Scholar Feed

Mixture-of-experts routing

LEMoE

LEMoE: Advanced Mixture of Experts Adaptor for Lifelong Model Editing of Large Language Models

Superseded baseline#36 of 1,370 most-superseded · first seen Jun 28, 2024

Superseded — cited as a baseline and beaten by newer methods

3 papers critique it · 1 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites LEMoE as a baseline.

However, LEMoE functions as a parameter-preserving framework that attaches external modules to a frozen backbone (typically dense). It addresses routing consistency within the added adaptor rather than the routing distribution shift of the base model itself.
— MoEEdit: Efficient and Routing-Stable Knowledge Editing for Mixture-of-Experts LLMs
“Additionally, since the routers in these methods are fine-tuned, even if previous experts are frozen, there can still be modifications to past knowledge.”
— LLaVA-CMoE: Towards Continual Mixture of Experts for Large Vision-Language Models
“LEMoE is based on MoE, but its greedy routing harms old experts' influence when integrating new ones.”
— Lifelong Knowledge Editing for Vision Language Models with Low-Rank Mixture-of-Experts

Beaten on benchmarks

Head-to-head results where a newer method reports beating LEMoE. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.