Help AI discover better algorithms for science.
Build a research environment around an important scientific problem, or collaborate with us to develop and validate new methods in your field.
Contribute a scientific research task.
Bring a fundamental, impactful problem from your field where better algorithms could make a difference. Turn it into a measurable task and build the complete environment for developing and testing new methods.
- Scientific valueExplain the advance a better method would enable.
- Room for discoveryShow where existing methods fall short and what can be improved.
- Meaningful evaluationDefine inputs, outputs, and metrics that measure scientific progress.
Have a task idea? Discuss it in an issue, or submit a draft PR with missing pieces clearly marked.
Scientific question
Optimization · Prediction · Classification · Generation
- Data pipelineRebuildable data from pinned sources, with train / development / test splits.
- Research runtimeDependencies and a workspace for running experiments.
- Reproduced baselineA reference method, runnable commands, and verified results.
- Development evaluationMetrics and tests for iterative research.
- Independent testingRerun the submission on held-out inputs to assess generalization.
- Submission interfaceA portable model or harness with defined inputs and outputs.
Collaborate on scientific discovery.
Partner with us on a research question in your field. Combine your scientific expertise, data, or experimental validation with AI researchers that design, implement, and test new methods.
- Shape the research questionDefine what progress would matter scientifically.
- Bring data or validationProvide datasets, simulators, or experiments that can test the ideas.
- Develop and assess methods togetherExplore algorithms with AI researchers and interpret the results.
Share your research question, the resources you can bring, and the collaboration you have in mind.
A shared scientific question
Domain researchers
Scientific judgment
Data & validation
AI researchers
Method design
Experiments & evaluation