A new index compiles accepted papers from ICML 2026, the machine learning conference held in Seoul, that come with publicly available code or data. The list was assembled by Paper Digest, a research platform based in New York City and founded in 2018, using an automated extraction process to find papers with linked repositories.
Each entry in the index shows the paper's title, its authors, a link to the code, and a short highlight excerpt describing the work. The compilers point to several examples. UniAR proposes a single visual tokenizer intended to bridge image understanding and image generation. tau²-Bench adds a dual-control telecom environment for evaluating AI agents. PlotCraft is a benchmark of 1,000 visualization tasks spanning multiple domains. Another paper uses physics simulators as a source of supervision for teaching large language models to reason about physics, while a controlled study examines the separate roles that pre-training, mid-training and reinforcement learning play in reasoning models.
The index is not exhaustive. Because it relies on automated extraction, it may miss some papers that do have public resources. In addition, some repositories may not become fully public until the conference itself begins, so the list reflects what could be found at the time of compilation rather than a final count.
For researchers and practitioners, a consolidated list of papers with usable code and data can make it easier to find and reproduce work presented at one of the field's major venues.
Credit. Based on reporting for Paper Digest. Is this your work? Write to us.
48
