Is CLIP-MoE superseded?

CLIP-MoE (Mixture-of-experts routing): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 2 beat it on benchmarks — #41 of 1370 most-superseded. Sub-problem: cluster led by Upcycling. Newer alternatives in the same sub-problem include MetaMoE, BAR, BERT-MoE Framework, null experts within token-choice MoE, MixtureKit.

Mixture-of-experts routing

CLIP-MoE

CLIP-MoE: Towards Building Mixture of Experts for CLIP with Diversified Multiplet Upcycling

Superseded baseline#41 of 1,370 most-superseded · first seen Sep 28, 2024

Superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 2 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites CLIP-MoE as a baseline.

However, scaling this method is challenging, as adding more experts increases complexity and necessitates additional training stages.

— CLIP-UP: A Simple and Efficient Mixture-of-Experts CLIP Training Recipe with Sparse Upcycling

Beaten on benchmarks

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

Upcycling beats CLIP-MoE
39.6 vs 38.2
Avg. · [ViT-B/32]
Enhancing Mixture-of-Experts Specialization via Cluster-Aware Upcycling
CLIP-ERMoE beats CLIP-MoE
94.7 vs 92.0
Recall@10 · [COCO I2T Recall@10]
ERMoE: Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization

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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.