Zaid Harchaoui
University of Washington, WA·Computer Science
Machine learning methods now do two very different kinds of work: generating text and images at enormous scale, and making sense of signals recorded from the living brain. Both raise the same underlying question of why a complex model behaves the way it does, and whether that behavior can be explained mathematically rather than accepted as a black box.
- Fields
Computer Science
Mathematics & Statistics
Neuroscience
- Topics
generative AI
large language models
diffusion models
- Grants
- 2 known grants, through Aug 2028
What they work on
- Identify the mathematical quantities that drive generalization in frontier generative models, including large language models, using frameworks from probability theory, variational analysis, mathematical statistics and information theory.
- Develop new mathematical analyses of diffusion-based generative models, which build data by reversing a noise process, and design data strategies that support zero-shot inference where no training data exist.
- Discover scaling laws that let models reach compute-optimal tradeoffs between cost and accuracy during inference and generation, addressing the limits of brute-force scale.
- Learn biologically plausible, flexible functional connectivity models, which capture how brain regions act together, from electrocorticography recordings taken from the cortical surface.
- Build a deep graph convolutional network that links electrocorticography signals to network-scale connectivity, and use it as an explainable guide for prefrontal brain stimulation and for tracking how networks change over time.
More about the research
Two projects pursue this question from opposite ends, one deriving the mathematical principles behind large generative models, the other building models whose reasoning can be inspected and used to steer a medical treatment.
Why it matters
Understanding mathematically why large generative models generalize could reduce the reliance on sheer scale and make inference in new, data-free situations more dependable. Separately, explainable models of how stimulation reshapes brain networks could support AI-guided neurostimulation aimed at improving higher cognitive functions such as working memory, attention and executive control in patients with neurological and psychiatric disorders.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Alejandro Ribeiro
University of Pennsylvania, PA·Engineering
Develop optimization and statistical foundations for aligning generative models by exploiting that the problem is strongly concave in distribution space even though it is highly nonconvex in the parameters of deep networks and transformers.
Huibin Zhou
Yale University, CT·Mathematics & Statistics
Determine whether the iterative rectified flow algorithm, which turns high-dimensional density estimation into repeated regression, achieves optimal density and transport estimation in just one step of regression.
Yusu Wang
University of California-San Diego, CA·Computer Science
Explain why trained flow-based and diffusion generative models generalize better than the theoretically optimal solution predicts, using geometry, ODE analysis, manifold learning and deep learning theory, to enable controlled generation of novel samples.
The school
University of Washington: 81 labs on PI Pages, $1.7B a year on research, 694 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.