בית הספר
להנדסת חשמל ומחשבים

From Subspace Matching to Neural Inversion: New Frontiers in Computational Signal Processing In the first part of the talk, I will introduce Signal Subspace Matching (SSM), a novel goodness-of-fit metric for comparing subspaces. Unlike traditional PCA-based methods, this approach is eigen decomposition-free and dimension-selection-free, as it avoids the need to explicitly choose a subspace dimension. The core of the method involves constructing soft-projection matrices from data using adaptive diagonal loading to ensure they remain robust estimates of the signal subspace. This metric effectively measures the distance between subspaces by implicitly utilizing the principal angles between them. I will demonstrate that this framework provides a powerful foundation for signal estimation and learning. In the second part of the talk, I will present my work on deep learning for solving 2D and 3D seismic inversion. I will describe how deep neural networks can replace computationally intensive Full Waveform Inversion to produce velocity models with millisecond latency. I will describe neural architectures for selecting a minimal subset of seismic sources, overcoming the "curse of dimensionality" in 2D and 3D inversion. Finally, I will present ongoing research on neural seismic fields, by learning a grid-less mapping from continuous coordinates to seismic field values, with applications such as redatuming (i.e., creating virtual data) in time-lapse imaging. Bio: Amir Adler is Associate Professor of Electrical Engineering at Braude College of Engineering, and a Visiting Associate Professor at the Department of Mathematics at MIT. He was a post-doctoral associate at the laboratory of Prof. Tomaso Poggio at the Center for Brains, Minds, and Machines (CBMM) at MIT, where he worked on deep learning and inverse problems. His research has been supported by grants and awards from the US-Israel Binational Science Foundation, Ministry of Science and Technology, TotalEnergies and Facebook (now Meta). His research interests are deep learning, signal processing, inverse problems and communication. Prof. Adler was named a 2025 Top Scholar by ScholarGPS, placing him in the top 0.5% of all scholars worldwide in the fields of signal processing and deep learning. He holds a Ph.D. in Computer Science, an M.Eng. in Electrical Engineering, and a B.Sc. in Electrical Engineering, all from the Technion –Israel Institute of Technology.