Abstract:
Numerical application is an integral part of computer science. Scientific applications rapidly advance science and improve people’s lives, from weather forecast, brain map atlas, to AI. However, developing accurate and efficient numerical computation techniques still remain a challenging research area due to limited-precision computation. A problem as simple as fast and accurate approximation of elementary functions is still challenging and mainstream math libraries did not produce correct results until recently.
This talk presents the RLibm project. This project generates efficient and correctly rounded elementary functions for various representations. It makes a case for approximating the correctly rounded results of an elementary function rather than the real value of an elementary function. Additionally, it uses a non-standard rounding mode known as round-to-odd to produce correctly rounded results for multiple representations with a single implementation. The results from the RLibm project has made a strong case for mandating correctly rounded math libraries and the mainstream compilers have begun to adopt correctly rounded libraries.
Biography:
Jay Lim joined UC Riverside as an assistant professor in computer science and engineering in July 2025 after receiving Ph.D at Rutgers University in 2021. His research focus is in programming languages, systems, and numerics. He has been working on verifying widely-used and often-trustd systems (e.g., compilers and open source libraries) are indeed correct. His current research topics include generating correct approximations of elementary functions, producing correct results of scientific algorithms using low-precision representations, and implementing accurate numerical programs with significant resource constraints. Before joining UC Riverside, Jay was a lecturer at Yale University for four years teaching several systems courses.
Abstract:
Numerical application is an integral part of computer science. Scientific applications rapidly advance science and improve people’s lives, from weather forecast, brain map atlas, to AI. However, developing accurate and efficient numerical computation techniques still remain a challenging research area due to limited-precision computation. A problem as simple as fast and accurate approximation of elementary functions is still challenging and mainstream math libraries did not produce correct results until recently.
This talk presents the RLibm project. This project generates efficient and correctly rounded elementary functions for various representations. It makes a case for approximating the correctly rounded results of an elementary function rather than the real value of an elementary function. Additionally, it uses a non-standard rounding mode known as round-to-odd to produce correctly rounded results for multiple representations with a single implementation. The results from the RLibm project has made a strong case for mandating correctly rounded math libraries and the mainstream compilers have begun to adopt correctly rounded libraries.
Biography:
Jay Lim joined UC Riverside as an assistant professor in computer science and engineering in July 2025 after receiving Ph.D at Rutgers University in 2021. His research focus is in programming languages, systems, and numerics. He has been working on verifying widely-used and often-trustd systems (e.g., compilers and open source libraries) are indeed correct. His current research topics include generating correct approximations of elementary functions, producing correct results of scientific algorithms using low-precision representations, and implementing accurate numerical programs with significant resource constraints. Before joining UC Riverside, Jay was a lecturer at Yale University for four years teaching several systems courses.