ICML 2026 — Program Notes (pre-meeting prep)¶
Internal working file. Today is May 11, 2026 — the meeting is ~2 months out. Main-paper submissions closed Jan 28, 2026; workshop final-notification deadline is May 15, 2026 (this week); accepted-paper lists post on OpenReview around the same window through camera-ready in June. Limited accepted-paper data exists yet. This file holds the scaffolding so we can move fast once the lists drop.
Verified facts¶
| Field | Value | Source |
|---|---|---|
| Conference | Forty-Third International Conference on Machine Learning | icml.cc |
| City | Seoul, South Korea | icml.cc/Conferences/2026 |
| Venue | COEX Convention & Exhibition Center | icml.cc/Conferences/2026 |
| Expo / Tutorial Day | Jul 6, 2026 | icml.cc |
| Main conference | Jul 7–9, 2026 | icml.cc |
| Workshops | Jul 10–11, 2026 | icml.cc |
| Submission site opened | Jan 8, 2026 — past | CallForPapers |
| Abstract submission deadline | Jan 23, 2026 (AoE) — past | CallForPapers |
| Full paper submission deadline | Jan 28, 2026 (AoE) — past | CallForPapers |
| Workshop application deadline | Feb 13, 2026 (AoE) — past | CallForWorkshops |
| Workshop acceptance notifications | Mar 20, 2026 — past | CallForWorkshops |
| Workshop contributions deadline | Apr 24, 2026 (AoE) — past | CallForWorkshops |
| Workshop final notification | May 15, 2026 (AoE) — this week | CallForWorkshops |
| Early registration deadline | May 24, 2026 (AoE) | icml.cc |
| Registration cancellation deadline | Jun 15, 2026 (8:59 PM KST) | icml.cc |
| Paper format | 8 pages main + unlimited references / impact / appendix; +1 page for camera-ready | CallForPapers |
| License | PMLR open-access (proceedings) + OpenReview hosting | PMLR / OpenReview |
Sources we will rely on¶
| Source | When usable | Coverage |
|---|---|---|
| ICML 2026 home | now | dates, venue, organizers, registration |
| Call for Papers | now | submission rules, paper-format spec |
| Call for Workshops | now | workshop timeline |
| OpenReview ICML group | from late May 2026 | accepted main-track papers + per-workshop venues |
| PMLR proceedings | from ~Jul 2026 | camera-ready, citeable |
| SlidesLive (ICML) | from ~Jul 15 onward | recorded orals + workshop talks |
| bioRxiv / arXiv | continuous | preprints, filterable by "to appear at ICML 2026" / "accepted at ICML 2026" |
#ICML2026 on Bluesky / Mastodon / X |
Jul 6–11 | real-time talk reactions |
| Lab blogs (DeepMind, Anthropic, Insitro, Recursion, Genentech prescient design, Isomorphic Labs, Profluent, Cradle, Latent Labs) | rolling | preprints + summary posts of ICML papers from these labs |
Filter criteria (locked — mirrors tools/index.md)¶
A paper enters this vault only if it matches one of three filters. Reproduced here for the pre-meeting screening pass.
Filter 1 — workshop track¶
Accept anything in these workshops (canonical short names; 2026 roster finalizes May 15):
- FM4LS / ML4LMS — Multi-modal Foundation Models and LLMs for Life Sciences
- GenBio — Generative AI for Biology (3rd edition anticipated)
- AI4Science — AI for Science (bio / chem subset only)
- FM4Health — Foundation Models for Health (anticipated entrant; may surface under a different rebrand)
- LCFM — Long-Context Foundation Models (genomic-LM contributions only)
Filter 2 — main-track keyword¶
Run this regex over titles + abstracts once the main-track accept list publishes:
(cancer|oncolog|tumor|metasta|carcinom|melanoma|leukemi|lymphoma)
| (genomic|epigenom|transcriptom|proteom|metabolom|methyl)
| (single[- ]cell|scRNA|scATAC|spatial transcriptom|spatial omic)
| (perturbation|CRISPR|knockout|knockdown|sgRNA)
| (drug discover|target ident|molecular property|ADMET|small molecul|biologic|de novo design|generative chemistry)
| (protein structur|protein function|protein language|equivariant|RFdiffusion|AlphaFold|Boltz|ESM)
| (foundation model.*(biolog|cell|genom|tissue|pathol|clinical|medical))
| (electronic health record|EHR|clinical NLP|medical imaging|pathology|histolog|radiolog|diagnos)
| (virtual cell|cell atlas|cellular state)
Filter 3 — agentic-AI / LLM-for-science subset¶
Independent of bio framing, accept any main-track or workshop paper on:
- LLM agents with tool-use, code-execution, or experiment-design loops
- Scientific-reasoning benchmarks (SWE-bench-class, sci-agent-class, lab-task benchmarks)
- Autonomous research / hypothesis-generation systems
- Multi-agent scientific workflows
Rationale: these adapt to oncology within ~6 months (the empirical pattern from 2024 → AACR 2025 and 2025 → AACR 2026) and feed the AACR agentic-AI axis directly.
Expected paper themes (5 anchors)¶
Pre-meeting predictions, to be reconciled against the actual list in late May / June. These are the buckets we expect tools/ entries to cluster into.
Theme 1 — Foundation models for biology, second generation¶
scGPT, Geneformer, and ESM are 2022–2024 vintage. ICML 2026 should feature the next-gen models: cross-modal (DNA + RNA + protein + tissue + clinical), multi-million-cell training corpora, instruction-tuned for downstream tasks (perturbation prediction, virtual-screen rerank), and benchmarked on virtual-cell tasks rather than embedding-quality proxies. Workshop home: FM4LS / ML4LMS, with main-track spillover. Expect heavy overlap with the NeurIPS 2026 LMRL roster six months later.
Theme 2 — Agentic AI for science¶
Tool-use LLMs that drive bioinformatics pipelines, design wet-lab experiments, or autonomously navigate clinical-decision trees. 2025 surfaced ChemCrow / Coscientist-class systems; 2026 should bring oncology-specific agents (tumor-board agents, trial-matching agents, multi-omics interpretation agents). Workshop home: main-track + FM4Health (if it runs); also AI4Science.
Theme 3 — Single-cell + spatial ML methods¶
Cell-state representation learning, perturbation-response prediction (Replogle-class, sci-Plex-class benchmarks), spatial-transcriptomics imputation, cell-cell communication inference, virtual-tissue generation. Workshop home: FM4LS / ML4LMS; main-track for the methods-heavy contributions. The 2025 FM4LS roster anchored this theme strongly; 2026 should extend it.
Theme 4 — Generative biology: de novo protein, antibody, and molecule design¶
De novo small-molecule generation conditioned on oncology targets, antibody / TCR / peptide design, equivariant diffusion models for protein–ligand co-design, RFdiffusion descendants, Boltz / AlphaFold3-class structure models, flow-matching applied to molecular design. Workshop home: GenBio (3rd edition anticipated); main-track for the headline releases.
Theme 5 — Long-context genomic and clinical language models¶
Evo / NT-class genomic LMs scaling to whole-chromosome / whole-genome context windows; clinical-trajectory LMs over multi-year EHR sequences; long-context FMs for whole-slide pathology / whole-trajectory radiology. Workshop home: LCFM (genomic-LM subset) + FM4LS / FM4Health spillover; main-track for the architecture papers.
Pre-meeting timeline (action items)¶
| Date | Action |
|---|---|
| Now → May 15, 2026 | passive monitoring; await workshop final-notification list |
| May 15, 2026 | workshop roster finalized — confirm which bio/health workshops actually run, identify any new entrants |
| Late May – Jun 2026 | main-track accepted-paper list publishes via OpenReview — run keyword regex, build candidate list |
| Jun 2026 | per-workshop accepted-paper lists drop — extend candidate list; cross-check author overlap with NeurIPS 2026 expected roster |
| Late Jun 2026 | scaffold likely tools/ pages from titles + bioRxiv mirrors |
| Jul 6–11, 2026 | meeting week — opportunistic real-time capture from social, especially agentic-AI and GenBio orals |
| Mid-Jul 2026 → Aug 2026 | bulk tools/ population from SlidesLive recordings + camera-ready / PMLR papers |
Caveats¶
- Partial accepted-paper data. Workshop accept lists and the main-track list publish on a rolling basis through May–Jun. Past-year patterns anchor the planning; the keyword filter is the safety net.
- 2026 workshop roster could shift. FM4LS appeared in 2025; the committee favors novel workshops over long-running iterations, so the same theme may rebrand. The keyword filter and theme anchors mitigate this.
- ICML / NeurIPS author overlap. Many tools and authors appear at both venues within a calendar year. Tool-entry consolidation policy is in
tools/index.mdopen question 3. - Agentic-AI scope creep risk. Filter 3 is broader than Filters 1–2 by design, but it can swell if applied loosely. Keep the "science / tool-use benchmark required" caveat enforced.
- Seoul-venue logistics. No bearing on the vault, but the meeting-week digest (if we run one) needs to account for KST time-zone offsets when capturing social reactions.