I am a principal scientist at Apple, focusing on foundation language models. Previously, I worked at ByteDance as head of the applied machine learning (AML) research team. Even earlier, I worked as a research scientist at Google. Even even earlier, I received my PhD from Computer Science Department, Princeton University, working with Prof. David Blei. email: mr.chongwang@[companyname].com

Some Awards:

• SIGKDD 2021 Test of Time Award
• Google PhD fellow and Siebel Scholar
• Paper awards or honorable mentions at NIPS, AISTATS, KDD

Highlights:

• A simple and efficient speculative decoding algrithm for faster large language model inference,
  • Recurrent Drafter for Fast Speculative Decoding in Large Language Models, Source code with Pytorch and TensorRT LLM.
• More efficient softmax attentions with linear complexity,
  • Linear Complexity Randomized Self-attention Mechanism, ICML 2022
  • Efficient Attention via Control Variates, ICLR 2023
  • Source code is available on GitHub.
• A more efficient nonuniform negative sampling algorithm,
  • Nonuniform Negative Sampling and Log Odds Correction with Rare Events Data, NeurIPS 2021
• Learning large-scale end-to-end retrieval models without resorting to approximate nearest neighbor search,
  • Deep Retrieval: An End-to-End Learnable Structure Model for Large-Scale Recommendations, CIKM 2021.
  • Full Index Deep Retrieval: End-to-End User and Item Structures for Cold-start and Long-tail Item Recommendation, RecSys 2023.
• Neural networks meet structured Bayesian methods,
  • Fully Supervised Speaker Diarization provides better accuracy and efficiency in speaker diarization, ICASSP 2019.
    • Official Google AI blog.
    • Media reports: VentureBeat, SiliconANGLE, InfoQ, futurism, cnBeta, Sina Tech, iThome, ChinaEmail, eepw, QbitAI, oschina.
    • Open source on github with 1400+ stars!
• A generic approach for sequence modeling through segmentations,
  • Sequence Modeling via Segmentations, ICML 2017;
  • Towards Neural Phrase-based Machine Translation (code), ICLR 2018;
  • Subgoal Discovery for Hierarchical Dialogue Policy Learning, EMNLP 2018.
  • Neural Phrase-to-Phrase Machine Translation on Arxiv.
• Lookahead convolution architecture (ICLR 2016 workshop) enabled the deployment of an end-to-end speech recognition system (ICML 2016) to benefit hundreds of millions of users (http://maps.baidu.com) by significantly reducing the latency. See a torch implementation.
• Collaborative topic models (KDD 2011) are used by New York Times for their recommendation engine.

Publications:

• Find me at Google scholar and LinkedIn.

Recent News:

• 05/2024: Recurrent Drafter for Fast Speculative Decoding in Large Language Models on arxiv, 2024.
• 01/2023: Efficient Attention via Control Variates, ICLR 2023 (oral presentation).
• 05/2022: Differentially Private Multi-Party Data Release for Linear Regression, UAI 2022.
• 05/2022: PathFlow: A Normalizing Flow Generator that Finds Transition Paths, UAI 2022.
• 05/2022: Fourier Learning with Cyclical Data , ICML 2022.
• 05/2022: Linear Complexity Randomized Self-attention Mechanism, ICML 2022.
• 03/2022: Differentially Private Label Protection in Split Learning on arxiv, 2022.
• 01/2022: Label Leakage and Protection in Two-party Split Learning, ICLR, 2022.
• 11/2021: Learning Large-Time-Step Molecular Dynamics with Graph Neural Networks presented at AI for Science workshop in NeurIPS 2021.
• 09/2021: Nonuniform Negative Sampling and Log Odds Correction with Rare Events Data, NeurIPS 2021.
• 08/2021: Deep Retrieval: An End-to-End Learnable Structure Model for Large-Scale Recommendations, CIKM 2021.
• 05/2019: Rate Distortion For Model Compression: From Theory To Practice, ICML 2019.
• 02/2019: Fully Supervised Speaker Diarization, ICASSP 2019.
• 12/2018: Neural Logic Machines, ICLR 2019.
• 11/2018: Invited talk at University of Washington, Seattle.
• 08/2018: Subgoal Discovery for Hierarchical Dialogue Policy Learning, EMNLP (oral presentation), 2018.
• 06/2018: Invited talk at machine learning thoery workshop, Peking University, Beijing, China.
• 03/2018: Invited talk at Caltech, Pasadena, CA.
• 02/2018: Thoracic Disease Identification and Localization with Limited Supervision, CVPR, 2018.
• 01/2018: Towards Neural Phrase-based Machine Translation, ICLR, 2018. (code)
• 01/2018: A General Method for Robust Bayesian Modeling, Journal of Bayesian Analysis.
• 09/2017: Q-LDA: Uncovering Latent Patterns in Text-based Sequential Decision Processes, NIPS, 2017.

Professional Activities:

• Area chairs for major machine learning conferences, including NeurIPS, ICML, AISTATS, etc.
• Associate Editor for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI).