סמינרים

• Understanding and Enhancing Deep Neural Networks with Automated Interpretability

  Abstract: Deep neural networks are becoming incredibly sophisticated; they can generate realistic images, engage in complex dialogues, analyze intricate data, and execute tasks that appear almost human-like. But how do such models achieve these abilities? In this talk, I will present a line of work that aims to explain the behaviors of deep neural networks.… Continue Reading Understanding and Enhancing Deep Neural Networks with Automated Interpretability

  ---

• Bayesian Persuasion in Networks: Divisibility and Network Irrelevance

  Abstract: We study a multiple-receiver Bayesian persuasion model in which the sender wants to persuade a critical mass of receivers. Receivers are connected in a network and can perfectly observe their immediate neighbors' signals, which complicates the problem of the sender. We simplify the problem by considering signaling schemes ("experiments") in which certain receivers are… Continue Reading Bayesian Persuasion in Networks: Divisibility and Network Irrelevance

  ---

• Predicting and Analyzing High-Level Cognitive Traits Using Computational Multiplex Networks and Vector Representations

  Abstract: High-level cognition, such as intelligence and creativity, are considered the hallmark of human cognition; however, their complexity hinders the identification of underlying common mechanisms. We focus on one such likely mechanism – mental navigation. We utilize converging computational methods to demonstrate how mental navigation – operationalized via verbal fluency tasks—predicts individual differences in creativity,… Continue Reading Predicting and Analyzing High-Level Cognitive Traits Using Computational Multiplex Networks and Vector Representations

  ---

• The edge-averaging process on graphs with random initial opinions

  Abstract:  In several settings (e.g., sensor networks and social networks), nodes of a graph are equipped with initial opinions, and the goal is to estimate the average of these opinions using local operations. A natural algorithm to achieve this is the edge-averaging process, where edges are repeatedly selected at random (according to independent Poisson clocks)… Continue Reading The edge-averaging process on graphs with random initial opinions

  ---

• Two Lenses on Deep Learning: Data Reconstruction and Transformer Structure – Job Talk

  Abstract: Despite the remarkable success of modern deep learning, our theoretical understanding remains limited. Many fundamental questions about how these models learn, what they memorize, and what their architectures can express are still largely open. In this talk, I focus on two such questions that offer complementary perspectives on the behavior of modern networks. First, I examine how… Continue Reading Two Lenses on Deep Learning: Data Reconstruction and Transformer Structure – Job Talk

  ---

• Interpreting the Inner Workings of Vision Models

  Abstract: In this talk, I present an approach for interpreting the internal computation in deep vision models. I show that these interpretations can be used to detect model bugs and to improve the performance of pre-trained deep neural networks (e.g., reducing hallucinations from image captioners and detecting and removing spurious correlations in CLIP) without any… Continue Reading Interpreting the Inner Workings of Vision Models

  ---

• Deep Learning for approximating solutions to hard computational problems indirectly

  Abstract: Computational reduction serves as a powerful tool in optimization by enabling the translation between different computational problems while preserving key structural and computational properties. This allows one problem to be reformulated in terms of another, often facilitating the solution of otherwise complex problems. Since reductions are mostly manual, it is plausible that alternative, better… Continue Reading Deep Learning for approximating solutions to hard computational problems indirectly

  ---

• Fundamentals of Aligning General-Purpose AI – Job Talk

  Abstract: The field of artificial intelligence (AI) is undergoing a paradigm shift, moving from neural networks trained for narrowly defined tasks (e.g., image classification and machine translation) to general-purpose models such as ChatGPT. These models are trained at unprecedented scales to perform a wide range of tasks, from providing travel recommendations to solving Olympiad-level math problems.… Continue Reading Fundamentals of Aligning General-Purpose AI – Job Talk

  ---

• Cooperation by Design

  ABSTRACT Cooperation failures often produce suboptimal outcomes across numerous interactions. How can we redirect interactions away from lose-lose and win-lose outcomes and toward mutually beneficial, win-win outcomes? This talk considers myriad pathways to promoting cooperation across multiple contexts, ranging from interactions between robbers and victims to those between wedding planners and their clients. Our research… Continue Reading Cooperation by Design

  ---

אנשים

• 

  אבי-יצחק בנימין (בני)

  In Memoriam: Benjamin Avi-Itzhak (1933–2017) Professor Emeritus Benjamin Avi-Itzhak—or Benny, as we knew him—was born in Jerusalem in 1933 and passed away in November 2017. Benny obtained a BSc degree in Mechanical Engineering in 1955, an additional BSc in Industrial Engineering and Management (IE&M) in 1960, an MSc in Operations Research also in 1960, and… Continue Reading אבי-יצחק בנימין (בני)

  ---