arXiv:2606.00684v3 Announce Type: replace-cross Abstract: We address the problem of out-of-distribution (OOD) detection for target observations embedded in a subspace of the high dimensional data spa...
arXiv:2602. 00797v2 Announce Type: replace-cross Abstract: Flow-based methods have achieved significant success in various generative modeling tasks, capturing nuanced details within complex data distributions.
The paper introduces entropy-ordered flows (EOFlows), a normalizing flow framework that adds an orthogonality regularizer on the decoder Jacobian to promote geometric disentanglement. EOFlows produce near‑orthogonal curvilinear coordinates whose features can be ranked by explained manifold entropy, providing a nonlinear analogue to PCA. The method yields an order‑of‑magnitude more stable features than existing approaches, categorizing them into global, local, and generic types, and can be trimmed to form an autoencoder with competitive rate‑distortion performance.
arXiv:2607. 14306v1 Announce Type: new Abstract: In this paper, we study the connection between an LLM's output distribution and the data used to train it.
arXiv:2512. 21815v4 Announce Type: replace-cross Abstract: Vision-language models (VLMs) achieve remarkable performance but remain vulnerable to adversarial attacks.
The paper introduces Contrastive Noise Alignment (CNA), a training-time method for generative flow models that dynamically aligns Gaussian noise with data samples using a cross-modal InfoNCE objective. By modeling noise as an interacting particle system and regularizing with angular entropy and radial norm penalties, CNA reduces arbitrary data-noise couplings and flow curvature. Empirical results show that CNA improves generation quality, lowering FID by over 50% for few-step pixel-space generation compared to standard rectified flow and outperforming optimal transport baselines by at least 24%.
The paper investigates how the order in which components are generated in GFlowNets affects learning and convergence. It shows that existing GFlowNet algorithms often fail to discover useful generation orders, and introduces a new training recipe—Learning-Order GFlowNet—that learns interpretable orders without explicit annotations. Experiments demonstrate that the new method improves convergence, and analysis of gradient signal-to-noise ratio suggests that higher SNR correlates with faster convergence and better generation orders.
arXiv:2605. 28021v2 Announce Type: replace Abstract: Out-of-distribution (OOD) detection is essential for deploying machine learning models in open-world and safety-critical scenarios, where test inputs may deviate from the training distribution and overconfident predictions on unknown samples can lead to unreliable decisions.
The paper introduces Priming Models, a two‑stage framework for training one‑step generative models. First, the model is trained on a reconstruction objective such as pixel‑space denoising at a fixed noise level, then it undergoes feature‑space distributional post‑training. This simple reconstruction stage alone is shown to eliminate the difficulty of distributional training, leading to significant improvements in training compute versus generation quality across tasks like class‑conditional image generation, text‑to‑image generation, and world modeling, with a state‑of‑the‑art FID of 1.19 on ImageNet 256×256 using about one‑tenth the compute of previous methods.
arXiv:2605. 00941v4 Announce Type: replace Abstract: Flow matching has become a leading framework for generative modeling, but quantifying the uncertainty of its samples remains an open problem.
PALATE is a new framework that improves the evaluation of deep generative models by applying the law of total expectation to accessible data. It builds on the Feature Likelihood Divergence (FLD) method but overcomes FLD’s computational limits, integrating with the MMD baseline metric and the DINOv3 feature extractor. Experiments show PALATE matches or surpasses state‑of‑the‑art methods while being more efficient and scalable, especially for detecting memorization and assessing generalization.
arXiv:2606. 07271v1 Announce Type: cross Abstract: Understanding what generative models retain from training data remains challenging, with implications for copyright and privacy.