analiza aiqbench
GReinSS introduces a policy learning framework that effectively reconstructs mechanistic latent states in combinatorial spaces, addressing key limitations of variational autoencoders that often generate artificial representations. By utilizing dynamically rescaled rewards, the method demonstrates superior performance in both simulated environments and real-world RNA sequencing tasks compared to standard algorithms. This advancement offers a principled approach for inferring unobserved states from indirect observations, which is critical for scaling generative modeling in complex scientific domains.
Generative Modeling of Discrete Latent Structures via Dynamic Policy Gradients
arXiv:2606.07400v1 Announce Type: new Abstract: Many scientific problems require inferring unobserved mechanistic latent states from indirect observations. While classical approaches, including expectation maximization, do not scale to combinatorially large spaces, deep learning approaches such as variational autoencoders typically form artificial latent states rather than reconstructing the mechanistic ground-truth states. Here, we introduce GReinSS, a policy learning framework that uses dynamically rescaled rewards to learn latent state distributions that maximize the observed data likelihood. We show that GReinSS accurately reconstructs simulated latent sets and latent graphs, outperforming alternative policy learning and generative modeling baselines. Additionally, GReinSS reconstructs isoforms from real short-read RNA sequencing data that better match isoforms detected by orthogonal long-read sequencing than the standard RSEM algorithm. Overall, GReinSS is a principled and practically effective approach for generative modeling and inference of combinatorial latent states from indirect observations.
Abstract:Many scientific problems require inferring unobserved mechanistic latent states from indirect observations. While classical approaches, including expectation maximization, do not scale to combinatorially large spaces, deep learning approaches such as variational autoencoders typically form artificial latent states rather than reconstructing the mechanistic ground-truth states. Here, we introduce GReinSS, a policy learning framework that uses dynamically rescaled rewards to learn latent state distributions that maximize the observed data likelihood. We show that GReinSS accurately reconstructs simulated latent sets and latent graphs, outperforming alternative policy learning and generative modeling baselines. Additionally, GReinSS reconstructs isoforms from real short-read RNA sequencing data that better match isoforms detected by orthogonal long-read sequencing than the standard RSEM algorithm. Overall, GReinSS is a principled and practically effective approach for generative modeling and inference of combinatorial latent states from indirect observations.
Machine Learning (cs.LG)
arXiv:2606.07400 [cs.LG]
(or arXiv:2606.07400v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.07400 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 227 cuvinte
Subiecte conexe
LMBuild: Evaluating LLM Agents for Generating Buildable and Functional Structures
arXiv · 7 oct. 2026
Rethinking Structured Generation: Can Graph-Based Reasoning Resolve Ambiguity?
arXiv · 27 oct. 2026
Dystruct: Dynamically Structured Diffusion Language Model Decoding via Bayesian Inference
arXiv · 3 nov. 2026
XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMs
arXiv · 6 aug. 2026