icml

ICML 2026 announces its award winners, including Outstanding Paper awards (research and position papers) and the Test of Time Award, with details on the selection process.

The Red Queen Gödel Machine enables recursive self-improvement in AI by co-evolving the agent and evaluator, achieving better coding performance with fewer tokens.

A new paper shows that scaling inference compute via methods like self-consistency improves LLM accuracy in math and code but fails to improve truthfulness in domains without external verifiers, as model errors are too correlated.

This paper presented at ICML explores how causal and statistical models can generalize to novel combinations of interacting objects, with a poster session scheduled at the conference.

This paper observes that token embeddings in small language models condense into a narrow cone-like subspace, a phenomenon termed embedding condensation, and proposes a dispersion loss to counteract it, improving generalization.

This paper proposes SpeechCombine, an instruction-following speech language model trained without any instruction tuning, using only speech pre-training and a weight combination strategy that transfers text LLM capabilities to the speech domain.

This paper systematically compares discrete, continuous, and hybrid value encoding strategies for transformers in electronic health record data, finding that hybrid token-based approaches with binning provide robust performance and are recommended as a practical default.

This paper introduces OpenAgent, a problem setting for tool-use agents in open-world scenarios with distributional shifts, and proposes Perturbation-Augmented Fine-Tuning to improve robustness. Experiments reveal that both SFT and RL agents degrade under environmental shifts.

NEO is a new type of world model that learns to discover reusable building blocks of explanation from raw observations without supervision or language, selected as an ICML 2026 oral presentation.

This paper revisits the volume hypothesis, which posits that generalization in over-parameterized networks is mainly due to the larger volume of good-generalizing regions in weight space rather than SGD's implicit bias. Through experiments with binary networks, the authors show that the generalization advantage of gradient learning over random sampling diminishes as training data size grows, potentially resolving contradictory prior findings.

Kyutai presents Hibiki-Zero, a real-time speech-to-speech translation model, at ICML 2026 in Seoul, with an oral presentation scheduled for July 8.

This paper presents the first comprehensive empirical study of safety impacts of benign multilingual fine-tuning on LLMs, showing that safety outcomes vary drastically by language and that assessing only English is insufficient.

Memora is a scalable memory system for AI agents that decouples storage from retrieval, enabling long-horizon tasks with up to 98% fewer context tokens while setting new state-of-the-art on benchmarks. The paper is published at ICML 2026.

This paper update presents a universal sequence preconditioning method achieving dimension-free regret bounds for marginally stable linear dynamical systems, using second-order VAW algorithm and Faber polynomials.

Introduces 'skill neologisms', a method for enabling LLMs to learn new skills without weight updates, addressing catastrophic forgetting. Presented at ICML.

This paper introduces LaViD, a framework that transfers semantic knowledge from a language-only LLM to a vision student model by generating multiple-choice questions as conceptual signatures, achieving superior fine-grained classification performance and robustness.

A visual web app that indexes over 6000 ICML papers, allowing users to explore the paper landscape by topic.

CAT-Q introduces a post-training ternary quantization method for LLMs that uses learnable modulation and softened ternarization, achieving superior performance over BitNet 1.58-bit while using only 512 calibration samples and scaling to 235B parameters.

LithoDreamer is the first physics-informed World Model framework for computational lithography, modeling the multi-stage lithography process as a decision-driven system. It achieves state-of-the-art performance in forward evolution and inverse planning for semiconductor manufacturing.

A tweet highlighting the most-cited papers and most-starred GitHub repos related to ICML 2026, with publicly available code and workflows to map papers to Semantic Scholar citation data and GitHub repos.