Method Drift›Mixture-of-experts routing
Tracked
DisagMoE
DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe ParallelismMixture-of-experts routing · first seen May 10, 2026
current frontier — recent, not yet superseded in the knowledge base
0 papers critique it · 0 beat it on benchmarks
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- ConceptM$^3$oEConceptM$^3$oE: Concept-Guided Multimodal Mixture of Experts for Interpretable Computational PathologyMay 23, 2026
- DisagMoEDisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe ParallelismMay 10, 2026
- PiperPiper: Efficient Large-Scale MoE Training via Resource Modeling and Pipelined Hybrid ParallelismMay 6, 2026
- GRACE-MoEGRACE-MoE: Grouping and Replication with Locality-Aware Routing for Efficient Distributed MoE InferenceMay 6, 2026
- Apr 21, 2026
- Feb 12, 2026
- Multi-Head LatentMoE and Head Parallel (HP)Multi-Head LatentMoE and Head Parallel: Communication-Efficient and Deterministic MoE ParallelismFeb 4, 2026
- Jan 29, 2026
- Rasterized Steered Mixture of ExpertsRasterized Steered Mixture of Experts for Efficient 2D Image RegressionOct 7, 2025
- Sep 30, 2025
- Sep 24, 2025