Yatong Bai
|
Doctor of Philosophy University of California, Berkeley Advisor: Somayeh Sojoudi Email: yatong_bai (at) berkeley.edu |
Biography
I am a fresh Ph.D. graduate from UC Berkeley's engineering department, advised by Professor Somayeh Sojoudi.
My research aims to make deep learning and generative AI more efficient, aligned, robust, and reliable. My interests span across generative models (particulaly audio/music), robust deep learning, (convex) optimization, reinforcement learning, and controls.
Specifically, my research areas so far include:
Diffusion Models (Audio/Music Generation).
- Accelerating diffusion-based text-to-audio generation with consistency distillation (ConsistencyTTA).
- Aligning text-to-music generation to human preferences by optimizing distributional rewards (DRAGON).
Robust Deep Learning.
- Analyzing the vulnerabilities of large language models (LLMs) when used in conversational search engines (RAGDOLL).
- Mixing classifiers to address the "accuracy-robustness trade-off" for image classifiers (MixedNUTS, Adaptive Smoothing).
Convex Neural Network Optimization.
Prior to joining Berkeley, I obtained Bachelor's degrees in computer engineering and mechanical engineering from Georgia Institute of Technology, where I researched with Professors Julien Meaud and Thomas Conte.
I have interned at Adobe Research, Microsoft, Scale AI, Honda Aircraft Company, and Tesla.
News
May 2025: I graduated with a Doctor of Philosophy in mechanical engineering from UC Berkeley with a cumulative GPA of 4.00 out of 4.00. My Berkeley email remains active.
May 2025: I filed my Ph.D. dissertation, titled "Efficient and Reliable Optimization for Deep Learning and Media Generation".
April 2025: New paper “DRAGON: Distributional Rewards Optimize Diffusion Generative Models” by Yatong Bai, Jonah Casebeer, Somayeh Sojoudi, and Nicholas J. Bryan. Project website with generated music examples here.
April 2025: I completed my Ph.D. dissertation defense talk (slides).
March 2025: I received the 2024-2025 Outstanding Graduate Student Instructor (OGSI) Award from UC Berkeley.
February 2025: I served as a reviewer and reviewed 6 papers for the ICML 2025 conference.
January 2025: I served as a reviewer and reviewed 1 paper for the SIMODS journal.
December 2024: I led the compilation of a knowledge summary about convex optimization based on Berkeley's EECS 127/227A lecture notes by Somayeh Sojoudi. It's available to everyone here.
December 2024: I served as a reviewer and reviewed 2 papers for the TMLR journal.
October 2024: I served as a reviewer and reviewed 3 papers for the ICLR 2025 conference.
September 2024: Our paper Ranking Manipulation for Conversational Search Engines has been accepted to Conference on Empirical Methods in Natural Language Processing (EMNLP 2024) with an Oral presentation format (top 10% accepted papers). During the November 12-16 conference in Miami, in addition to giving an oral presentation, we will present this poster.
August 2024: Our paper “MixedNUTS: Training-Free Accuracy-Robustness Balance via Nonlinearly Mixed Classifiers” has been accepted to Transactions on Machine Learning Research (TMLR).
July 2024: I served as a reviewer and reviewed 6 papers for the NeurIPS 2024 conference.
June 2024: New paper “Ranking Manipulation for Conversational Search Engines” by Samuel Pfrommer, Yatong Bai, Tanmay Gautam, and Somayeh Sojoudi. Project code on GitHub. This work proposes the "RAGDOLL" dataset, which is available on Huggingface Datasets.
June 2024: Our paper “ConsistencyTTA: Accelerating Diffusion-Based Text-to-Audio Generation with Consistency Distillation” has been accepted to the INTERSPEECH conference of 2024. Live demo available at Huggingface. Please try it out!
May 2024: I have received the Henry Lurie Family Fellowship.
March 2024: Our paper Improving the Accuracy-Robustness Trade-Off of Classifiers via Adaptive Smoothing has been accepted to the SIAM Journal on Mathematics of Data Science (SIMODS).
March 2024: I served as a reviewer and reviewed 4 papers for the ICML 2024 conference.
March 2024: Our paper Mixing Classifiers to Alleviate the Accuracy-Robustness Trade-Off has been accepted to the 6th Annual Learning for Dynamics & Control Conference.
February 2024: We have open-sourced the training and evaluation code of the paper “ConsistencyTTA: Accelerating Diffusion-Based Text-to-Audio Generation with Consistency Distillation” at github.com/Bai-YT/ConsistencyTTA. The model checkpoints are shared on Huggingface.
February 2024: New paper “MixedNUTS: Training-Free Accuracy-Robustness Balance via Nonlinearly Mixed Classifiers” by Yatong Bai, Mo Zhou, Vishal M. Patel, and Somayeh Sojoudi. Project code github.com/Bai-YT/MixedNUTS.
February 2024: I prepared summarization notes about Nonlinear Optimization for Professor Javad Lavaei's IEOR 160 course and made it publicly available on this website.
December 2023: I will join Adobe Research as a Research Intern in May 2024. I will be working with Nicholas J. Bryan.
October 2023: I served as a reviewer and reviewed 4 papers for the ICLR 2024 conference.
September 2023: New paper “ConsistencyTTA: Accelerating Diffusion-Based Text-to-Audio Generation with Consistency Distillation” by Yatong Bai, Trung Dang, Dung Tran, Kazuhito Koishida, and Somayeh Sojoudi. This work accelerates text-to-audio generation by up to 400x. Project website consistency-tta.github.io. Code is open-sourced here.
August 2023: I served as a Graduate Student Instructor (GSI) for IEOR 160: Nonlinear and Discrete Optimization in the Fall 2023 semester.
July 2023: Our paper Initial State Interventions for Deconfounded Imitation Learning has been accepted to the IEEE Conference on Decision and Control (CDC).
June 2023: I served as a reviewer and reviewed 6 papers for the NeurIPS 2023 conference.
April 2023: I served as a reviewer and reviewed 5 papers for the 2023 IEEE Conference on Decision and Control (CDC).
February 2023: I served as a reviewer for the ICML 2023 (4 papers) and IEEE CCTA 2023 (1 paper) conferences.
January 2023: New paper “Improving the Accuracy-Robustness Trade-Off of Classifiers via Local Adaptive Smoothing” by Yatong Bai, Brendon G. Anderson, Aerin Kim, and Somayeh Sojoudi.
January 2023: New paper “Let's Go Shopping (LGS) — Web-scale Image-text Dataset for Visual Concept Understanding” by Yatong Bai, Utsav Garg, Erhan Bas, Isidora Filipovic, Apaar Shanker, Haoming Zhang, Samyak Parajuli, Amelia N. Chu, Eugenia D Fomitcheva, Elliot Branson, Aerin Kim, Somayeh Sojoudi, and Kyunghyun Cho.
December 2022: I will join Microsoft Applied Science as a Research Intern in May 2023. I will be working with Kazuhito Koishida.
December 2022: I presented our poster on convex neural network training at the year-end TBSI Symposium.
October 2022: Our paper “Efficient Global Optimization of Two-layer ReLU Networks: Adversarial Training and Quadratic-time Algorithms” has been accepted to the SIAM Journal on Mathematics of Data Science (SIMODS).
August 2022: I am serving as a Graduate Student Instructor (GSI) for IEOR 160: Nonlinear and Discrete Optimization of the Fall 2022 semester.
May 2022: I have joined Scale AI as a Machine Learning Research Intern. I will be working on a bi-modal dataset project with Aerin Kim.
May 2022: I have passed the Qualifying Examination and advanced to candidacy as a Ph.D. candidate.
April 2022: I served as a reviewer and reviewed 1 paper for the 2022 IEEE Conference on Decision and Control (CDC).
January 2022: Our paper “Practical Convex Formulation of Robust Two-layer Neural Network Training” has been accepted to the American Control Conference of 2022.
January 2022: I am serving as a Graduate Student Instructor (GSI) for IEOR 160: Nonlinear and Discrete Optimization of the Spring 2022 semester.
December 2021: New paper “Efficient Global Optimization of Two-layer ReLU Networks: Adversarial Training and Quadratic-time Algorithms” by Yatong bai, Tanmay Gautam, and Somayeh Sojoudi.
October 2021: I gave a talk on convex neural network adversarial training at INFORMS 2021.
October 2021: The paper “Efficient Global Optimization of Two-layer ReLU Networks: Adversarial Training and Quadratic-time Algorithms” received the 2021 INFORMS Data Mining Best Student Paper Award Runner-Up (2nd out of 48 papers).
August 2021: I gave a talk on convex neural network adversarial training at MOPTA 2021.
August 2021: I have passed the Preliminary Examination.
April 2021: I received the ECE Senior Scholar Award from the Roger P. Webb Awards Program at Georgia Tech.
December 2020: New paper on convex neural network adversarial training: “Practical Convex Formulation of Robust Two-layer Neural Network Training” by Yatong Bai, Tanmay Gautam, Yu Gai, and Somayeh Sojoudi.
August 2020: I have joined UC Berkeley as a Ph.D. student.
August 2020: I have graduated from Georgia Institute of Technology with B. S. degrees in Computer Engineering and Mechanical Engineering with a cumulative GPA of 4.00 out of 4.00.