Yatong Bai
Scholar

Yatong Bai

Google Scholar ID: A9h086cAAAAJ
University of California, Berkeley
OptimizationDeep LearningAdversarial RobustnessAudio GenerationMusic Generation
Citations & Impact
All-time
Citations
144
 
H-index
6
 
i10-index
6
 
Publications
13
 
Co-authors
19
list available
Resume (English only)
Academic Achievements
  • Paper 'DRAGON: Distributional Rewards Optimize Diffusion Generative Models' accepted to Transactions on Machine Learning Research (TMLR) in October 2025.
  • Paper 'Ranking Manipulation for Conversational Search Engines' accepted to EMNLP 2024 with Oral presentation (top 10%).
  • Ph.D. dissertation titled 'Efficient and Reliable Optimization for Deep Learning and Media Generation' filed and defended in May 2025.
  • Developed key projects: ConsistencyTTA (accelerating diffusion-based text-to-audio), DRAGON (aligning text-to-music via distributional rewards), RAGDOLL (analyzing LLM vulnerabilities in conversational search), MixedNUTS and Adaptive Smoothing (addressing accuracy-robustness trade-off), and convex optimization algorithms for neural networks with global optimality guarantees.
  • Recipient of UC Berkeley’s Outstanding Graduate Student Instructor (OGSI) Award for 2024–2025.
  • Served as reviewer for ICML 2025 (6 papers), ICLR 2025 (3 papers), TMLR (2 papers), and SIMODS (1 paper).