Live intelligence

Latest AI updates

2 updates
Research

Flow Matching from Non-Gaussian Sources: A Theory of Efficient Sampling

The paper develops a nonasymptotic convergence theory for Euler discretization of flow matching with linear interpolation when the source distribution is non-Gaussian. Under assumptions including an everywhere positive continuously differentiable density, finite total variance, an L-Lipschitz score, and a mild target second-moment condition, it bounds total variation error by epsilon in O-tilde(L d TotVar_base / epsilon) iterations. The assumptions cover multivariate Student's t sources with more than two degrees of freedom and finite Gaussian mixtures with shared covariance, while allowing target distributions without smooth densities or bounded support.

Flow matchingGenerative modelsSampling theory
Research

UniFlow: Unified Multi-Scale Flow Matching for Time Series Generation

UniFlow is a new framework for time series generation that integrates multi-scale representation learning with non-autoregressive generation via a shared-semantic discrete interface. It uses a multi-scale tokenizer with shared-index voting to consolidate temporal information into a single public-ID sequence, and geometry-aware categorical flow matching to generate these IDs efficiently. Experiments on four benchmarks show improved fidelity, cross-scale consistency, and faster inference than autoregressive and diffusion baselines.

time series generationflow matchingmulti-scale representation learning