![[2602.17827] Avoid What You Know: Divergent Trajectory Balance for GFlowNets](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17827] Avoid What You Know: Divergent Trajectory Balance for GFlowNets
Summary
The paper presents Adaptive Complementary Exploration (ACE), an algorithm designed to enhance the efficiency of Generative Flow Networks (GFlowNets) by improving exploration of high-reward states during training.
Why It Matters
This research addresses a critical limitation in GFlowNets, which struggle with efficient exploration of diverse state spaces. By proposing ACE, the authors aim to improve the learning process for generative models, which has significant implications for various applications in machine learning and AI.
Key Takeaways
- ACE enhances exploration efficiency in GFlowNets.
- The algorithm focuses on discovering high-reward states.
- Extensive experiments show improved approximation accuracy.
- Curiosity-driven search methods may waste samples on known regions.
- ACE represents a significant advancement in generative modeling techniques.
Computer Science > Machine Learning arXiv:2602.17827 (cs) [Submitted on 19 Feb 2026] Title:Avoid What You Know: Divergent Trajectory Balance for GFlowNets Authors:Pedro Dall'Antonia, Tiago da Silva, Daniel Csillag, Salem Lahlou, Diego Mesquita View a PDF of the paper titled Avoid What You Know: Divergent Trajectory Balance for GFlowNets, by Pedro Dall'Antonia and Tiago da Silva and Daniel Csillag and Salem Lahlou and Diego Mesquita View PDF HTML (experimental) Abstract:Generative Flow Networks (GFlowNets) are a flexible family of amortized samplers trained to generate discrete and compositional objects with probability proportional to a reward function. However, learning efficiency is constrained by the model's ability to rapidly explore diverse high-probability regions during training. To mitigate this issue, recent works have focused on incentivizing the exploration of unvisited and valuable states via curiosity-driven search and self-supervised random network distillation, which tend to waste samples on already well-approximated regions of the state space. In this context, we propose Adaptive Complementary Exploration (ACE), a principled algorithm for the effective exploration of novel and high-probability regions when learning GFlowNets. To achieve this, ACE introduces an exploration GFlowNet explicitly trained to search for high-reward states in regions underexplored by the canonical GFlowNet, which learns to sample from the target distribution. Through extensive expe...