Introducing Dr. Amir Weiss

Imagine a smart home with a virtual assistant (like Amazon Alexa), several smartphones, and a smart watch. Each has a built-in microphone, collects data, and can share this information — or at least some of it — with us. “These devices operate independently. They’re originally designed to do specific things, not to serve as a network of sensors that operate harmoniously for a mutual purpose. But they can, if we connect them all through a central computing unit, even just momentarily. In that brief moment of collaboration, what we want them to convey is not the raw data they measure, but only the most vital information concerning the specific parameter required by the computing unit,” explains Dr. Amir Weiss. “Translating this to mathematics, this problem can be described as ‘joint data compression and estimation’ or ‘estimation-oriented compression’. This is one of the main projects I’m working on these days. Surprisingly, it’s a relatively underexplored research area. The traditional purpose of compression is to produce the best possible recovered version of the compressed signal. However, we’re not interested in how well we can recover the original signal, but rather how accurately we can estimate — or make a statistically educated guess — the parameter in which we’re interested, such as a speaker’s location in the room.”

Dr. Weiss arrived at the Faculty of Engineering in April 2024 fresh out of reserve duty. He lives in Tel-Aviv, is married to Lior, and has two children — Aylon and Shaked. He began his academic career at the Faculty of Engineering at Tel-Aviv University. “I was a career officer, so I started relatively late. I did the fast-track graduate program in the field of statistical signal processing while also working part-time at Elisra, a company specializing in developing electronic warfare solutions. My original plan was to continue working in the industry, but toward the end of my master’s degree, my supervisor, Prof. Arie Yeredor, proposed that I continue working with him on another project for which he was looking for a PhD student. I remember considering my options and the decision wasn’t immediate — people around me had been working for years in the industry, and a PhD meant four more years, at least. What tilted the scales was the fact that I was about to complete my thesis and was already missing doing research — that and the fact that I really enjoyed working with Arie. He is a wonderful person whom I highly appreciate, both personally and professionally. To this day, I feel his influence and draw inspiration from him, for instance in how I approach teaching and in the way I aspire to advise my students.

His doctoral research dealt with blind source separation. “Source separation deals with problems where only mixtures of multiple signals can be measured, and the goal is to separate them and obtain a clean version of each signal. The term ‘blind’ refers to a framework where one makes as few assumptions about the model as possible — to see or know as little as possible in advance — so as to lead to a solution that is as general and robust as possible, which will work in a wide range of conditions with minimum assumptions. For example, if a system has one receiver and multiple antennas, each antenna receives some mixture of several signals; this mixture varies from one antenna to the other depending on factors such as the antenna’s location. Sometimes the antenna array configuration is known in advance — but sometime it is not, so an algorithm must be designed to operate under the assumption that this information is unknown. It's a beautiful and challenging mathematical problem, and I focused on its more theoretical aspects.”

After completing his PhD Amir took on a one-year postdoctoral position in the Department of Applied Mathematics and Computer Science at the Weizmann Institute of Science, working with Prof. Boaz Nadler. He then continued to a three-year postdoctoral position at the Massachusetts Institute of Technology (MIT), working with Prof. Greg Wornell, leading of the “Signals, Information and Algorithms Laboratory” at the Research Laboratory of Electronics (RLE). “I worked on several interesting projects. In one of them, we examined how machine learning could be used to develop technologies for the U.S. Navy, specifically in underwater acoustic localization. The physics of the underwater environment is relatively complicated to describe analytically because it involves the interaction of acoustic waves with the dynamic environment of ocean waves and the seabed. It’s a fascinating and complex problem, and it seems that combining classical signal processing approaches with modern machine learning techniques could lead to scientific breakthroughs that could benefit both defense and civilian applications in areas such as traffic monitoring and management in port environments, area surveying with autonomous underwater vehicles, and vessel detection and classification above and below water.

"In another project, we similarly explored machine learning applications for the U.S. Air Force. In this collaboration with MIT Lincoln Laboratory, we worked on single-channel RF signal separation, mainly for communication systems, using machine learning tools. Interestingly, it turned out that this is closely related to my PhD in terms of the problem — but here, the approach to the solution was entirely different, relying on deep artificial neural networks and modern tools that, until recently, were more commonly used in the fields of image and natural language processing. In this context, the question constantly arises as to whether it is even possible to make further breakthroughs in the field of communications. Many argue that this field, in which we have achieved so much scientifically and technologically over the past few decades, is already quite saturated, as there are fundamental limitations that cannot be surpassed and that we appear to be approaching. Then again, history is filled with people who have said: ‘That's it, there’s nothing more we can do,’ only for another breakthrough to suddenly occur. I my opinion, the use of machine learning in the field of communications hasn’t been fully explored yet. There is a significant challenge here, along with substantial potential for improvement, and in this project, I delved deeper into machine learning and expanded the range of problems and solution approaches I work on.”

In December 2022, after two years in Boston, Dr. Weiss returned to Israel to remotely continue his MIT postdoctoral research. When the war broke out, he was immediately drafted and spent six months away from home. During that time he learned that he had been accepted to the communications track of the electrical engineering program at Bar-Ilan University. “I continue to explore the things I did during my postdoctoral research, in addition to other problems related to data compression,” he says. Dr. Weiss is now seeking master’s and doctoral students with a strong mathematical background, a passion for signal processing and machine learning, and preferably practical experience in machine learning. Sounds interesting? Contact Dr. Weiss by email: amir.weiss@biu.ac.il