Metraipolis › archive › dc_ca7cb37e1b4b

JudgeMoE: Distributional Aggregation for LLM-as-a-Judge — arXiv:2610.07109

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.07109
License (as archived)
CC0-1.0
Archived
2026-10-07T11:23:00.573Z
SHA-256 of the archived text
76c2f313a16b89defea2e7d94eba2e55b2cae9c5ffa57ec4f7f3475dccbcea40
Stable URL of this page
https://metraipolis.com/documento/dc_ca7cb37e1b4b

How to cite

JudgeMoE: Distributional Aggregation for LLM-as-a-Judge — arXiv:2610.07109 — arXiv (en), https://arxiv.org/abs/2610.07109 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ca7cb37e1b4b (Metraipolis archive dc_ca7cb37e1b4b, sha256 76c2f313a16b89defea2e7d94eba2e55b2cae9c5ffa57ec4f7f3475dccbcea40)
BibTeX
@misc{dcca7cb37e1b4b,
  title = {JudgeMoE: Distributional Aggregation for LLM-as-a-Judge — arXiv:2610.07109},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.07109},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ca7cb37e1b4b; sha256 76c2f313a16b89defea2e7d94eba2e55b2cae9c5ffa57ec4f7f3475dccbcea40},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.07109. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

When an LLM judge scores an output, its score distribution retains uncertainty and disagreement information that is lost after scalar compression. We introduce JudgeMoE, a lightweight aggregator that assigns example-specific weights to cached judge score distributions and fuses them before computing a final score. A protocol study shows that score-range choice is unstable across judge--dataset settings and that soft scoring usually outperforms hard decoding. On the original 10-cell benchmark, JudgeMoE improves mean Spearman over uniform log pooling by $+0.079$. Applying the same configuration to six additional cells yields a $+0.0393$ mean gain over the strongest local single judge across 16 cells, with positive differences in 12/16 cells and a one-sided Wilcoxon signed-rank $p=0.0091$. Validation-based analyses further show that the preferred aggregation method depends on the task and judge pool.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.07109 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T11:15:39.411Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).