Bio

Alicia is currently a Ph.D. student at the Department of Electrical Engineering and Computer Science at UC Berkeley, formally advised by Laurent El Ghaoui and affiliated with Berkeley Artificial Intelligence Research (BAIR) Lab. She received a B.B.A. degree in International Business from National Taiwan University and an M.S. degree in Information Management and Systems from UC Berkeley. Her primary research interests include gaining a better understanding of the theory, optimization, and behavior of AI systems, and the robustness and sparsity issues in ML algorithms. 

Alicia has served as the executive board member of Women in Machine Learning, the founding board member of Taiwan Data Science Association, and the founder of Women in Data Science (WiDS) Taipei. She has worked as a research intern at Google DeepMind, an ML engineer intern at Apple, an applied scientist intern at Amazon, and an investment associate at Cherubic Ventures.

Research Statement

Recently, a new and promising paradigm in deep learning has emerged, known as implicit models, which are different from the conventional feedforward structures found in complex deep learning models. These implicit models rely on an "equilibrium" equation to make predictions, rather than relying on recurrence through multiple layers. Unlike traditional deep learning models that are based on feedforward structures without loops, implicit models introduce the possibility of loops, which is a departure from the established neural network architecture. This departure from feedforward structures may hold the key to modeling complex higher-level reasoning, a challenge that conventional deep learning has struggled with.

However, this shift introduces a fundamental concern related to well-posedness, as equilibrium equations may have no unique solution or multiple solutions. My research delves into the theoretical aspects of implicit models, beginning with a unified "state-space" representation that simplifies notation. Additionally, my work explores various training challenges associated with implicit models, including problems amenable to convex optimization. These investigations reveal connections to critical topics such as architecture optimization, model compression, and robustness.

In addition to theoretical exploration, my research extends the potential applications of implicit models to various problem domains. This includes exploring their utility in tasks like parameter reduction, feature elimination, and mathematical reasoning. Implicit models hold promise as a new avenue for advancing deep learning capabilities.