ICLR 2026¶

Fourteenth International Conference on Learning Representations — Riocentro Convention and Event Center, Rio de Janeiro, Brazil, April 23–27, 2026 (main conference Apr 23–25; workshops Apr 26–27).

Status: Scaffold — meeting concluded ~2 weeks ago (today is May 11, 2026). Accepted-paper list and workshop rosters are public on OpenReview; SlidesLive recordings and camera-ready PDFs are rolling out now. This is a light build, not full coverage — see "Scope" below. Bulk tools/ extraction starts this month.

Why this is in the vault¶

ICLR is the third of the "big three" general-purpose ML conferences (alongside NeurIPS and ICML) and is the canonical venue for representation learning — the methodological substrate underneath every foundation model that later gets adapted to cancer biology. Like NeurIPS and ICML, the overwhelming majority of its program is general-purpose ML with no direct oncology relevance, so we are not trying to cover it like ISMB or EuroBioC. ICLR lives in this vault for the same reason its siblings do: it is an upstream methods source for the AACR vault's agentic-AI and virtual-cells / foundation-model axes.

Three reasons to keep a scoped presence:

Representation learning is the bottleneck for virtual cells. scGPT, Geneformer, UCE, and the next-generation cross-modal cell models all are representation-learning contributions before they are biology contributions; ICLR is where their methodological cousins (contrastive objectives, masked modeling regimes, equivariant architectures, multimodal alignment) first land.
Bio-FM cross-pollination is heavier than at NeurIPS / ICML. ICLR 2025 hosted MLGenX and GEM; ICLR 2026 hosts four bio-themed workshops (LMRL, MLGenX 3rd ed., Gen², GEM) — a denser bio footprint than ICML's. Many authors and tools overlap directly with the NeurIPS LMRL / MLSB and ICML FM4LS / GenBio circuits.
Agentic AI for science. ICLR 2026 features explicit MLGenX framing of "AI Agents and the Biological Lab" plus a Foundation Models for Science workshop; the same ~6-month adapt-to-oncology pattern applies — an ICLR April result tends to surface as a TissueAgent / OncoAgent-class demo at SITC, ASH, or AACR within the same calendar year.

Scope (what this vault covers — and what it does not)¶

In scope (light build):

The recurring bio / health / drug-discovery / structural-biology workshops at ICLR 2026 (see roster below).
Main-track papers that match a cancer / genomics / single-cell / spatial / drug-discovery / clinical-NLP keyword filter (anticipate a few dozen out of ~3,700 main-track papers).
The agentic-AI / LLM-for-science subset of the main track, regardless of explicit bio framing.

Out of scope:

General-purpose ML methods papers (vision-only, NLP-only, RL-only, theory, optimization, fairness, robotics, etc.) with no plausible bio downstream — the >95% bulk.
The full ~3,700-paper main proceedings; bulk extraction is explicitly not planned.
Tutorials, expo events, and socials outside the bio / health niches.

Bio / health-relevant workshops (verified, 2026 roster)¶

ICLR 2026 accepted 40 workshops; four are directly bio-themed plus a fifth science-FM workshop with a substantial bio subset. Workshop dates are Apr 26–27, 2026 at Riocentro.

Workshop	Theme	Canonical short name	Edition
Learning Meaningful Representations of Life	foundation models for genomics, single-cell, spatial; virtual cells as the headline framing	LMRL	recurring (ICLR + NeurIPS)
Machine Learning for Genomics Explorations	"From Reasoning to Experimentation: Closing the Loop Between AI Agents and the Biological Lab" — target ID, perturbation reasoning, BRChallenge agent benchmark	MLGenX	3rd edition
Generative AI in Genomics: Barriers and Frontiers	directed evolution in protein science; engineering cellular and tissue states	Gen² / Gen2	1st edition
Integrating Generative and Experimental Platforms for Biomolecular Design	proteins, ligands, nucleic acids, cells; bridging computational and experimental biology	GEM	recurring
Foundation Models for Science: Real-World Impact and Science-First Design	problem-driven FMs for climate, materials, biology — bio subset only is in scope	FM4Science	2nd edition

This is a denser bio footprint than ICML 2026's (2–3 bio workshops) and comparable to NeurIPS 2026's expected 5–6 bio workshops — partly because OpenReview's transparent acceptance lifecycle gives ICLR a small advantage for bio-FM submissions that prefer public review.

OpenReview advantage — bulk extraction is easier here¶

ICLR's review process is fully public on OpenReview from submission through camera-ready: every accepted paper has a public review thread, author-reviewer discussion, and (for ~90% of papers) a permissive license on the PDF itself. This makes ICLR the easiest of the big-three to bulk-extract, compared to CVPR (camera-ready only) or NeurIPS (OpenReview-public but with later workshop posting cadence). Practical implications:

Accepted-paper list for ICLR 2026 was public from Jan 25, 2026 (notifications) onward; full camera-ready PDFs are public now.
Each workshop runs its own OpenReview venue with accepted papers posted ahead of the meeting (Mar–Apr 2026) — most are already public.
Author–reviewer review threads are scrapable for sentiment / contention signals.
bioRxiv mirrors exist for the majority of bio-themed contributions; the lag is typically 1–4 weeks behind OpenReview.

What we have to work with¶

Source	Coverage	Notes
Main proceedings	~3,700 accepted papers	OpenReview group `ICLR.cc/2026/Conference`; all camera-ready PDFs are public. CC-BY for ~90% of submissions.
Workshop pages	one OpenReview venue per workshop	All 40 workshops have their accept lists public; bio-themed five already finalized.
bioRxiv mirrors	most bio-relevant ICLR papers have a preprint	Filter bioRxiv for "ICLR 2026" mentions; lag of 1–4 weeks behind OpenReview.
SlidesLive + YouTube	recorded talks	ICLR records orals + most workshop talks; posts rolling, typically within 1–3 weeks of the meeting. Most are landing on SlidesLive now.
Twitter/Bluesky/Mastodon	post-meeting reactions	`#ICLR2026`; useful for catching which agentic-AI / FM-bio papers got traction.
Cross-link with NeurIPS / ICML	author + tool overlap	Many ICLR bio papers reappear at ICML (Jul) or NeurIPS (Dec) within the same year — single tool entry should consolidate or explicitly cross-link.

Organization¶

conferences/iclr-2026/
├── index.md             # this page
├── program-notes.md     # post-meeting prep — verified facts, filter criteria, expected themes
└── tools/               # per-paper / per-tool entries, populated from May 2026 onward
    └── index.md         # template + scope-filter note

No digest/, no themes.md, no keynotes/. The light-build pattern mirrors ICML 2026 and NeurIPS 2026: only tools/ gets populated, and only with the filtered subset.

What's different from full-conference vaults¶

Compared to ISMB 2026, RECOMB 2026, AACR 2026, ASCO 2026, or EuroBioC 2025 — which aim at near-complete program coverage — ICLR 2026 is scope-limited by design:

No per-session digests, no per-day digests, no themes synthesis. The general-purpose ML bulk is irrelevant.
No keynote / plenary coverage unless the speaker is a known bio-FM author.
Tool entries are filtered down to a few dozen, not exhaustive. The filter is keyed on the AACR axes, not on ICLR's own taxonomy.
Most ICLR 2026 tool entries should explicitly cross-link to an ICML 2026 (Jul) or NeurIPS 2026 (Dec) cousin paper — author and method overlap is the rule, not the exception across the big-three ML venues within a calendar year.

Key dates¶

Date	Event
Sep 19, 2025	Abstract submission deadline (AoE) — past
Sep 24, 2025	Paper submission deadline (AoE) — past
Nov 11, 2025	Reviews released — past
Dec 3, 2025	Discussion period ends — past
Jan 25, 2026	Decision notification (AoE) — past
Feb 28, 2026	Workshop acceptance notification — past
Mar 14, 2026	Early registration deadline — past
Apr 2, 2026	Registration cancellation deadline — past
Apr 23–25, 2026	Main conference (Riocentro, Rio de Janeiro) — past
Apr 26–27, 2026	Workshops — past
May 2026 → ongoing	SlidesLive recordings + camera-ready PDFs posting

Sources¶

Next step¶

May 2026: run the keyword filter against the public main-track accept list; identify the ~few-dozen bio-relevant papers; cross-check against the five bio workshop rosters.
May–Jun 2026: populate tools/ with the filtered subset; pull bioRxiv mirrors and SlidesLive recordings as they post; tag each entry with its cross-link to AACR 2026 agentic-AI or virtual-cells axes.
Jul 2026 (ICML overlap): cross-link ICLR 2026 bio tools to their ICML 2026 cousins where the same lab ships an evolved version.
Dec 2026 (NeurIPS overlap): repeat for NeurIPS LMRL / MLSB / AI4D3 cousin papers; consolidate the full-year ML-for-biology arc across the big-three venues.