arXiv Machine Learning
Aug 6
GFlowNet Training by Policy Gradients
arXiv:2408. 05885v3 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) have been shown effective to generate combinatorial objects with desired properties.
Rethinking GFlowNet Training: Negative Reinforce Dynamics
The paper introduces Negative Reinforce Dynamics (NRD), a new training approach for Generative Flow Networks (GFlowNets) that addresses the exploration–correctness trade‑off inherent in standard semi‑gradient updates. By linking the full gradient to a min‑min dynamic between a GFlowNet and a decoupled sampler, NRD effectively reduces balance errors in high‑error regions while preserving a single‑policy framework. Experiments on standard benchmarks show that NRD speeds up convergence and improves exploration efficiency compared to existing methods, without requiring a separate teacher or auxiliary policy.
Related stories
arXiv AI
Jun 6
Retry Policy Gradients in Continuous Action Spaces
arXiv:2606. 05888v1 Announce Type: new Abstract: Retry-based objectives such as pass@K and max@K optimize the best return obtained from multiple sampled trajectories, and recent work has shown that they can promote exploration without explicit exploration bonuses.
arXiv Machine Learning
Aug 6
Stable GFlowNets with TV Monitoring and Probabilistic Guarantees
arXiv:2605. 01729v2 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) learn to sample states proportional to an unnormalized reward.
arXiv Machine Learning
Aug 7
IFlowNets: Extending Generative Samplers to Learn Strategies in Incomplete Information Games
arXiv:2608. 05422v1 Announce Type: new Abstract: While many algorithms blend reinforcement learning (RL) with counterfactual regret (CFR) methods to leverage tradeoffs in computational speed and performance, there are fewer investigations into generative sampling frameworks in game theoretic applications in incomplete information games.
arXiv Machine Learning
Oct 12
Towards Understanding the Training of GFlowNets: A Theoretical Perspective
The paper presents a theoretical framework linking balance objectives in Generative Flow Networks (GFlowNets) to sampling accuracy. It establishes bounds on terminal symmetric KL divergence in terms of expected balance loss and shows how approximate stationarity controls distribution error, accounting for parameterization incompleteness. The authors prove finite‑time convergence rates under various sampling and backward‑policy regimes, clarifying the roles of coverage, parameterization, and moving components in GFlowNet training.
arXiv Machine Learning
Aug 31
Trajectory balance: Improved credit assignment in GFlowNets
The paper introduces a new learning objective called trajectory balance for Generative Flow Networks (GFlowNets), aiming to improve credit assignment across long action sequences. It demonstrates that minimizing this objective yields a policy that samples exactly from the target distribution. Experiments on four domains show that trajectory balance enhances convergence, sample diversity, and robustness to long sequences and large action spaces.
arXiv Machine Learning
Sep 1
Expected flow networks in stochastic environments and two-player zero-sum games
arXiv:2310.02779v3 Announce Type: replace Abstract: Generative flow networks (GFlowNets) are sequential sampling models trained to match a given distribution. GFlowNets have been successfully applied...
arXiv Machine Learning
Sep 24
WTF?! Simulation-Free Reinforcement Learning with Wasserstein-Tilted Flow Maps
The paper introduces Wasserstein‑Tilted Flow Maps (WTF), a simulation‑free reinforcement learning method that fine‑tunes pre‑trained flow‑based generative models by adding an optimal transport regularizer derived from the model’s drift. Unlike traditional KL‑reward tilting, WTF transports individual samples toward higher reward, framing the problem as a deterministic optimal control task on the flow map. Experiments on ImageNet‑256 and text‑to‑image demonstrate that WTF achieves higher reward and comparable or better diversity while reducing training compute by up to 280×.
arXiv AI
Jun 16
Proximal Policy Optimization for Amortized Discrete Sampling
arXiv:2606. 15793v1 Announce Type: cross Abstract: This paper explores policy gradient algorithms for training stochastic policies to sample from structured discrete probability distributions under the Generative Flow Network (GFlowNet) framework.
arXiv Machine Learning
Oct 12
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
DiffusionART introduces a new reinforcement learning framework for diffusion models that aggregates prompt rewards into a shared direction, eliminating consistency forces inherent in prior methods. It employs a trust‑region‑masked second update for drift‑adaptive restoration, removing the need for EMA tuning and differentiable reward models. Experiments on image and video generation show that DiffusionART trains more efficiently and stably, achieving higher mean rewards with significantly fewer GPU hours compared to existing approaches.
arXiv AI
Sep 3
Action abstractions for amortized sampling
The paper introduces a method that integrates action abstraction into policy optimization for reinforcement learning and generative flow networks. By iteratively identifying frequently used action subsequences in high‑reward trajectories and treating them as single high‑level actions, the approach expands the action space and improves sample efficiency. Experiments on synthetic and real‑world tasks show that this technique discovers diverse high‑reward states more effectively, especially on challenging exploration problems, and yields interpretable abstract actions that reflect the underlying reward structure.
arXiv AI
Sep 30
Learn from the Gap: Differential-Aware Advantage Pruning with Adaptive Rollout Sampling for GRPO
The paper introduces FastRL, a reinforcement learning framework designed to enhance the efficiency of Group Relative Policy Optimization (GRPO) and its variants. FastRL employs an advantage-aware pruning strategy that retains high-advantage trajectories while preserving gradient diversity, and an adaptive rollout sampling mechanism that adjusts sampling scale during training based on historical pruning data. Experiments show that FastRL can be integrated into GRPO, DAPO, and GSPO, yielding a 2.07× speedup on Geometry3K and GeoQA8K-R1V and a 1.64% accuracy improvement on visual reasoning benchmarks.