About Me

My name is Daniel Israel, and I am a Computer Science PhD student at UCLA, co-advised by Guy Van den Broeck and Aditya Grover.

Accelerating LLM inference is often viewed as an exercise in rote optimization. For me, this work is driven by a curiosity about the fundamental limits of computation. I want to understand the theoretical bounds on how much memory or parallelism an inference algorithm can inherently exploit. I use principles from computer science and probability theory to narrow the search space of possible algorithms and identify promising directions to evaluate empirically. An important theme of my research is understanding tradeoffs and trying to push the Pareto frontier of speed and quality. I am interested in applying these ideas to diffusion language models, speculative decoding, and other probabilistic frameworks.

If you would like to collaborate, please reach out to me via email.