Research on algorithms, machine learning, AI evaluation, and accountable systems. Explore computational experiments, software, and technical evidence interfaces.
SSRN Working Papers present developed research. Some are maintained as standalone working papers by design. Research maturity and publication status are recorded separately.
10 papers · Academic publication first · record updated 2026-09-06
An empirical regime map of which structural constraints a Minecraft voxel generator can learn.
FDG 2026 17th PCG Workshop · Accepted 2026-05-02 · Published 2026-08-09
Two AI systems can reach the same scientific conclusion for different reasons: one may generate the candidate sooner, another may check it more cheaply, and a third may obtain decisive evidence earlier.
A portable evidence interface linking AI-agent logging obligations to independent audit.
These working papers include early ideas, initial drafts, and work awaiting further development or verification. They are shared as work in progress.
We report covariance-induced irreversible drift, a failure mode of CMA-ES in barrier-constrained optimization.
The natural proofs barrier (Razborov--Rudich 1997) has had no machine-verified formalization in any proof assistant.
We study the geometry of circuit-SAT solution spaces over the complete binary basis B₂.
AI agents access sensitive data across heterogeneous platforms, yet no open specification defines the exposure event semantics needed to make this access auditable.
Models trained on canonical representatives of equivalence classes under group symmetry can exploit representation artifacts rather than learning invariant structure.
We introduce the first RL-based dynamic algorithm configuration (DAC) system for MaxSAT local search.
Can a fixed, user-deployable tactic string — requiring no solver modification — materially improve Z3's performance on nonlinear integer arithmetic? We show that the answer is yes.
