Reward Transport Controls Molecular Properties in Flow Matching.

Kehan Guo, Yili Shen, Yujun Zhou, Yue Huang, Chujie Gao, Shiyi Du, Xiangliang Zhang· July 13, 2026 View original

Summary

Reward Transport introduces a novel method to control molecular properties in flow matching models by aligning a scalar noise-space coordinate with molecular rewards during training. This allows for steering generated distributions at inference time without needing an oracle or gradient guidance.

Researchers have developed "Reward Transport," an innovative technique that leverages the coupling mechanism in flow matching models to embed controllable structure directly into learned flow fields. Traditionally, this coupling, which pairs noise vectors with data points, is seen as a computational detail. Reward Transport redefines this coupling as an alignment interface. By using optimal transport coupling during training, it aligns a scalar coordinate in the noise space with specific molecular rewards. This alignment enables precise control over generated molecular properties during inference. At inference, simply varying this noise-space coordinate allows for steering the generated distribution towards desired properties, eliminating the need for external oracles, reward models, or gradient guidance. The method has been empirically validated on ZINC-250K and GuacaMol datasets, demonstrating monotone control of logP and consistent QED control, with the same control knob producing opposite structural responses for different targets, confirming its specific property-steering capability.

Why it matters

This breakthrough offers a more direct and efficient way to design molecules with desired properties, accelerating drug discovery, materials science, and chemical engineering processes.

How to implement this in your domain

  1. 1Explore integrating Reward Transport into existing generative AI pipelines for molecular design or materials discovery.
  2. 2Identify specific molecular properties that are critical for current research or product development and define corresponding reward functions.
  3. 3Experiment with varying the noise-space coordinate during inference to generate molecules with a controlled range of desired properties.
  4. 4Validate the generated molecules through simulations or experimental synthesis to confirm the efficacy of the property control.

Who benefits

PharmaceuticalsBiotechnologyMaterials ScienceChemical EngineeringAI/ML Development

Key takeaways

  • Reward Transport enables direct control of molecular properties in flow matching models.
  • It aligns noise-space coordinates with molecular rewards during training.
  • This allows for steering generated distributions at inference without external guidance.
  • The method shows effective control over properties like logP and QED in molecular generation.

Original post by Kehan Guo, Yili Shen, Yujun Zhou, Yue Huang, Chujie Gao, Shiyi Du, Xiangliang Zhang

"arXiv:2607.08781v1 Announce Type: cross Abstract: The coupling in flow matching -- the rule pairing noise vectors with data points -- is typically treated as a computational choice. We show that this coupling can instead serve as an alignment interface: by matching noise and data…"

View on X

Originally posted by Kehan Guo, Yili Shen, Yujun Zhou, Yue Huang, Chujie Gao, Shiyi Du, Xiangliang Zhang on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Data-Efficient Deep Learning Guidelines for Inertial Sensor Classification

This study provides empirical guidelines for estimating the minimum training set size needed for deep learning models in inertial sensor classification tasks. It reveals that accuracy follows a consistent logarithmic growth pattern, allowing for data-efficient planning of recording campaigns and achieving practical stability with fewer samples than traditionally assumed.

Ofir Kruzel, Itzik KlienJul 13, 2026
AI Engineering & DevToolsAI Research

On-Device Adaptive AI Boosts EV Battery Power Prediction

Researchers developed a novel approach for on-device learning in electric vehicles (EVs) that continuously adapts pretrained battery power prediction models to new data. This method significantly improves forecasting performance, reducing mean absolute errors by up to 14.88% with offline adaptation and 7.49% with online adaptation.

Avik Bhatnagar, Anton Paule, Tobias Schuermann, Sebastian Reiter, Oliver BringmannJul 13, 2026
AI ResearchAI Engineering & DevTools

New Differentiable Logic Networks Outperform Fixed-Connection Models

Researchers introduce a novel method for optimizing connections in deep differentiable logic gate networks (LGNs) and lookup table networks (LUTNs), achieving superior performance with significantly fewer gates. The approach allows for parallel learning of optimal gate types and LUT entries, demonstrating improved accuracy on benchmarks like MNIST.

Wout Mommen, Lars Keuninckx, Matthias Hartmann, Werner Van Leekwijck, Piet WambacqJul 13, 2026