As a Research Assistant under Dr. Helmuth, I led the implementation and enhancement of Probabilistic Lexicase Selection, building upon the groundbreaking work by Ding et al. (2023). My implementation in the Clojure-based CBGP/PushGP systems achieved remarkable results: a 18% increase in median solve-rate across a 10-benchmark symbolic regression suite and a 23% improvement in population diversity. Through innovative optimization techniques, including transients and batched fitness evaluation, I reduced per-generation runtime by 22%, enabling 10× larger hyper-parameter sweeps on our 64-core HPC cluster.

The project's scale was unprecedented, with 2.4 million evolutionary runs automated across 20 composite problems using SLURM and Datomic logging. I developed comprehensive Jupyter dashboards for real-time convergence analytics and selection-pressure heatmaps, providing unprecedented insights into evolutionary dynamics. My work extended beyond implementation, as I presented technical talks on genetic programming advancements at Push Language Discourse sessions, engaging faculty and PhD candidates from UMass Amherst and Amherst College.

• Research PaperProbabilistic Lexicase Selection (Ding et al., 2023)
• Implementationhttps://github.com/kitan23/cbgp-lite
• TechnologiesClojure, Python, SLURM, Datomic, Jupyter
• Key Achievements
  • • 18% increase in median solve-rate
  • • 23% improvement in population diversity
  • • 22% reduction in per-generation runtime
  • • 2.4M evolutionary runs automated
  • • 10× larger hyper-parameter sweeps enabled