(The talk will be given in English)

Speaker:     Dr. Daniel Amir

                                Viterbi Faculty of Electrical and Computing Engineering, Technion

Monday, April 7^th, 2025

13:00 - 14:00

Optimal Oblivious Reconfigurable Networks

Abstract

As datacenter network traffic grows explosively, outpacing Moore's law, packet switching capabilities are struggling to meet the demands of modern data centers. Recent hardware advances have enabled the new switching technology of fast, nanosecond-scale circuit switches. These have the potential to fully replace packet switches, but realizing this potential requires novel network designs.

In this talk, I will present my research into the Oblivious Reconfigurable Network (ORN) design paradigm, which is ideally suited to this new switching technology. I will present Shale, the first ORN design capable of supporting a datacenter-scale network, achieving orders of magnitude better latency and memory requirements than prior designs at this scale. Shale achieves a tunable tradeoff between throughput and latency, and is enabled by novel techniques including interleaving, congestion control for multi-hop ORNs, and unique hardware optimizations. For each of Shale's tunings, the tradeoff between throughput and latency scaling is Pareto optimal among all ORN designs. Finally, I will discuss follow-on work on Semi-Oblivious Reconfigurable Networks (SORNs), which enhance ORNs through periodic optimization based on network demand, further improving the performance possible on fast circuit-switched networks.

Short Bio

Daniel Amir is a Postdoctoral Fellow at the Technion, supervised by Mark Silberstein. His research focuses on developing next-generation networks that leverage cutting-edge hardware to address the network demands of the future. He recently completed his Ph.D. in Computer Science at Cornell University, where he was advised by Hakim Weatherspoon and collaborated extensively with Robert Kleinberg. His dissertation work focused on developing Shale, the first oblivious reconfigurable network design which can scale to an entire datacenter. Prior to his Ph.D., he earned undergraduate degrees in Computer Science and in Engineering Physics from the University of Illinois Urbana-Champaign.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא במודל בתחילת הסמינר

Abstract
Generative AI has made tremendous strides over the last few years in a wide variety of fields, including text, images, audio, and video. In this talk, we discuss the use of Generative AI techniques in the realm of molecules, emphasizing the incorporation of invariances to transformation groups, and covering three applications. In the first, we show an approach to the problem of generating molecules which will bind to a particular receptor molecule, a problem with strong applications in drug design. We design specialized normalizing flows which respect the physical invariances inherent in the problem, through the use of semi-equivariant networks. In the second application, we show how to adapt diffusion models to deal with this same problem. In particular, we address the size disparity between the receptor and the generated molecule, which can be problematic for learning as the receptor can overwhelm the training; we do so by creating a small sketch of the receptor, dubbed a “virtual receptor”. In the final scenario, we address a fundamental problem with applications in chemistry, biochemistry and materials science: computing the quantum ground state of a molecule. We demonstrate an efficient method of solving the Electronic Schrodinger Equation by using a carefully designed antisymmetric normalizing flow to construct the wavefunction ansatz.

  -סמינר זה יחשב כסמינר שמיעה לתלמידי תואר שני ושלישי-  This Seminar Is Considered A Hearing Seminar For Msc/Phd Students