Events

I will discuss Borel equivalence relation and hyperfinitness following [Kerr-Li] Section 4.8,  [Loh1] Sections 3.1.1, 3.2.1, and [Fur1] Section 4.

For more details and the exact reference, click here.

Some of the most important open problems in group theory concern whether all groups can be metrically approximated by certain classes of groups. In particular, it is currently unknown whether there exist groups that are not (linear) sofic, hyperlinear, or MF.

We introduce the general problem of approximation of groups, as well as the related problem of stability, which asks whether almost homomorphisms from a group are close to actual homomorphisms. We recall a cohomological tool for stability introduced by De-Chiffre, Glebsky, Lubotzky, and Thom, which was used to prove the existence of groups that are not Frobenius-norm approximated. Moreover, we propose a weaker notion of semi-stability, which relates the different classes of approximated groups, and discuss how this cohomological tool can be adapted for this purpose.

For more details, check out our website.

We look forward to seeing you in one or both sessions next Monday.

In this talk, we examine the benefits of semantic vector representations for computer vision, highlighting their advantages over pixel-space representations in various applications. We first consider the problem of human motion anomaly detection and demonstrate the advantage of doing that over human poses. Then, we introduce a novel category-agnostic approach, termed GraphCape, that enables pose estimation across any category. Finally, we will explore further improvements for structure-based CAPE networks, dynamically predicting useful connections.

Bio: Or Hirschorn is a PhD candidate at Tel Aviv University, advised by Prof. Shai Avidan. His research interests are in developing methods for learning semantic vector representations of images and their applications for a variety of vision tasks. His MSc work won him the Weinstein scholarship for outstanding signal processing research.

TBD

The rapid growth of generative models allows an ever-increasing variety of capabilities. Yet, these models may also produce undesired content such as unsafe images, private information, or copyrighted material.

In this talk, I will discuss practical methods to prevent undesired generations. First, I will show how the challenge of avoiding undesired generations manifested itself in a simple Capture-the-Flag LLM setting, where even our top defense strategy was breached. Next, I will demonstrate a similar vulnerability in state-of-the-art concept erasure methods for Text-to-Image models. Finally, I will describe the notion of ‘Unconditional Concept Erasure’ aiming to mitigate such vulnerabilities. I will show that Task Vectors can achieve Unconditional Concept Erasure, and discuss the challenge of applying Task Vectors in practice.

Bio: Niv is a postdoctoral researcher at New York University hosted by Prof. Chinmay Hegde. He received a BSc in mathematics with physics as part of the Technion Excellence Program. He received his PhD in computer science from the Hebrew University of Jerusalem, advised by Prof. Yedid Hoshen. Niv was awarded the Israeli data science scholarship for outstanding postdoctoral fellows (VATAT). He is interested in anomaly detection, model personalization, and AI safety for Vision & Language models. 

TBD

TBD

TBD

Vizing's Theorem from 1964 states that any n-vertex m-edge graph of maximum degree Δ can be edge colored using at most Δ+1 different colors.

Vizing's original proof is algorithmic and implies that such an edge coloring can be found in O(mn) time.

In this talk, I'll present a randomized algorithm that computes a (Δ+1)-edge coloring in near-linear time -- in fact, only O(mlogΔ) time -- with high probability.