Semantics for 2D Rasterization

This feed bundles several recent arXiv papers across programming languages, formal methods, ML systems, and security-oriented program analysis. Highlights: (1) μSkia — a mechanized formal semantics (in Lean) for the Skia 2D rasterization library and a verified optimizer yielding ~18.7% speedups on real web workloads; (2) DVM — a bytecode-based real-time operator compiler and fusion framework for dynamic AI models that greatly reduces compilation latency and improves operator efficiency; (3) Likelihood hacking in probabilistic program synthesis — defines ‘likelihood hacking’ where RL-trained LM