[Abstract]

Continuous computations, which involve continuous data and operations, are fundamental to many areas, including machine learning and scientific computing. Examples include evaluating math functions (e.g., numpy.exp(x)) and computing derivatives (e.g., jax.grad(f)). In practice, we typically assume these computations are “correct” in an informal sense. However, when analyzed rigorously, they often turn out to be “incorrect” in a formal sense due to the use of floating-point numbers (instead of the reals), non-differentiable functions (instead of differentiable ones), etc. I have studied this gap between the theory and practice of continuous computations, aiming to make these computations more reliable and rigorous. In this talk, I will explain the nature of this gap and how it can be addressed, focusing on several classes of computations such as function evaluation and differentiation.

[Biography]

Wonyeol Lee is an Assistant Professor of Computer Science at POSTECH. Before joining POSTECH, he was a Postdoctoral Associate at CMU, working with Feras Saad. He received his PhD degree in Computer Science from Stanford University, advised by Alex Aiken. During his PhD, he also spent time at KAIST for military service, working with Hongseok Yang. He obtained his BS degree in Computer Science and Mathematics from POSTECH. His research spans programming languages and machine learning, with a focus on making continuous computations more reliable and scalable.