Tag: benchmark

A new benchmark called MATE-Hard shows that cooperative multi-agent reinforcement learning policies lose roughly half their advantage over random behavior ...

A new 200,000-question benchmark from Carnegie Mellon University reveals that leading vision-language models perform at random chance when interpreting engineering ...

A new benchmark called MathEval unifies 22 mathematical datasets and uses annually refreshed Gaokao exam problems to measure large language ...

A new benchmark reveals that drug synergy prediction models often rely on memorizing historical data rather than learning generalizable pharmacological ...

A new benchmarking framework called SpaBEAT defines four types of batch effects in spatial transcriptomics and shows that no correction ...