(The talk will be given in English)

Speaker: Dr. Ron Rothblum

Weizmann Institute

Monday, January 2nd, 2017 15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

How to prove the correctness of computations

Abstract

Efficient proof verification is at the heart of the study of computation. Seminal results such as the IP=PSPACE Theorem [LFKN92,Shamir92] and the PCP theorem [AS92,ALMSS92] show that even highly complicated statements can be verified extremely efficiently. We study the complexity of proving statements using interactive protocols. Specifically, what statements can be proved by a polynomial-time prover to a super-efficient verifier. Our main results show that these proof-system are remarkably powerful: it is possible to prove the correctness of general computations such that (1) the prover runs in polynomial-time, (2) the verifier runs in linear-time (and in some conditions in sublinear-time) and (3) the interaction consists of only a constant number of communication rounds (and in some settings just a single round). These proof-systems are motivated by, and have applications for, delegating computations in a cloud computing environment, and guaranteeing that they were performed correctly.

(The talk will be given in English)

Speaker:     Dr. Daniel Soudry
                    Department of Statistics, Columbia University

Wednesday, January 18^th, 2017
15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

Resource Efficient deep Learning

Abstract

Background: The recent success of deep neural networks (DNNs) relies on large computational resources (memory, energy, area and processing power). These resources pose a major bottleneck in our ability to train better models, and to use these models on low power devices (e.g., mobile phones). However, current generation DNNs seem tremendously wasteful, especially in comparison to the brain (which consumes only 12W). For example, typical DNNs use 32bit floating point operations, while the brain typically operates using binary spikes and with limited synaptic precision. Achieving such low precision in DNNs can significantly improve memory, speed and energy. However, until recently, 8 bits appeared to be the lowest possible limit.

Results: We show that it is possible to significantly quantize (even down to 1 bit) the activations and weights of DNNs trained by a variant of the backpropagation algorithm, while preserving good performance on various benchmarks (e.g., MNIST, ImageNet). Interestingly, the algorithm originated from first principles: we developed a closed form analytical approximation to the Bayes rule update of the posterior distribution of the binary DNN weights. At run-time, such a binarized DNN requires 32-fold less memory, is 7 times faster (using dedicated GPU kernels), and is at least 10-fold more energy efficient (using dedicated hardware). This can enable the use of trained DNNs in low power devices. Additional benefits are expected at train-time by further quantizing the gradients, potentially allowing larger and more sophisticated models to be trained.

 (The talk will be given in English)

Speaker:     Dr. Peter Winzer
                    Bell Labs

Monday, January 16^th, 2016
15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

From scaling disparities to integrated parallelism: Space-division multiplexing in fiber-optic communications

Abstract
The evolution of network traffic and communication technologies over the past 10+ years and their projections into the coming 10+ years reveal increasingly pronounced scaling disparities between technologies used to create and process data and technologies used to transport data. In core networks, we expect the need for 10+ Terabit/s transponders working over Petabit/s systems within the coming decade. However, with the help of digital coherent detection, advanced multi-dimensional modulation, shaping, and coding, these systems are now rapidly approaching recently established estimates for the Shannon capacity of the nonlinear fiber-optic channel. By 2020, leading-edge network operators will require capacities that are physically impossible to implement using conventional optical transmission technologies. Highly integrated spatially parallel optical transmission solutions (Space-Division Multiplexing, SDM) seem to be the only long-term viable path to overcome the looming optical networks capacity crunch. We discuss the implications of ultimately unavoidable spatial crosstalk in highly parallel SDM systems and examine how multiple-input-multiple-output (MIMO) digital signal processing, well established in wireless communications (albeit on a different set of boundary conditions), can be used to scale optical core networks. A look at information theoretic security enabled by SDM-based fiber-optic transmission systems rounds off our discussion.

Bio
Peter J. Winzer received his Ph.D. in electrical engineering from the Vienna University of Technology, Austria, in 1998. Supported by the European Space Agency (ESA), he investigated photon-starved space-borne Doppler lidar and laser communications using high-sensitivity digital modulation and detection. At Bell Labs since 2000, he has focused on various aspects of high-bandwidth fiber-optic communication systems, including Raman amplification, advanced optical modulation formats, multiplexing schemes, and receiver concepts, digital signal processing and coding, as well as on robust network architectures for dynamic data services. He contributed to several high-speed and high-capacity optical transmission records with interface rates from 10 Gb/s to 1 Tb/s, including the first 100G and the first 400G electronically multiplexed optical transmission systems and the first field trial of live 100G video traffic over an existing carrier network. Since 2008 he has been investigating and internationally promoting spatial multiplexing as a promising option to scale optical transport systems beyond the capacity limits of single-mode fiber. He currently heads the Optical Transmission Systems and Networks Research Department at Bell Labs in Holmdel, NJ. He has widely published and patented and is actively involved in technical and organizational tasks with the IEEE Photonics Society and the Optical Society of America (OSA), currently serving as Editor-in-Chief of the IEEE/OSA Journal of Lightwave Technology. He has been Program Chair of the 2009 European Conference on Optical Communications (ECOC) and Program Chair and General Chair of the 2015 and 2017 Optical Fiber Communication Conference (OFC). Dr. Winzer is a Bell Labs Fellow, a Fellow of the IEEE and the OSA, and a 2015 Thomson Reuters Highly Cited Researcher, the only one from industry in Engineering.
 

(The talk will be given in Hebrew)

Speaker:     Dr. Oren Eliezer
                       PHAZR

Wednesday, December 21^st, 2016
15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

5th Generation Mobile Communications - An Overview of Research and Challenges in the Related RF-Circuitry

Abstract

5th generation cellular communication systems are expected to be deployed in 2020, following the pace of one generation per decade that has been experienced ever since the first generation. While the 3GPP standardization group efforts are ongoing, some pre-standard systems are already being developed, demonstrating data-rates that can exceed 1Gpbs per user.

This seminar will present some of the RF-circuitry related challenges associated with the design and deployment of 5th generation cellular systems, and will highlight several different fields of ongoing research, which may offer career/business/research opportunities. The work of several leading researchers in the field will be discussed and some industry examples will be shown.

In particular, the challenges associated with efficient use of spectrum and power-efficient transmission will be discussed.

Bio
Dr. Oren Eliezer has 28 years of experience in the design and productization of wireless communication systems and ICs.

He received his BSEE and MSEE degrees from the Tel-Aviv University in 1988 and in 1997, focusing on communication systems and signal processing, and his PhD in microelectronics from the University of Texas at Dallas in 2008.

After serving for 6 years as an engineer in the IDF, he co-founded Butterfly Communications, which was acquired by Texas Instruments (TI) in 1999.

He was relocated by TI to Dallas in 2002, where he was elected senior member of the technical staff, and took part in the development of TI’s digital-radio-processor (DRP) technology.

Since 2009 he has been involved in academic research and teaching at the University of Texas at Dallas and the University of North Texas, as well as in entrepreneurship with several startup companies in the Dallas area.

He is currently with PHAZR, a pioneer in millimeter-wave wireless systems for 5th generation cellular communications.

He has authored and coauthored over 55 journal and conference papers and over 45 patents, and has given over 50 tutorials related to communication system and IC design and productization.

He is a senior member of the IEEE, regularly reviews papers for various IEEE publications and serves on the organizing committees of 4 different IEEE conferences.