I am a scientist at Amazon AGI Foundation, we recently launched Nova 2, Amazon’s unified multimodal foundation model.

I obtained a Ph.D. from CMU’s
Machine Learning Department and Neuroscience Institute 🧠, advised by Prof. Byron Yu, Prof. Matthew Smith, and Prof. Brent Doiron. My research focused on machine learning methods for accelerating scientific discovery and decision-making. I have also collaborated with Prof. Leila Wehbe and Prof. Aaditya Ramdas on neural data analysis, and with Prof. Woody Zhu on causal diffusion models.

Before CMU, I studied mathematics and statistics at Columbia University, working with Prof. Liam Paninski on a machine learning pipeline for neural signal processing. I also spent two years studying computing mathematics at City University of Hong Kong before entering the CityU–Columbia joint degree program.

Education

Ph.D. in Machine Learning and Neural Computation, Carnegie Mellon University, 2018–2023
Thesis committee: Byron Yu, Matthew Smith, Brent Doiron, Chengcheng Huang, Robert Kass, Tatiana Engel

B.A. in Mathematics–Statistics, Columbia University, 2015–2018
Summa Cum Laude · Phi Beta Kappa · Statistics Department Honor

B.S. in Computing Mathematics, City University of Hong Kong, 2013–2015
Mainland Student Full Tuition Scholarship · First Class Honor

Work Experience

Applied Scientist, Amazon AGI, 2025–Now
Working on Nova2 Omni, a unified multimodal foundation model.

Research Scientist, ByteDance Seed AI-for-Science, 2023–2025
Foundation model for protein structure prediction.

Research Scientist Intern, Applied Machine Learning Group, TikTok (ByteDance), 2022
Developed a reinforcement learning framework based on graph neural networks to accelerate molecular simulation.

Preprints

📄 Amazon Nova 2: Multimodal Reasoning and Generation Models.
Technical Report
Keywords: foundation model, unified multimodal model, image generation.

📄 Protenix — Advancing Structure Prediction Through a Comprehensive AlphaFold3 Reproduction.
Paper · Code
ByteDance AML AI4Science Team, Xinshi Chen, Yuxuan Zhang, Chan Lu, Wenzhi Ma, Jiaqi Guan, Chengyue Gong, Jincai Yang, Hanyu Zhang, Ke Zhang, Shenghao Wu, Kuangqi Zhou, Yanping Yang, Zhenyu Liu, Lan Wang, Bo Shi, Shaochen Shi, Wenzhi Xiao
Keywords: foundation model, diffusion transformer, AI4Science.

📄 YASS: Yet Another Spike Sorter applied to large-scale multi-electrode array recordings in primate retina.
Paper · Code
JinHyung Lee, Catalin Mitelut, Hooshmand Shokri, Ian Kinsella, Nishchal Dethe, Shenghao Wu, Kevin Li, Eduardo Blancas Reyes, Denis Turcu, Eleanor Batty, Young Joon Kim, Nora Brackbill, Alexandra Kling, Georges Goetz, E.J. Chichilnisky, David Carlson, Liam Paninski
Keywords: signal processing, convolutional neural networks, generative models, spike sorting.

Journal Publications

📄 Deconstructing The Ethics of Large Language Models From Long-standing Issues to New-emerging Dilemmas.
AI and Ethics
Paper
Chengyuan Deng, Yiqun Duan, Xin Jin, Heng Chang, Yijun Tian, Han Liu, Henry Peng Zou, Yiqiao Jin, Yijia Xiao, Yichen Wang, Shenghao Wu, Zongxing Xie, Kuofeng Gao, Sihong He, Jun Zhuang, Lu Cheng, Haohan Wang
Keywords: LLM, differential privacy, security.

📄 Automated customization of large-scale spiking network models to neuronal population activity.
Nature Computational Science
Paper · Code
Shenghao Wu, Chengcheng Huang, Adam Snyder, Matthew Smith, Brent Doiron, Byron Yu
Keywords: Bayesian optimization, dimensionality reduction, spatiotemporal data, spiking neural networks, brain-computer interface.

📄 Brainprints: identifying individuals from magnetoencephalograms.
Nature Communications Biology
Paper · Code
Shenghao Wu, Aaditya Ramdas, Leila Wehbe
Keywords: neuroimaging, data privacy, feature engineering, multi-modal recordings.

Conference Publications and Presentations

📄 Perceptual Inductive Bias Is What You Need Before Contrastive Learning.
CVPR 2025
Paper
Junru Zhao, Tianqin Li, Dunhan Jiang, Shenghao Wu, Alan Ramirez, Tai Sing Lee
Keywords: computer vision, biology inspired algorithm, contrastive learning, pretraining.

📄 Counterfactual Generative Models for Time-Varying Treatments.
KDD 2024 (talk); NeurIPS 2023 DGMH Workshop (spotlight)
Paper · Code
Shenghao Wu, Wenbin Zhou, Minshuo Chen, Shixiang Zhu
Keywords: diffusion models, causal inference, longitudinal data, counterfactual prediction.

📄 Interpreting area-to-area differences in spiking variability using spiking network models.
Society for Neuroscience, 2023
Shenghao Wu, Adam Snyder, Chengcheng Huang, Matthew Smith, Brent Doiron, Byron Yu
Keywords: deep neural emulators, mechanistic models, spiking neural networks.

📄 RLCG: When Reinforcement Learning Meets Coarse Graining.
NeurIPS 2022 AI4Science Workshop
Paper
Shenghao Wu, Tianyi Liu, Zhirui Wang, Wen Yan, Yingxiang Yang
Keywords: graph neural networks, reinforcement learning, molecular dynamics.

📄 Automatic fitting of spiking network models to neuronal activity reveals limits of model flexibility.
Cosyne 2020
Abstract
Shenghao Wu, Chengcheng Huang, Adam Snyder, Matthew Smith, Brent Doiron, Byron Yu

📄 Neural networks for sorting neurons.
Cosyne 2020
Abstract
JinHyung Lee, Catalin Mitelut, Ian Kinsella, Shenghao Wu, Eleanor Batty, Liam Paninski, Hooshmand Shokri, Ari Pakman, Yueqi Wang, Nishchal Dethe, Kevin Li, Eduardo Blancas Reyes, Alexandra Tikidji-Hamburyan, Georges Goetz, Ej Chichilnisky, David Carlson

📄 Riskalyzer: Inferring Individual Risk-Taking Propensity Using Phone Metadata.
ACM Ubicomp 2018
Paper
Vivek Singh, Rushil Goyal, Shenghao Wu

Services

PC Member:
NeurIPS (2024), KDD (2024, 2023), ICML/NeurIPS AI4Science Workshop (2024, 2023),
NeurIPS Generative AI & Biology Workshop (2023), Brain Informatics (2023), ACAIN (2021–2023).

Journal Reviewer:
Nature Communications Biology, Neurocomputing, Information Fusion.