’ adaptation pathways and exacerbate forgetting. |
| Approach: | They propose a LoRA-induced subspace that reflects which low-rank pathway directions an input activates in each expert, providing a capability-aligned coordinate system for routing and preservation. |
| Outcome: | The proposed method outperforms conventional continual learning baselines and MoE–LoRA variants in accuracy and resistance to forgetting, without increasing model parameters. |
Similar Papers
Dynamic Expert Specialization: Towards Catastrophic Forgetting-Free Multi-Domain MoE Adaptation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to adapt Mixture-of-Experts models to multiple domains are prohibitive computation, cross-domain interference or require separate runs per domain. |
| Approach: | They propose a dynamic expert specialization framework for multi-domain adaptation of Mixture-of-Experts models. |
| Outcome: | The proposed framework reduces forgetting by 89% compared to full fine-tuning as domains scale from 2 to 6 and achieves faster convergence than conventional methods. |
From Pseudo-Balancing to True Specialization: Memory-Aware Routing for Mixture-of-Experts (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to optimize expert-centered load balancing fail to account for pseudo-balance phenomenon . severe knowledge overlap among experts leads to redundant representations and inefficient parameter utilization . |
| Approach: | They propose a method that prioritizes expert utilization over semantic alignment . they use memory-aware routing to ensure expert load balancing is consistent . |
| Outcome: | Experimental results show that MAR improves expert specialization by 35% and accuracy by 2%-25% . MAR matches baseline performance with only half the experts . |
From Outcome to Process: Optimizing MoE Load Balancing with MCTS (2026.findings-acl)
Copied to clipboard
Wenjun Ke, Hengyuan Xu, Ziyu Shang, Yao He, Jiahao Wang, Zijie Xu, Peng Wang, Yuhang Lou, Jiajun Liu
| Challenge: | Existing balancing strategies focus on constraining the final distribution of expert usage, but overlook the routing decisions made at each layer. |
| Approach: | They propose a three-stage framework that leverages process-level rewards to guide balanced expert routing. |
| Outcome: | Extensive experiments show that LayerMoE improves the performance of state-of-the-art LoRA-MoA baselines, yielding an average accuracy gain of 1.39%. |
SLIM: Let LLM Learn More and Forget Less with Soft LoRA and Identity Mixture (2025.naacl-long)
Copied to clipboard
| Challenge: | balancing the training budget, downstream performance, and general capabilities of large language models remains a challenge in many applications. |
| Approach: | They propose a mixture of expert framework based on Soft LoRA and Identity Mixture . SLIM allows dynamic routing between LoRA adapters and identity layers . |
| Outcome: | The proposed framework reduces training cost while maintaining general capabilities . it can be open-sourced upon publication. |
MiLoRA: Efficient Mixture of Low-Rank Adaptation for Large Language Models Fine-tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-rank adaptation and its mixture-of-experts (MOE) methods are highly effective but introduce significant latency in multi-tenant settings due to the LoRA modules and MOE routers added to multiple linear modules. |
| Approach: | They propose a low-rank adaptation variant that considers each LoRA module as an expert and employs a prompt-aware routing mechanism. |
| Outcome: | Extensive analysis on commonsense reasoning tasks and math reasoning tasks show that MiLoRA outperforms strong PEFT baselines with comparable tunable parameter budgets. |
SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for multitask learning fail to match input semantics with expert capabilities, leading to weak expert specialization. |
| Approach: | They propose a parameter-efficient mixture-of-experts framework for task-adaptive learning that aligns textual semantics with the most suitable experts for precise routing. |
| Outcome: | The proposed framework outperforms the state-of-the-art methods and holds excellent task generalization capabilities. |
Faster MoE LLM Inference for Extremely Large Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing inference optimizations for coarse-grained Mixture-of-Experts models implicitly assume a fixed activation budget, which is poorly understood. |
| Approach: | They propose a training-free policy that adapts token-level activation using router confidence and entropy while remaining within the model’s original budget. |
| Outcome: | The proposed skipping policy can provide substantial throughput gains, but optimal static schedules vary significantly across models and routing mechanisms. |
Awakening Dormant Experts:Counterfactual Routing to Mitigate MoE Hallucinations (2026.acl-long)
Copied to clipboard
Wentao Hu, Yanbo Zhai, Xiaohui Hu, Mingkuan Zhao, Shanhong yu, Xue Liu, Kaidong Yu, Shuangyong Song, Xuelong Li
| Challenge: | Sparse Mixture-of-Experts models are vulnerable to hallucinations, authors say . static Top-k routing leaves "specialist experts" under-prioritized for specific tokens . |
| Approach: | They propose a training-free inference framework to awaken dormant experts . they propose 'counterfactual routing' to shift computational resources from syntax-dominant to knowledge-intensive layers . |
| Outcome: | Experiments show that CoR improves factual accuracy by 3.1% without increasing the inference budget. |
Advancing SMoE for Continuous Domain Adaptation of MLLMs: Adaptive Router and Domain-Specific Loss (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies have explored Continual Instruction Tuning (CIT) in Multimodal Large Language Models (MLLMs), with a primary focus on Task-incremental CIT, where MLLM are required to continuously acquire new tasks. |
| Approach: | They propose a Sparse Mixture of Expert (SMoE) based method for domain-incremental CIT in Multimodal Large Language Models (MLLMs) . they equip the SMoA module with a domain-specific autoregressive loss (DSAL) they establish a new benchmark to evaluate the efficacy of their method . |
| Outcome: | The proposed method outperforms all baselines and is based on a Sparse Mixture of Experts (SMoE) module . |
LoRAMoE: Alleviating World Knowledge Forgetting in Large Language Models via MoE-Style Plugin (2024.acl-long)
Copied to clipboard
Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Experimental results show that, as the instruction data increases, LoRAMoE can significantly improve the ability to process downstream tasks, while maintaining the world knowledge stored in the LLM. |
| Approach: | They propose a framework that introduces several low-rank adapters and integrates them by using a router network to freeze the backbone model and force a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks. |
| Outcome: | The proposed framework freezes the backbone model and forces a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks, to alleviate world knowledge forgetting. |