Computer Science Speakers Calendar

View Full Calendar

Research Short Talk: Brighten Godfrey, "Networking at the Speed of Light" & Edgar Solomonik, "Tensor Decomposition Algorithms and Applications: Linear Regression to AlphaTensor"

Event Type
Seminar/Symposium
Sponsor
Illinois Computer Science
Location
Only for CS Faculty
Date
Nov 7, 2022   12:00 pm  
Views
31

Brighten Godfrey:

Abstract: Low latency is a requirement for a variety of interactive network applications, and it's fundamentally a difficult goal. I'll talk about our group's efforts to improve network latency in two areas.  First, in the wide area, we explore the design of wide-area networks that move data at nearly the speed of light in vacuum. Our cISP design augments the Internet’s fiber with free-space microwave wireless connectivity over paths very close to great-circle paths. cISP addresses the fundamental challenge of simultaneously providing ultra-low latency -- within 5% of the speed of light -- while accounting for numerous practical factors ranging from transmission tower availability to packet queuing to weather interference.  Second, latency is also a challenge at the edge for interactive applications.  I'll discuss a method to combine multiple communication channels -- low latency and high bandwidth -- to improve application performance in 5g wireless, with applications to web browsing, mobile apps, and emerging applications.

Edgar Solomonik:

Abstract: Tensor decompositions have proven effective in unraveling structure and providing approximate models of multidimensional datasets in a variety of application domains. This talk will focus on applications and optimization algorithms for the canonical polyadic (CP) tensor decomposition, which provides the most natural generalization of matrix rank to tensors. A unique application of finding low-rank CP decompositions is searching for fast bilinear algorithms (e.g., FFT for convolution, Strassen's algorithm for matrix multiplication). We review a recent breakthrough in fast matrix multiplication algorithms (AlphaTensor, Nature 2022), based on numerical computation of CP decompositions.

link for robots only