Ali Hussaini Umar

Local algorithms can outperform theoretical predictions in constrained optimization scenarios, revealing a critical gap between finite and asymptotic performance.

Human-readable prompts can be generated with significantly lower perplexity using a novel Bayesian approach, transforming how we interact with LLMs.

Representational alignment doesn't guarantee better generalization, and can actually be *minimized* near the interpolation threshold, challenging the assumption that similar representations always lead to better models.