(The talk will be given in English)

Speaker:     Dr. Gal Vardi

TTI-Chicago and the Hebrew University

hall, Electrical Engineering-Kitot Building‏ 011

Monday, January 15^th, 2024

15:00 - 16:00

On Implicit Bias and Benign Overfitting in Neural Networks

Abstract

When training large neural networks, there are typically many solutions that perfectly fit the training data. Nevertheless, gradient-based methods often have a tendency to reach those which generalize well, namely, perform well on test data, and understanding this “implicit bias” has been a subject of extensive research. Surprisingly, trained networks often generalize well even when perfectly fitting noisy training data (i.e., data with label noise), a phenomenon called “benign overfitting”. In the first part of the talk, I will discuss the implicit bias and its implications. I will show how the implicit bias can lead to good generalization performance, but also has negative implications in the context of susceptibility to adversarial examples and privacy attacks. In the second part of the talk, I will discuss benign overfitting and the settings in which it occurs in neural networks.

Short Bio

Gal is a postdoctoral researcher at TTI-Chicago and the Hebrew University, hosted by Nati Srebro and Amit Daniely as part of the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning. Prior to that, he was a postdoc at the Weizmann Institute, hosted by Ohad Shamir, and a PhD student at the Hebrew University, advised by Orna Kupferman. His research focuses on theoretical machine learning, with an emphasis on deep-learning theory.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בטופס הנוכחות שיועבר באולם במהלך הסמינר

The talk will be given in English)

Speaker:     Dr. Joachim Neu

Stanford University

 hall, Electrical Engineering-Kitot Building‏ 011

Wednesday, January 10^th, 2024

15:00 - 16:00

Internet-Scale Consensus in the Blockchain Era

Abstract

Blockchains have ignited interest in Internet-scale consensus as a fundamental building block for decentralized applications and services, which promise egalitarian access and robustness to faults and abuse. While the study of consensus has a 40+ year tradition, the new Internet-scale setting requires a fundamental rethinking of models, desiderata, and protocols.

A key challenge is to simultaneously serve users with different security requirements. I will focus on two examples from my work: (1) Some users want uninterrupted service-availability even as protocol participants unexpectedly come and go, while other users demand that perpetrators can be held accountable in case of service-consistency breaches. We show that no traditional protocol can satisfy both simultaneously. To resolve this dilemma, we develop the ebb-and-flow consensus paradigm, which has been adopted in Ethereum (the largest freely-programmable blockchain). (2) Some users wish to harden consistency even at the expense of availability, while others prefer to stick to traditional consensus protocols' equal resilience to consistency and availability attacks. We present the first protocols that allow optimal availability-consistency tradeoff for every user simultaneously.

Short Bio

Joachim Neu is a PhD candidate at Stanford University advised by David Tse. His research focuses on the science and engineering of Internet-scale consensus as a fundamental building block for decentralized systems, using tools from distributed systems, probabilistic systems analysis, networking and communications, and applied cryptography. While a Masters student at Technical University of Munich, he worked in information and coding theory. Joachim has received the Protocol Labs PhD Fellowship and the Stanford Graduate Fellowship.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בטופס הנוכחות שיועבר באולם במהלך הסמינר