Normalizing Flows
PulseAugur coverage of Normalizing Flows — every cluster mentioning Normalizing Flows across labs, papers, and developer communities, ranked by signal.
8 day(s) with sentiment data
-
LLMs and deep generative models advance PDE solving capabilities
Researchers have developed new methods for solving complex partial differential equations (PDEs) using large language models (LLMs) and deep generative models. One approach, OpInf-LLM, uses operator inference with LLMs …
-
New Rènyi Regularization Tackles Mode Collapse in Bayesian Inference
Researchers have developed a new method called Rènyi-Regularized Flow-Based Variational Inference to address mode collapse in Bayesian model calibration. This technique aims to improve posterior approximation by introdu…
-
New methods improve Bayesian inference for agent-based models
Researchers have developed new methods for Bayesian inference in agent-based models (ABMs), which are commonly used in economics and social sciences. The first paper introduces a framework called sequential neural poste…
-
Research: Counterintuitive Likelihood Behavior Rare in Tabular Anomaly Detection
A new research paper explores why counterintuitive likelihood behavior, where anomalous data receives higher likelihood scores, is rare in tabular anomaly detection compared to image data. The study introduces a domain-…
-
New FSKA framework learns nonstationary attention kernels for efficient ML
Researchers have introduced Flow-based Spectral Kernel Attention (FSKA), a novel framework designed to learn nonstationary attention kernels for improved self-attention mechanisms in machine learning. FSKA utilizes spec…
-
New Method Quantifies Uncertainty in Density-Based Clustering
Researchers have developed a new method for quantifying uncertainty in density-based clustering by integrating martingale posterior distributions with modern differentiable density estimators like normalizing flows. Thi…
-
Normalizing flows prove effective for Blume-Capel model phase diagram analysis · arXiv paper
Researchers have utilized normalizing flows, a type of generative model, to effectively sample and analyze the phase diagrams of a continuous spin variant of the Blume-Capel model. This study, published on arXiv, demons…
-
New NFDiag method enhances MCMC convergence diagnosis in high dimensions
Researchers have introduced NFDiag, a novel method for diagnosing the convergence of Markov chain Monte Carlo (MCMC) simulations, particularly in high-dimensional scenarios. Unlike traditional diagnostics that focus on …
-
New research tackles variational inference challenges in AI
Two new research papers address challenges in variational inference, a method used for approximating complex probability distributions. The first paper, "Mode Collapse of Mean-Field Variational Inference," provides a th…
-
OrthoGen: New Generative Orthogonal Learner for Time-Varying Treatments
Researchers have developed OrthoGen, a novel generative orthogonal learner designed to estimate conditional distributional potential outcomes (CDPOs) in scenarios with time-varying treatments. This method addresses chal…
-
New perspective on likelihood approximation for complex simulation models
Researchers have introduced a novel perspective on likelihood approximation for complex simulation models, utilizing a chain of aggregated normalizing flows. This approach aims to facilitate scalable data analysis, effi…
-
New method uses normalizing flows for SU(N) lattice gauge theory sampling
Researchers have developed a novel method for sampling SU(N) gauge theory on a 2D lattice using normalizing flows. This approach addresses the challenge of mapping sampled plaquettes to link variables in lattice gauge t…
-
Diffusion models advance with new architectures and training methods · 10 sources tracked
Researchers are exploring advanced diffusion models across various domains, including tri-modal generation for text, images, and audio by Apple, and data-free training methods for lattice field theories using diffusion …
-
New framework enhances low-resolution image representation using normalizing flows
Researchers have developed LR2Flow, a novel framework that enhances low-resolution image representation by combining wavelet tight frames with normalizing flows. This approach aims to preserve essential visual content w…
-
CDFlow introduces efficient invertible layers for generative models
Researchers have developed CDFlow, a novel approach to building invertible layers for deep generative models using circulant and diagonal matrices. This method reduces parameter complexity and computational cost for mat…
-
Normalizing Flows Used to Recover Weak Scientific Signals
Researchers have developed a novel method using normalizing flow models to reconstruct weak signals that are obscured by stronger nuisance signals in scientific data. This technique addresses the inherent distortion tha…
-
New MAGT method offers direct generative transport for high-dimensional data
Researchers have introduced Manifold-Aligned Generative Transport (MAGT), a novel method for generating high-dimensional data that concentrates near a low-dimensional structure. Unlike diffusion models that require iter…
-
New statistical method uses normalizing flows for likelihood-free inference
A new statistical method has been developed for likelihood-free inference, particularly useful when dealing with nuisance parameters. This approach utilizes a neural-network-based normalizing flow to identify a pivotal …
-
New framework models real-world image noise using normalizing flows
Researchers have developed a new normalizing flows (NF) framework to model real-world image noise more effectively. Unlike previous methods that rely on camera metadata even during the sampling phase, this new framework…
-
Apple unveils STARFlow2 for unified text-image generation
Apple researchers have developed STARFlow2, a novel architecture that unifies multimodal generation by bridging language models and normalizing flows. This approach allows for continuous, single-pass, and purely causal …