51 citations
4 papers
ClosureBench: A Constructive Benchmark for Compositional Graph Reasoning
Stefano Goria
Large language models fail on multi-step compositional reasoning, but measuring that failure is hard, because new models are trained on the benchmarks used to evaluate them. A fixe…
Prosecution of complex criminal networks: a multilevel ERGMs approach to CICIG's judicial cases
H. Waxenecker, I. Luna-Pla, J. R. Nicolás-Carlock
Prosecutors are essential in combating organized crime, making key decisions about prosecution, target selection, and structuring imputation strategies. Despite their importance, t…
FreeRide: Harvesting Bubbles in Pipeline Parallelism
Jiashu Zhang, Zihan Pan, Molly +3
The occurrence of bubbles in pipeline parallelism is an inherent limitation that can account for more than 40% of the large language model (LLM) training time and is one of the mai…
Superconductivity and magnetism in Iron Sulfides Intercalated by Metal Hydroxides
Xiuquan Zhou, Christopher Eckberg, Brandon Wilfong +4
Inspired by naturally occurring sulfide minerals, we present a new family of iron-based superconductors. A metastable form of FeS known as the mineral mackinawite forms two-dimensi…