Frontiers in Electrical Engineering

===

Wenqi Jiang

(Presenting 9:30 am - 10:30 am)

Title: Beyond Model Acceleration in Next-Generation Machine Learning Systems

Abstract: Despite the recent popularity of large language models (LLMs), the transformer neural network invented eight years ago has remained largely unchanged. It prompts the question of whether machine leanring (ML) systems research is solely about improving hardware and software for tensor operations. In this talk, I will argue that the future of machine learning systems extends far beyond model acceleration. Using the increasingly popular retrieval-augmented generation (RAG) paradigm as an example, I will show that the growing complexity of ML systems demands a deeply collaborative effort spanning data management, systems, computer architecture, and ML.

I will present RAGO and Chameleon, two pioneering works in this field. RAGO is the first systematic performance study of retrieval-augmented generation. It uncovers the intricate interactions between vector data systems and models, revealing drastically different performance characteristics across various RAG workloads. To navigate this complex landscape, RAGO introduces a system optimization framework to explore optimal system configurations for arbitrary RAG algorithms. Building on these insights, I will introduce Chameleon, the first heterogeneous accelerator system for RAG. Chameleon combines LLM and retrieval accelerators within a disaggregated architecture. The heterogeneity ensures efficient serving of both LLM inference and retrievals, while the disaggregation enables independent scaling of different system components to accommodate diverse RAG workload requirements. I will conclude the talk by emphasizing the necessity of cross-stack co-design for future ML systems and the abundant of opporutnities ahead of us.

Bio: Wenqi Jiang is a final-year PhD student at ETH Zurich, advised by Gustavo Alonso and Torsten Hoefler. He aims to enable more efficient, next-generation machine learning systems. Rather than focusing on a single layer in the computing stack, Wenqi's research spans the intersections of data management, computer systems, and computer architecture. His work has driven advancements in several areas, including retrieval-augmented generation (RAG), vector search, and recommender systems. These contributions have earned him recognition as one of the ML and Systems Rising Stars, as well as the AMD HACC Outstanding Researcher Award.

===

Jennifer Tang

(Presenting 10:45 am - 11:45 am)

Title: Foundations for Efficient Information Usage

Abstract: A defining feature of the modern information age is the widespread adoption of technologies which rely on, and generate, vast amounts of data. This 'data deluge' puts a corresponding burden on computing infrastructure supporting these algorithms, which must store, communicate, and infer from this data. Thus, a key challenge is to develop techniques for reducing this burden by making efficient use of the information contained in the data. This talk looks at this from a theoretical viewpoint.

I will discuss work on developing the mathematical and engineering foundations for addressing these challenges, rooted in a technique called Kullback-Leibler divergence covering, and how this information theory perspective can be used to design algorithms and show rigorous bounds for various problems in communication, learning, and efficient information storage.

Bio: Jennifer Tang is a Postdoctoral Associate at MIT in the Institute of Data, Systems and Society (IDSS) and the Laboratory for Information and Decision Systems (LIDS), working with Professor Ali Jadbabaie. She received her Ph.D and S.M in EECS at MIT, advised by Yury Polyanskiy and a B.S.E in Electrical Engineering from Princeton University.  Jennifer won the 2022 ISIT Best Student Paper Award. She also has received the Irwin Mark Jacobs and Joan Klein Jacobs Presidential Fellowship at MIT and is the Shannon Centennial Celebration Student Competition Winner. Her research focuses on finding theoretical guarantees on problems in information and data science, at the intersection of information theory, applied probability, networks, and collective social phenomena.

===

Rahul Parhi

(Presenting 1:30 pm - 2:30 pm)

Title: Function-Space Models for Deep Learning

Abstract: Deep learning has been wildly successful in practice and most state-of-the-art artificial intelligence systems are based on neural networks. Lacking, however, is a rigorous mathematical theory that adequately explains the amazing performance of deep neural networks. In this talk, I present a new mathematical framework that provides the beginning of a deeper understanding of deep learning. This framework precisely characterizes the functional properties of trained neural networks. The key mathematical tools which support this framework include transform-domain sparse regularization, the Radon transform of computed tomography, and approximation theory. This framework explains the effect of weight decay regularization in neural network training, the importance of skip connections and low-rank weight matrices in network architectures, the role of sparsity in neural networks, and explains why neural networks can perform well in high-dimensional problems. At the end of the talk we shall conclude with a number of open problems and interesting research directions.

Bio: Rahul Parhi is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of California, San Diego. Prior to joining UCSD, he was a Postdoctoral Researcher at the École Polytechnique Fédérale de Lausanne (EPFL), where he worked from 2022 to 2024. He completed his PhD in Electrical Engineering at the University of Wisconsin-Madison in 2022, where he was supported by an NSF Graduate Research Fellowship. His research interests lie at the interface between functional analysis, signal processing, machine learning, and nonparametric statistics, with a particular emphasis on advancing the mathematical foundations of neural networks and deep learning.

===

Ilija Radosavovic

(Presenting 2:45 pm - 3:45 pm)

Title: Robotics as Sensorimotor Sequence Modeling

Abstract: Over the last decade the overarching paradigm of large language models has provided a unified framework for natural language processing (NLP). In contrast to NLP, we do not yet have a unified framework for robotics. In this talk, I will show that the paradigm of modern language models, when sufficiently generalized, can serve as a foundation for robotics. What is a good "language" for robotics? I will argue that one such choice are sensorimotor sequences. I will present results demonstrating the generality of this approach in both locomotion and manipulation settings. In particular, this enables humanoid locomotion over challenging terrain, manipulation from pixels, and offers appealing properties such as in-context adaptation.

Bio: Ilija Radosavovic is a PhD student in EECS at UC Berkeley, advised by Jitendra Malik. His research interests are in the areas of robotics, computer vision, and machine learning. Ilija is a recipient of the PAMI Mark Everingham Award (2021) and his work has been deployed across the industry and adopted by major corporations, including Facebook and Tesla.

===