معرفی
David Duvenaud is an Associate Professor at the University of Toronto, holding a Canada Research Chair in Generative Models and a Schwartz Reisman Chair in Technology and Society. He is cross-appointed to the Department of Computer Science and Department of Statistical Sciences. A Sloan Research Fellow and founding member of the Vector Institute, his work bridges deep probabilistic models, AI safety, and scientific computing.
- PhD in Machine Learning (University of Cambridge, 2014)
- Postdoc in Hyperparameter Optimization (Harvard University, 2016)
- Co-founded Invenia (energy forecasting company)
His research spans foundational Neural Ordinary Differential Equations (NeurIPS 2018 Best Paper) and Automatic Chemical Design (ACS Central Science 2018) to recent work on AGI governance (2025) and AI safety (2024). Key contributions include stochastic variational inference, implicit differentiation frameworks, and antisymmetrization layers for quantum Monte Carlo.
Recent publications (2024-2025) focus on systemic existential risks from AI, many-shot jailbreaking attacks, and epistemic uncertainty quantification. His group trains energy-based models with scalable MCMC samplers and develops invertible neural architectures (e.g., Residual Flows NeurIPS 2019). He also explores human-AI alignment through LLM Processes (NeurIPS 2024) and Sycophancy in Language Models (ICLR 2024).
- Canada Research Chair (2025)
- NSERC Grant (2025)
- Sloan Research Fellow (2021)
- Schwartz Reisman Chair (2021)
- Best Paper Award (NeurIPS 2018)
- Distinguished Paper Award (ICFP 2021)
His students include James Requeima, Jesse Bettencourt, and Raymond Douglas. He teaches courses on Statistical Methods for Machine Learning and Differentiable Inference. Current work (2025) investigates systemic human disempowerment through incremental AI capabilities and sabotage risk mitigation via hyperparameter-aware evaluations.
