Uploaded August 2026 | Updated November 2026, 5 days ago
Jiaxin Shi (Meta)
https://simons.berkeley.edu/talks/jiaxin-shi-meta-2026-08-06
Diffusion Generative Modeling: Progress and Next Steps
Modern generative AI is often viewed through two dominant paradigms: autoregressive models, which underpin much of language modeling, and diffusion models, which have set the standard for high-quality image and video synthesis. These paradigms are usually treated as fundamentally different: autoregressive models are associated with compression and sequential computation, while diffusion models are associated with perceptual sample quality and parallel generation.
In this talk, I will argue that this divide is less fundamental than it may appear. I will use masked diffusion models (MDMs) as a lens: this class of discrete diffusion models is theoretically equivalent to random-order autoregressive models. From this perspective, many empirical differences between autoregression and diffusion, such as parallelism and perceptual quality, arise not from a fundamental distinction between the model classes themselves, but from details of model specification and training techniques. For example, I will present evidence that the loss reweighting technique from diffusion, when applied to any-order MDM/AR models, can yield perceptual sample quality comparable to strong continuous diffusion models.
Next, I will talk about moving beyond both (fixed-order) autoregression and (random-order) diffusion to find new powerful ways of generation. Specifically, I will focus on models that adaptively determine their generation order based on partial context. I will highlight the important role of insertion in flexible generation and planning, and a principled framework we established for scalable maximum likelihood learning of insertion-based generative models.
Jiaxin Shi (Meta)
https://simons.berkeley.edu/talks/jiaxin-shi-meta-2026-08-06
Diffusion Generative Modeling: Progress and Next Steps
Modern generative AI is often viewed through two dominant paradigms: autoregressive models, which underpin much of language modeling, and diffusion models, which have set the standard for high-quality image and video synthesis. These paradigms are usually treated as fundamentally different: autoregressive models are associated with compression and sequential computation, while diffusion models are associated with perceptual sample quality and parallel generation.
In this talk, I will argue that this divide is less fundamental than it may appear. I will use masked diffusion models (MDMs) as a lens: this class of discrete diffusion models is theoretically equivalent to random-order autoregressive models. From this perspective, many empirical differences between autoregression and diffusion, such as parallelism and perceptual quality, arise not from a fundamental distinction between the model classes themselves, but from details of model specification and training techniques. For example, I will present evidence that the loss reweighting technique from diffusion, when applied to any-order MDM/AR models, can yield perceptual sample quality comparable to strong continuous diffusion models.
Next, I will talk about moving beyond both (fixed-order) autoregression and (random-order) diffusion to find new powerful ways of generation. Specifically, I will focus on models that adaptively determine their generation order based on partial context. I will highlight the important role of insertion in flexible generation and planning, and a principled framework we established for scalable maximum likelihood learning of insertion-based generative models.

![Ground Energy estimation of Quantum Impurity model is in BQP
Jiaqing Jiang (UC Berkeley)
https://simons.berkeley.edu/talks/jiaqing-jiang-uc-berkeley-2026-07-24
Quantum Summer Cluster Final Workshop
Estimating the ground-state energy of quantum many-body systems is a leading application of quantum computing. For general Hamiltonians, however, the problem is QMA-complete and is therefore unlikely to admit an efficient quantum algorithm. This motivates the search for natural classes of Hamiltonians whose ground-energy problem lies in BQP but is not known to lie in BPP.
In this talk, we identify quantum impurity Hamiltonians as one such class. These Hamiltonians are central to dynamical mean-field theory (DMFT), widely regarded as a gold-standard framework for studying strongly interacting quantum systems. Specifically, by explicitly constructing a guiding state, we give a polynomial-time quantum algorithm that estimates the ground-state energy of quantum impurity Hamiltonians to any inverse polynomial precision. Previously, the best unconditional upper bounds were a quasi-polynomial-time classical algorithm and containment of the corresponding decision problem in QCMA [Bravyi&Gosset arxiv:1609.00735]. Ground Energy estimation of Quantum Impurity model is in BQP](https://i.ytimg.com/vi/krqOy27VIu0/mqdefault.jpg)








