Simons Institute | Talk By Tal Herman @SimonsInstituteTOC | Uploaded 1 week ago | Updated 7 hours ago
Tal Herman (Weizmann Institute of Science)
https://simons.berkeley.edu/talks/tal-herman-weizmann-institute-science-2024-08-06
Workshop on Local Algorithms (WoLA)

Whistle variability and social acoustic interactions in bottlenose dolphins

The Role of Prior Data in Rapid Learning of Motor Skills

Data is as Data Does: The Influence of Computation on Inference

The approximate structure of triangle-free graphs

What neural machinery is needed for language acquisition (Virtual Talk)

Faces, Flukes, Fins, and Flanks: How Multispecies Re-ID Models are Transforming Our Approach to AI..

What Does Machine Learning Have to Offer Mathematics? | Theoretically Speaking

$Sublinear Insights: A Faster (Classical) Algorithm for Edge Coloring Sepehr Assadi (University of Waterloo and Rutgers University) https://simons.berkeley.edu/talks/sepehr-assadi-university-waterloo-rutgers-university-2024-06-20 Extroverted Sublinear Algorithms Vizings theorem states that every graph with maximum degree Delta admits a proper (Delta+1) edge coloring. The state-of-the-art algorithms for finding such a coloring on n-vertex m-edge graphs take O(msqrt(n)) time and date back to the 80s. In this talk, we present a very recent randomized algorithm for finding a Delta+O(log n) coloring in O(m*log(Delta)) time, namely, a near-linear time algorithm for a near-Vizings theorem. This algorithm is inspired by several tools developed in the study of sublinear algorithms. As a corollary of this result and Vizings original approach, we also obtain an O(n^2 log(n)) expected time algorithm for the (Delta+1) edge coloring problem. This is nearly optimal for dense graphs and improves upon the longstanding O(n^{2.5}) bounds of prior work for Vizings theorem in almost four decades. Based on https://arxiv.org/abs/2405.13371 (see also https://arxiv.org/abs/2405.15449 for an independent and concurrent work that obtains an ~O(m*n^{1/3}) time randomized algorithm for Vizings theorem)$

Plenary Talk: Privately Evaluating Untrusted Black-Box Functions

ML Efficiency for Large Models: From Data Efficiency to Faster Transformers

Talk By Tal Herman @SimonsInstituteTOC