Pith. sign in

REVIEW 6 cited by

Heavy-Tailed Diffusion Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.14171 v2 pith:OBBLZXPS submitted 2024-10-18 cs.LG stat.ML

classification cs.LGstat.ML
keywords diffusionmodelsheavy-taileddistributionsframeworkstandardstudent-tcapture
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

Diffusion models achieve state-of-the-art generation quality across many applications, but their ability to capture rare or extreme events in heavy-tailed distributions remains unclear. In this work, we show that traditional diffusion and flow-matching models with standard Gaussian priors fail to capture heavy-tailed behavior. We address this by repurposing the diffusion framework for heavy-tail estimation using multivariate Student-t distributions. We develop a tailored perturbation kernel and derive the denoising posterior based on the conditional Student-t distribution for the backward process. Inspired by $\gamma$-divergence for heavy-tailed distributions, we derive a training objective for heavy-tailed denoisers. The resulting framework introduces controllable tail generation using only a single scalar hyperparameter, making it easily tunable for diverse real-world distributions. As specific instantiations of our framework, we introduce t-EDM and t-Flow, extensions of existing diffusion and flow models that employ a Student-t prior. Remarkably, our approach is readily compatible with standard Gaussian diffusion models and requires only minimal code changes. Empirically, we show that our t-EDM and t-Flow outperform standard diffusion models in heavy-tail estimation on high-resolution weather datasets in which generating rare and extreme events is crucial.

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. An Extreme Value Perspective on Learning Stress Laws

    q-fin.RM 2026-07 conditional novelty 6.5 of 10

    SS-GEN splices an explicit radial tail law with a DGM-learned angular law so standard generative models produce asymptotically exact multivariate extremes and rare-event probabilities beyond the data.

  2. Denoising Subordinated Probabilistic Models: Diffusion with a Tempered-Stable Volatility Clock, and What the Noise Mechanism Actually Controls

    q-fin.MF 2026-07 conditional novelty 6.0 of 10

    Heavy-tailed noise in diffusion can be exactly calibrated via a tempered-stable volatility chain, but is provably a nuisance whenever the denoiser sees it and the chain is drawn independently of the data.

  3. Diversify Diffusion with Temperature Sampling and Variance-Corrective Time Shifting

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Variance-corrective time shifting cancels the variance inflation of naive temperature sampling so pretrained diffusion models can raise diversity without retraining.

  4. Heavy-Tailed Flow Matching via Random Clocks

    cs.LG 2026-07 conditional novelty 5.0 of 10

    HTFM trains flow-matching models with heavy-tailed source distributions obtained by mixing Gaussian sources over a random clock path, improving tail-statistic recovery and low-NFE sampling.

  5. Offline Reinforcement Learning with Penalized Action Noise Injection

    cs.LG 2025-07 conditional novelty 5.0 of 10

    Injecting noise-perturbed actions into offline Q-learning with a distance penalty improves D4RL performance over IQL and TD3 baselines, formalized as Q-learning in a Noisy Action MDP.

  6. Perspectives on Tsallis Statistics for Artificial Intelligence

    cs.AI 2026-08 conditional novelty 3.0 of 10

    Independent AI methods, including sparsemax attention, Tsallis-entropy reinforcement learning, Student-t generative models, and robust losses, are instances of a single 'q-dial' deformation of Boltzmann-Gibbs statisti...

Pith tools