Scholar Feed

Mixture-of-experts routing

MoLE

MoLE : Mixture of Language Experts for Multi-Lingual Automatic Speech Recognition

Superseded baseline#31 of 1,370 most-superseded · first seen Feb 27, 2023

Superseded — cited as a baseline and beaten by newer methods

4 papers critique it · 1 beat it on benchmarks

What papers say

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

Mixture-of-Lookup-Experts (MoLE) jie2025mole radically replaces MLPs with parameter-free lookup tables, sacrificing expressive power for efficiency.
— Breaking the MoE LLM Trilemma: Dynamic Expert Clustering with Structured Compression
“However, the absence of nonlinear activation on the expert outputs may limit the expressive capacity of the resulting representations.”
— Scaling Machine Learning Interatomic Potentials with Mixtures of Experts
“Compared to existing state-of-the-art MoE baselines (Switch Transformer, MoLE, HydraLoRA), HiLoMoE consistently shows superior efficiency and effectiveness.”
— Hierarchical LoRA MoE for Efficient CTR Model Scaling
“because the routing is global, the resulting potential applies the same effective weights to every atom in the simulation domain”
— Mixture of Experts Framework in Machine Learning Interatomic Potentials for Atomistic Simulations

Beaten on benchmarks

Head-to-head results where a newer method reports beating MoLE. 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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.