English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Generative Flow Networks (GFlowNets) are amortized samplers that learnstochastic policies to sequentially generate compositional objects from a givenunnormalized reward distribution. They can generate diverse sets of high-rewardobjects, which is an important consideration in scientific discovery tasks.However, as they are typically trained from a given extrinsic reward function,it remains an important open challenge about how to leverage the power ofpre-training and train GFlowNets in an unsupervised fashion for efficientadaptation to downstream tasks. Inspired by recent successes of unsupervisedpre-training in various domains, we introduce a novel approach for reward-freepre-training of GFlowNets. By framing the training as a self-supervisedproblem, we propose an outcome-conditioned GFlowNet (OC-GFN) that learns toexplore the candidate space. Specifically, OC-GFN learns to reach any targetedoutcomes, akin to goal-conditioned policies in reinforcement learning. We showthat the pre-trained OC-GFN model can allow for a direct extraction of a policycapable of sampling from any new reward functions in downstream tasks.Nonetheless, adapting OC-GFN on a downstream task-specific reward involves anintractable marginalization over possible outcomes. We propose a novel way toapproximate this marginalization by learning an amortized predictor enablingefficient fine-tuning. Extensive experimental results validate the efficacy ofour approach, demonstrating the effectiveness of pre-training the OC-GFN, andits ability to swiftly adapt to downstream tasks and discover modes moreefficiently. This work may serve as a foundation for further exploration ofpre-training strategies in the context of GFlowNets.

Pre-Training and Fine-Tuning Generative Flow Networks