Procedural Generation Of Algorithm Discovery Tasks in Machine Learning

Automating the development of machine learning algorithms has the potential to unlock new breakthroughs. However, our ability to improve and evaluate algorithm discovery systems has thus far been limited by existing task suites. They suffer from many issues, such as: poor evaluation methodologies; data contamination; and containing saturated or very similar problems. Here, we introduce DiscoGen, a procedural generator of algorithm discovery tasks for machine learning, such as developing optimisers for reinforcement learning or loss functions for image classification. Motivated by the success of procedural generation in reinforcement learning, DiscoGen spans billions of tasks of varying difficulty and complexity from a range of machine learning fields. These tasks are specified by a small number of configuration parameters and can be used to optimise algorithm discovery agents (ADAs). We present DiscoBench, a fixed, small subset of DiscoGen tasks for principled evaluation of ADAs. Finally, we propose a number of ambitious, impactful research directions enabled by DiscoGen, and demonstrate its use for ADA optimisation through scaling experiments for automated prompt tuning. DiscoGen is released open-source.

Usings agents to automatically discover machine learning algorithms has the potential to transform the field. However, optimising agents for this automatic discovery has proven difficult; in large part because there are not enough diverse tasks to learn from. We build DiscoGen, a system which procedurally generates tasks, rather than relying on an inherently limited fixed set of tasks. In addition to massively expanding the number of tasks we can sample, DiscoGen is designed to generate tasks with many desirable properties. We show that agents can be optimised in DiscoGen, and suggest a number of other research ideas that DiscoGen enables people to work on.