Latest posts tagged with #FlowMatching on Bluesky
6/
📄 Paper: arxiv.org/pdf/2601.18005
💻 Code: github.com/berczig/Flow...
Built at
@au.dk
and
@maxplanck.de.
Happy to discuss methodology, results, or potential applications!
#AI4Mathematics #AI4Science #FlowMatching
📢Out now! @stemartiniani.bsky.social and colleagues present PropMolFlow, a flow-matching method for property-guided molecule generation. #MoleculeDiscovery #FlowMatching www.nature.com/articles/s43... #chemsky
🔓 rdcu.be/eZ5cG
Value Gradient Guidance for Flow Matching Alignment
Carles Domingo-Enrich, Dinghuai Zhang et al.
Paper
Details
#ValueGradient #FlowMatching #Alignment
Flow-Matching Refiner Boosts Molecular Conformer Generation
The flow‑matching refiner improves molecular conformer quality while cutting denoising steps, and testing on GEOM‑QM9 and GEOM‑Drugs benchmarks showed improvements. Read more: getnews.me/flow-matching-refiner-bo... #flowmatching #conformer
Flow Matching Advances MRI‑CT and CBCT‑CT Synthetic Imaging
Researchers unveiled a 3‑D Flow Matching framework that creates synthetic CT from MRI or CBCT. It placed in the SynthRAD2025 Challenge and the pre‑print was posted on 6 Oct 2025. Read more: getnews.me/flow-matching-advances-m... #flowmatching #syntheticct
Flow-Matching Warm Start Boosts Robotic Arm Motion Planning
A flow-matching warm-start model using a single-view depth camera point cloud was tested on a Universal Robots UR5e arm, achieving higher success rates than random or conventional initializers. getnews.me/flow-matching-warm-start... #flowmatching #ur5e
Smart-GRPO Optimizes Noise Sampling for RL in Flow-Matching Models
Smart-GRPO improves reward optimization and visual quality in flow-matching models, outperforming baseline random-noise methods, the authors report on 3 Oct 2025. Read more: getnews.me/smart-grpo-optimizes-noi... #smartgrpo #flowmatching
Purrception: Variational Flow Matching for VQ Image Generation
Purrception offers variational flow matching for vector‑quantized image generation, delivering faster training and competitive FID on 256×256 pixel data. Submitted 1 Oct 2025. Read more: getnews.me/purrception-variational-... #flowmatching #vq
Vocoder-Free Audio Super-Resolution Achieves State-of-the-Art Quality
A vocoder‑free framework using flow‑matching generates 48 kHz waveforms directly, delivering state‑of‑the‑art audio super‑resolution across upsampling factors. Read more: getnews.me/vocoder-free-audio-super... #audiosuperresolution #flowmatching
Flow Matching Boosts Simulation-Based Inference under Misspecification
FMCPE adds a flow‑matching correction with real calibration samples, cutting inference error and improving uncertainty calibration on synthetic benchmarks. getnews.me/flow-matching-boosts-sim... #simulationinference #flowmatching
Score Distillation Extends to Text-to-Image Flow-Matching Models
Score‑distillation now works on flow‑matching text‑to‑image models such as SANA and FLUX.1‑dev, delivering single‑step generation without extra training. Read more: getnews.me/score-distillation-exten... #scoredistillation #flowmatching #texttoimage
OAT-FM Advances Flow Matching with Optimal Acceleration Transport
Researchers introduced OAT-FM, a method that improves flow matching by optimizing acceleration transport rather than constant velocity, and have released the code openly on GitHub. Read more: getnews.me/oat-fm-advances-flow-mat... #oatfm #flowmatching
Diffusion Bridge vs Flow Matching: Unified Theory and Comparative Experiments
A unified theory shows Diffusion Bridge keeps visual fidelity on limited data, while Flow Matching trains faster with more data. September 2025. Read more: getnews.me/diffusion-bridge-vs-flow... #diffusionbridge #flowmatching
Physics‑Informed ML Reduces Carbon Footprint in Forecasting
Adding physics‑based knowledge to ML models cuts compute needs and carbon emissions, boosting forecast accuracy, and flow‑matching method released on GitHub enables spatio‑temporal forecasting. getnews.me/physics-informed-ml-redu... #physicsml #flowmatching
Inversion-Free Method Boosts AI Video Editing with Flow-Based Models
IF-V2V eliminates inversion for flow‑matching image‑to‑video models, using rectification and deviation caching to keep visual fidelity. It works as a plug‑and‑play for I2V. Read more: getnews.me/inversion-free-method-bo... #ifv2v #flowmatching
Shortcut Flow Matching Enables Real‑Time AI Speech Enhancement
SFMSE achieves single‑step speech enhancement with a real‑time factor of 0.013 on a GPU, delivering audio 77× faster than real time while matching a diffusion baseline needing 60 NFEs. getnews.me/shortcut-flow-matching-e... #speechenhancement #flowmatching
8/
✨ Transcriptomes from images ✨
Using optimal transport + flow matching, scPortrait generates gene expression directly from CODEX images, capturing canonical marker expression like TCL1A in germinal centers in the tonsil
#CODEX #flowmatching #OT
Unified Diffusion Bridge Framework Merges Flow Matching and Schrödinger Methods
A new framework unifies flow matching and Schrödinger bridge methods, reproducing Flow Matching, optimal‑transport FM, mini‑batch Schrödinger FM and DSBM. Read more: getnews.me/unified-diffusion-bridge... #flowmatching #schrodingerbridge
At #ICML2025 this week?
Come check out our work on controlled generation, simulation-free latent SDEs, and risk monitoring in test-time adaptation, and chat with the awesome students that made it happen!
#SDE #Diffusion #FlowMatching #TTA #UncertaintyQuantification
🚨🚀
Come hear @gbarto.bsky.social talk about SDE Matching tomorrow!
SDE Matching is a highly efficient and scalable training framework for Latent/Neural SDEs.
You no longer have to discretize or simulate your SDE models when fitting them to data.
#SDE #Diffusion #FlowMatching #ML
Exploring the intersection of Diffusion and Flow Matching models! 🚀 1. Flow models intro. 2. Clears up the confusion, showing how Gaussian flow matching and diffusion models align:
1. 🔗 mlg.eng.cam.ac.uk/blog/2024/01...
2. 🔗 diffusionflow.github.io
#FlowMatching #DiffusionModels #GenerativeModels
🚀 New preprint!
We enhance the efficiency and performance of diffusion models by introducing a new Total-Variance / Signal-to-Noise Ratio (TV/SNR) disentangled framework, allowing independent control over both.
🔗 arXiv: arxiv.org/abs/2502.08598
@bifold.berlin #ML #DiffusionModels #FlowMatching
Thanks all who helped out and commented in this thread with tips and tricks!
I gave a talked recently on some work we have been doing in the lab that incorporated aspects from several of the responses.
docs.google.com/presentation...
#DiffusionModels #FlowMatching #StochasticInterpolants
Fellow #Diffusion and #FlowMatching enthusiasts, any tips for presenting this class of generative models to a broader ML audience? Any examples of slides that you have seen (or made) that you really like? 📊
So exciting to see everyone here. A brief self-intro for others:
I am currently a PhD student with @florianjug.bsky.social. My work involves the application of AI in bioimage analysis. I am interested in the application of #GenAI to solve inverse problems with #DiffusionModels #FlowMatching #VAEs.
Twitter saw it first but to get to grips with discrete flow matching paper, I've put together a little notebook to test it on MNIST. I kept the article's notations so it's easy to follow alongside the article! #flowmatching #generativeAI
🐍 notebook: github.com/gle-bellier/...
