Peter Tang

About Peter Tang

A numerical analyst by training, I most enjoy working in a variety of problems with diverse teams where each member contributes unique insight to advance a common goal. Machine learning is becoming an all important field and I have collaborated with academia as well as colleagues within my company. With Northwestern University, we publish a paper on the adverse effect on test accuracy when training with large batch (ICLR 2017) which prompted active research efforts to mitigate this problem. Last year, we also published an adaptation of the second-order L-BFGS algorithm for deep learning training (ICML 2018). A new paradigm in neural network is spiking neural network which has a potential to offer significant power saving. Together with my colleagues at Intel Corporation, we designed a first gradient-based and mathematically rigorous learning algorithm using only local synaptic information (ICLR 2019).In high-performance scientific computing I focus on algorithm design, convergence proof and accuracy analysis. Examples include a new FFT algorithm (Supercomputing 2012, best paper), convergence proof of a new eigenvalues/vectors algorithm, FEAST (SIMAX 2014), and a fast synthetic aperture algorithm for high-resolution radar (Supercomputing 2012, best paper nominee).Attracted by a novel and interdisciplinary computer design project, I joined D E Shaw Research to work on Anton 2, a supercomputer for molecular dynamics. I designed the computation method and data path for the machine\'s arithmetic unit (PPIP) in a hardware-software-application co-design environment, I am most proud of my invention of the topological ID method inside PPIP. The teams\'s work won a Gordon Bell Prize in 2014I worked also in computer arithmetic and proposed the software friendly BID encoding for floating-point decimals to the 754 committee which subsequently standardized it in 75•••08. I was elected SIAM fellow (2015) for my diverse contributions.

Experience

1. Principal Member of Technical Staff

   Rivos Inc

   Sep 2022 — Present · Mountain View, CA, US

Education

• University of California, Berkeley

  Ph. D., Mathematics

  1980 — 1987

Skills

• Parallel Computing
• Algorithms
• Simulations
• Mpi
• Signal Processing
• Computer Architecture
• Distributed Systems
• Parallel Programming
• Computer Science
• Cuda
• Scientific Computing
• High Performance Computing
• Image Processing