Publications
Research papers.
- 1.Mixture-of-Steering Vectors (MoSV): Sparse Gating for Compositional Hallucination Mitigation
ICML 2026 Workshop on Mechanistic Interpretability · 2026
Vedant Srinivas, Daniel Lee, Feolu Kolawole
A framework that reduces LLM hallucinations by dynamically selecting from multiple learned correction vectors per prompt, improving factual accuracy by +2.4pp where prior methods gained only +0.3pp, and automatically discovering distinct hallucination types without manual labels across 10,615 items spanning 8 knowledge domains.
- 2.Cross-Game Semantic Alignment of Latent Action Representations
IJCAI 2026 Workshop on Generalizing from Limited Resources in the Open World (GLOW) · 2026
Feolu Kolawole, Khizer Khaderi
Studies when a single inverse dynamics model trained on multiple visually-disparate games produces action embeddings that align by action class rather than game identity. Across racing, Atari, and first-person domains, identifies a closable “calibration gap” as the primary failure mode, and shows a 50-label-per-class centroid calibration recovers 91.1% accuracy on Pole Position — within 0.3pp of a fully-supervised single-game classifier.