{
  "annotation_id": "071f6be6-edcc-43ae-8ed6-26f33c765cba",
  "date_created": "2026-02-17T05:53:20.468000Z",
  "date_modified": "2026-02-17T05:53:20.468000Z",
  "file_hash": "5b7374da5baf95ece504e2dfdcc2f9140e70be570688a245db23489537b174c2",
  "private": false,
  "record": {
    "abstract": "Horseshoe mixtures-of-experts (HS-MoE) models provide a Bayesian framework for sparse expert selection in mixture-of-experts architectures. We combine the horseshoe prior\u0027s adaptive global-local shrinkage with input-dependent gating, yielding data-adaptive sparsity in expert usage. Our primary methodological contribution is a particle learning algorithm for sequential inference, in which the filter is propagated forward in time while tracking only sufficient statistics. We also discuss how HS-MoE relates to modern mixture-of-experts layers in large language models, which are deployed under extreme sparsity constraints (e.g., activating a small number of experts per token out of a large pool).",
    "arxiv_id": "2601.09043",
    "authors": [
      "Nick Polson",
      "Vadim Sokolov"
    ],
    "categories": [
      "stat.ML",
      "cs.LG"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "title": "Horseshoe Mixtures-of-Experts (HS-MoE)",
    "url": "https://arxiv.org/abs/2601.09043",
    "version": "v1"
  },
  "schema_id": "dorsal/arxiv",
  "source": {
    "execution_id": "73e57463-aaa2-45db-a655-487de8124dae",
    "id": "arXiv Dataset",
    "type": "Model",
    "variant": "snapshot-2026-01-17",
    "version": "0.1.0"
  },
  "user_id": 1000002
}