#generative-models

36 curated events

papersTODAY 04:00 UTC

Backward SDE Approach Targets Physics Consistency in Diffusion Models

A new arXiv preprint proposes using backward stochastic differential equations to enforce physics and measurement consistency when score-based diffusion models are applied to inverse problems. The authors argue that existing methods, such as heuristic guidance, periodic projections, or task-specific conditional training, are less principled. The work positions the approach as a general way to keep pretrained diffusion priors aligned with physical constraints.

diffusion/research/physics/generative-models

papersTODAY 04:00 UTC

DepthBenchCAD Examines Whether More Auditing Checks Improve Generative CAD Evaluations

A new arXiv preprint introduces DepthBenchCAD, a benchmark studying how the number of edit checks affects the reliability of evaluations for generative CAD models. The work focuses on behavioral correctness after parameter edits and asks whether auditing more programs under a fixed budget actually leads to firmer conclusions. It questions the common assumption that adding edit checks is a straightforward path to more trustworthy evaluation.

benchmark/cad/evaluation/generative-models

papersTODAY 04:00 UTC

Paper proposes lexicographic preference steering for pretrained robot policies

A new arXiv preprint introduces a method for adjusting pretrained generative robot policies at deployment time using lexicographic preferences. This lets operators express prioritized requirements that were not captured during training, instead of retraining or fine-tuning the policy. The approach targets situations where deployed robots face new constraints or user preferences.

robotics/preferences/generative-models/arxiv

papersTODAY 04:00 UTC

Study finds video models store correct physics but fail to apply it

A new arXiv paper asks whether video generators that produce physically implausible motion never learned the correct dynamics or merely fail to use what they learned. The authors introduce a notion of causal writability to probe this, and report that the correct motion is still represented inside the model and can be made to steer generation. The finding points to a gap between internal knowledge and how it is used, rather than a simple absence of physical understanding.

video-models/physics/interpretability/generative-models

papersTODAY 04:00 UTC

Ensemble-Conditioned Molecular Design Accounts for Conformer Distributions

A new arXiv preprint argues that molecular design should not be reduced to finding candidates that lock into one bioactive shape, since real molecules exist across a range of conformations. The authors propose an ensemble-conditioned method that designs molecules against this distribution of shapes rather than a single structure, aiming to better capture the properties that determine whether a candidate succeeds.

molecular-design/arxiv/generative-models/chemistry

papersTODAY 04:00 UTC

Energy-Based Generative Model Proposed for Signal Unmixing and Curve Resolution

A new arXiv preprint introduces EB-gMCR, an energy-based generative approach for separating the individual component signals contained in a single mixed measurement, such as a chemical reaction mixture or tissue sample. The method aims to recover both each component's profile and its concentration from data where only the weighted sum is observed. It is positioned as a tool for signal unmixing and multivariate curve resolution tasks in chemistry and related fields.

research/generative-models/signal-processing/chemistry

papersTODAY 04:00 UTC

Variational Incompressible Optimal Transport Operator for Flow-Based Generation

Researchers introduce the Variational Incompressible Optimal Transport (VIOT) operator, a generative neural operator designed to transport densities in an amortized way. Given a new pair of source and target densities, VIOT outputs a divergence-free velocity field to carry out the transformation. The approach targets incompressible flow settings, where the velocity field must remain divergence-free.

optimal-transport/generative-models/neural-operators/fluid-dynamics

papersTODAY 04:00 UTC

arXiv paper proposes training paradigm for fast long video generation

A new arXiv preprint addresses the difficulty of generating coherent minute-long videos, noting that while short clips are plentiful and high quality, long-form training data is scarce and confined to a few domains. The authors propose a training approach that combines mode-seeking and mean-seeking objectives to speed up long video generation. The work is positioned as a way to overcome the data bottleneck that limits scaling from seconds to minutes.

video-generation/research/training/generative-models

papersSEP 11 04:00 UTC

Paper links continuous and discrete flow matching through argmax projection

A new arXiv preprint argues that continuous and discrete flow matching, typically developed as separate frameworks, are dual formulations of the same underlying construction. The authors show that applying a position-wise argmax to continuous convex-interpolant flows with one-hot targets recovers the discrete counterpart, and they examine how the choice of source distribution shapes categorical generation. A revised version of the submission has replaced the initial preprint.

flow-matching/generative-models/discrete-diffusion/research

papersTODAY 04:00 UTC

Autoregressive Fragment Model Generates Molecules at Specified Attachment Sites

A new arXiv preprint introduces Fraglingo, a generative approach that builds molecules by mimicking common medicinal-chemistry edits such as extending a scaffold or swapping a substituent. The method generates fragments conditioned on a chosen attachment point, letting users direct where a new piece is added while optimizing desired molecular properties. It falls under fragment-based molecular design research.

molecular-design/generative-models/chemistry/research

papersTODAY 04:00 UTC

Cross-Block Conditioning in Deep Boltzmann Machines for Statistical Data Fusion

The paper introduces cross-block conditioning as a way to train deep Boltzmann machines on statistically fused data, where two panels share covariates but record different outcome blocks. Traditional data fusion setups never observe both outcomes for the same row, which prevents using a standard discriminative objective. The authors propose an approach that works around this missing-row problem, and the work appears as a cross-listing in arXiv's AI and machine learning categories.

deep-boltzmann-machines/data-fusion/arxiv/generative-models

papersTODAY 04:00 UTC

IsingFormer: Learned Proposals Augment Parallel Tempering for MCMC

A new arXiv paper proposes adding a global proposal move to parallel tempering, in which a learned model suggests state changes across finite temperatures. The aim is to improve mixing in Markov chain Monte Carlo sampling and optimization, where generative models have shown promise but remain difficult to integrate with standard MCMC. The authors present this as a way to combine learned proposals with established tempering methods rather than replace them.

mcmc/sampling/generative-models/research

papersTODAY 04:00 UTC

Convergence rate analysis of generative drifting flows

A new arXiv paper examines whether drifting models, which learn a gradual transport process during training but generate samples in a single step, can converge quickly to a target distribution. The authors identify obstructions to fast convergence at fixed scale and propose a multihead approach that improves convergence rates. The work is theoretical, focused on the training dynamics rather than a deployed system.

generative-models/training/research/theory

papersTODAY 04:00 UTC

AbFlow: Paratope-Centric Antibody Design via Interaction-Enhanced Flow Matching

A preprint introduces AbFlow, a generative approach that models full-atom antibody structures end-to-end while focusing on the paratope, the region that contacts the antigen. The method uses flow matching augmented with interaction information to guide design. The authors position it as addressing gaps in existing antibody design pipelines, which have lacked a unified generative framework at this structural resolution.

protein-design/flow-matching/generative-models/research

papersTODAY 04:00 UTC

ProteinZero Uses Online Reinforcement Learning for Self-Improving Protein Design

A new arXiv preprint introduces ProteinZero, a method that applies online reinforcement learning to protein generative models so they can improve without depending on curated sequence-structure datasets. The authors argue that current supervised training objectives are misaligned with actual protein design goals, and that their approach addresses this gap. The work appears as a replacement submission on arXiv's machine learning category.

protein-design/reinforcement-learning/generative-models/arxiv

papersTODAY 04:00 UTC

Reinforcement learning approach generates crystals with target symmetry and properties

A new arXiv preprint presents a reinforcement learning method for inverse design of crystalline materials that treats crystal symmetry as a constraint alongside desired physical properties. The authors argue that a good numerical property value is only meaningful when the structure's underlying symmetry is appropriate, so their approach generates candidates that satisfy both. The work targets functional materials discovery where symmetry-aware generation matters.

materials-science/reinforcement-learning/generative-models/arxiv

papersTODAY 04:00 UTC

Conditional Quantum Flow Matching Proposed for Physiological Signal Augmentation

Researchers propose a conditional quantum flow matching approach for generating synthetic physiological signals when labeled data is scarce. Unlike earlier quantum generative models that begin from uninformative noise, the method incorporates class structure already present in the data. The work targets label-scarce physiological signal classification tasks.

quantum/generative-models/data-augmentation/research

papersTODAY 04:00 UTC

Cooperative EBM-AE Framework Combines Energy Refinement and Manifold Projection

A new arXiv paper proposes pairing an energy-based model with an autoencoder in a cooperative training setup. The method alternates between refining the learned energy landscape and projecting samples back onto a data manifold, aiming to overcome common training difficulties in energy-based generative models. The authors report the framework assigns low energy to realistic samples and higher energy to unlikely ones.

research/generative-models/energy-based-models/arxiv

papersTODAY 04:00 UTC

Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling

A new arXiv paper introduces Chamaileon, a method for designing protein binders that works across different target contexts. The approach combines contextualized modeling with a mixed sampling strategy, building on recent generative techniques that jointly model protein sequence and structure. The authors position it as an alternative to existing end-to-end generation and hallucination-based pipelines, which they describe as limited in cross-context settings.

protein-design/generative-models/arxiv/structural-biology

papersTODAY 04:00 UTC

Bidirectional Neural Method Learns Quadratic Optimal Transport Maps from Unpaired Data

Researchers propose CyclOT, a bidirectional neural framework that recovers both forward and reverse quadratic optimal transport maps from unpaired samples in high dimensions. The learned maps induce forward and backward displacement interpolants that are synchronized during training. The work targets optimal transport estimation, a task that becomes difficult as dimensionality grows.

optimal-transport/research/generative-models/arxiv

papersSEP 11 04:00 UTC

Partial GFlowNet Method Aims to Speed Convergence in Large State Spaces

A revised arXiv paper proposes Partial GFlowNet, an approach that divides large state spaces into partitions to improve training convergence. The authors argue that standard GFlowNets, which explore state spaces freely, struggle to converge as those spaces grow. The work targets generative flow networks that sample candidates in proportion to their rewards.

gflownet/research/arxiv/generative-models

papersSEP 11 04:00 UTC

Fixed-Dimensional Latent Flow Model Generates Variable-Size 3D Molecules

Researchers present a transformer-based autoencoder that maps 3D molecules of differing sizes into a single fixed-dimensional latent space, removing the need to fix molecule size in advance. This matters because molecular size is tied to composition, structure and other target properties in drug and materials discovery. The approach is described as equivariant-free, meaning it avoids the geometric constraints commonly used in 3D molecular generators.

molecules/generative-models/3d-generation/research

papersSEP 10 04:00 UTC

New arXiv paper proposes tail-likelihood approach to reinforcement learning

A research paper on arXiv introduces a reinforcement learning method that goes beyond optimizing average reward. The authors argue that mean-based objectives can obscure differences between generative policies with equal averages but different probabilities of producing rare, high-reward outputs. The work focuses on optimizing the likelihood of these tail outcomes instead.

reinforcement-learning/research/generative-models

papersSEP 10 04:00 UTC

SurF: A Generative Model for Multivariate Irregular Time Series Forecasting

Researchers introduce SurF, a generative model designed for multivariate event streams that are sampled at irregular intervals. The work argues that tokenization-based approaches struggle when the gaps between events span orders of magnitude, and proposes an alternative suited to such data. The paper is a revised arXiv submission in machine learning.

time-series/generative-models/arxiv/forecasting

papersSEP 10 04:00 UTC

Jump-Diffusion Framework Introduced for Generating Irregularly Sampled Time Series

A research paper presents a method for training generative models on continuous-time data that is recorded unevenly and out of sync across sources. The approach builds on generator matching and can represent trajectories with sudden jumps rather than only smooth paths, backed by closed-form expressions for diffusion components. It could be useful in domains where measurements arrive at irregular intervals, such as healthcare monitoring or sensor networks.

time-series/generative-models/diffusion/research

papersSEP 10 04:00 UTC

Paper introduces latent bridge matching for albedo estimation in intrinsic image decomposition

A new arXiv preprint presents a latent bridge matching technique for estimating albedo, the reflectance component separated from lighting in intrinsic image decomposition. The authors note that generative approaches to this task have struggled with weak physical plausibility and computationally heavy inference, and their method is designed to tackle both shortcomings. The work is cross-listed under arXiv's artificial intelligence category.

computer-vision/generative-models/image-decomposition

papersSEP 10 04:00 UTC

Researchers propose manifold-aligned generative transport for low-dimensional data structures

A machine learning preprint on arXiv introduces a generative transport method aimed at datasets that concentrate near a low-dimensional structure embedded in a high-dimensional space. The approach seeks to limit probability mass leaking away from the data-supporting manifold while staying computationally practical, in contrast to the iterative sampling used by diffusion models. The paper was updated as a v2 cross-list replacement.

generative-models/diffusion/manifolds/research

papersSEP 10 04:00 UTC

uFlowCSP: Mean Flow Generative Models for Crystal Structure Prediction

Researchers introduce uFlowCSP, a generative approach to crystal structure prediction built on mean flow models. The method addresses a key drawback of diffusion- and flow-matching-based systems, which typically require many iterative sampling steps during inference. The paper situates the work next to existing generative CSP models such as CDVAE, DiffCSP, FlowMM, and CrystalFlow.

materials-science/generative-models/flow-matching/chemistry

papersSEP 10 04:00 UTC

GLOSS: Geometric Local Self-Similarity Learning for Reference-Guided 3D Texture Generation

A new arXiv paper introduces GLOSS, a method for generating complete textures on 3D objects guided by a single reference view. The authors observe that current conditional image generators struggle to cover a full object surface while staying faithful to the reference look. GLOSS addresses this by exploiting local geometric self-similarity across the shape to propagate texture detail consistently.

3d-generation/texturing/generative-models/arxiv

papersSEP 10 04:00 UTC

New Paper Addresses Mode Coverage Gaps in Normalizing Flow Boltzmann Generators

A new arXiv study examines why normalizing flow Boltzmann generators trained with forward KL divergence can miss portions of a target distribution when the available training samples are biased or incomplete. The authors propose using variation in the log-ratio of importance weights as a signal to detect when a flow's samples fail to cover target modes. The approach aims to improve the reliability of flow-based samplers used in statistical physics and molecular simulation.

generative-models/sampling/physics/machine-learning

papersSEP 12 04:00 UTC

Logit Refiner Targets Intra-Scale Dependencies in Visual Autoregressive Models

A new arXiv paper analyzes visual autoregressive models, which generate images by predicting one scale of tokens at a time and emitting all tokens in a scale in parallel. The authors argue this parallel decoding acts as a mean-field-style approximation that drops spatial dependencies within each scale. They propose a Logit Refiner method that models these intra-scale relationships to improve generation quality.

image-generation/autoregressive/research/generative-models

papersSEP 12 04:00 UTC

Model-Aware Diffusion Schedules Derived via Optimal Transport

A new arXiv paper argues that the schedules controlling how signal and noise are mixed along diffusion and flow-matching paths can be optimized by minimizing a kinetic action borrowed from optimal transport theory. The authors show that making these schedules depend on the specific model, rather than using fixed hand-tuned coefficients, improves generation quality. The work offers a theoretical framing for why certain noise schedules perform better than others.

diffusion/optimal-transport/generative-models/arxiv

papersSEP 12 04:00 UTC

ScaleResfusion combines residual rectified flow with residual vector fields for image restoration

A new arXiv paper introduces ScaleResfusion, a method that uses a residual vector field to guide residual rectified flow for real-world image restoration. The approach targets recovering high-quality images from complex, unknown degradations, a setting where diffusion-based methods have improved perceptual quality but still face remaining obstacles. The abstract excerpt available does not detail the full experimental results.

image-restoration/diffusion/research/generative-models

papersSEP 12 04:00 UTC

EgoGenEval benchmark targets physical consistency of image generators under ego-motion

A new arXiv paper introduces EgoGenEval, a benchmark aimed at measuring how well visual generators maintain physical consistency when a viewpoint moves, rather than judging output on image quality alone. The authors argue that current generators can produce realistic-looking images yet break physics under ego-motion, which limits their usefulness for spatial reasoning and embodied planning. Existing benchmarks, they note, mostly assess single images or single-step quality.

benchmark/evaluation/embodied-ai/generative-models

papersSEP 12 04:00 UTC

Information-Theoretic Framework Unifies Generalization Bounds for VAEs and Diffusion Models

A new arXiv paper derives generalization guarantees for both variational autoencoders and diffusion models within a single information-theoretic framework. The analysis exploits the encoder-generator structure shared by the two model families, which earlier theoretical work had largely treated separately. The authors report bounds that clarify how the shared architecture affects performance on unseen data.

research/diffusion/generative-models/theory

papersSEP 12 04:00 UTC

LoaDiff: Conditional Generation of Electricity Consumption Time Series

A new arXiv preprint introduces LoaDiff, a method for conditionally generating residential electricity consumption time series. The work is motivated by the energy transition, where distributed generation, electrified appliances and demand-response programs are shifting how households use power. The authors argue that granular synthetic consumption data can support energy analytics; the abstract is truncated in this report.

time-series/energy/generative-models/arxiv