Beyond Diffusion: Tensor-Train Decomposition for Exact AR Joints in High Dimensions

kalomaze · x · 2026-08-12

The author suggests that "next token prediction" can be generalized to "next joint prediction" without relying on diffusion, MSE regression, or flow matching. Backed by personal experiments, they propose that tensor-train style decomposition offers a tractable path for modeling exact autoregressive joints over high-dimensional data, effectively scaling to much larger action spaces than previously expected.

Related event: Tensor Train Decomposition Enables Exact High-Dimensional Modeling(4 posts)→

Original post →

More from Research

Research channel →