Add papers and researchers from GitHub paper issues - 2026-11-04
Adds 11 abstract-confirmed papers from the open paper issues.
- Data selection: curation diagnostics, agentic dataset search, group-level synthetic curation, model-aware selection, speech embeddings, OCRia data curation, FACTS, Domino, and gradient-based training-data attribution.
- Training strategy: unified foundation-model training and DISCOVER online prompt discovery/replay.
- Source issues: #91, #90, #89, #87, #84, #82, #77, #76, #69, #68, #43. Issue #89 contains the wrong arXiv link; resolved its title to https://arxiv.org/abs/2610.00779.
- #44 remains open: idSCD studies training-dataset membership inference; placement within the current lifecycle taxonomy remains unresolved.
Evidence: canonical arXiv abstracts and author lists; OCRia official technical-report abstract; DISCOVER author-hosted full text and NeurIPS 2026 program. No identifier duplicates found. Notes describe methods without generalizing benchmark-specific numerical claims.
Author follow-up adds 10 researchers and 20 supporting papers. Qualification requires at least three distinct abstract-confirmed relevant papers across topics, including the seed. All ten are new researcher identities in this snapshot; none replaces an existing entry.
| Researcher | Verified relevant papers (seed + new) | Topics |
|---|---|---|
| Cathy Jiao | 3 (1 + 2) | data_selection |
| Chenyan Xiong | 5 (1 + 4) | data_selection |
| Jiaxing Wang | 3 (1 + 2) | data_selection |
| Judy Hoffman | 3 (1 + 2) | data_selection, training_strategy |
| Viraj Prabhu | 3 (1 + 2) | data_selection, training_strategy |
| James Zou | 3 (1 + 2) | data_selection, training_strategy |
| Christopher Ré | 3 (1 + 2) | data_selection, labeling |
| Yuda Song | 3 (1 + 2) | training_strategy |
| David Grangier | 5 (1 + 4) | data_selection, training_strategy |
| Benjamin Roth | 3 (1 + 2) | data_selection, training_strategy |
Affiliations verified from current personal/institutional pages on 2026-11-04. In particular, Judy Hoffman is now at UC Irvine, with an adjunct Georgia Tech appointment. Author identity is linked through publication pages listing the qualifying works; full author lists and abstracts were retrieved for all supporting papers. Embedding-based seed papers and CLUE are placed under the embedding subsections.
Coverage: all 12 original paper issues were enumerated; eleven were imported and closed after PR creation, while #44 remains open pending relevance/placement. The 11 imported papers have 73 distinct authors. This follow-up used the default 30-author probe ceiling, prioritizing first/last seed authors and selected coauthors. Ten researchers qualified; twenty other probes remain unresolved or did not establish three relevant papers in the checked evidence. This is not a claim that those authors lack relevant work. Another 43 seed authors and the recursive author queues remain deferred at the author ceiling. The limit can be raised or removed to resume. Depth limit: two; default ceilings: 30 author checks, 200 candidate abstracts, 300 requests. Abstract and request ceilings were not reached.
Examples retained for follow-up include Loris Schoenegger (not reached within the author budget), the other BLADE/TANDEM coauthors, and unresolved Qixia/Qixiang Jiang identity linkage. Below-threshold probes were not imported as bibliography additions.
Validation: git diff --check, balanced HTML blocks, canonical-ID deduplication, researcher name/count checks, and researcher-to-bibliography link consistency. Caller checkout untouched. Worktree: .worktree/add-new-paper-2026-11-04, based on freshly fetched origin/main (08ba697). Local evidence and queues: /tmp/pr92-researcher-additions.json and /tmp/author-paper-evidence.json.