We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Unsupervised Resource Allocation with Graph Neural Networks

Authors: Miles Cranmer (Princeton), Peter Melchior (Princeton), Brian Nord (Fermilab)
Abstract: We present an approach for maximizing a global utility function by learning how to allocate resources in an unsupervised way. We expect interactions between allocation targets to be important and therefore propose to learn the reward structure for near-optimal allocation policies with a GNN. By relaxing the resource constraint, we can employ gradient-based optimization in contrast to more standard evolutionary algorithms. Our algorithm is motivated by a problem in modern astronomy, where one needs to select-based on limited initial information-among $10^9$ galaxies those whose detailed measurement will lead to optimal inference of the composition of the universe. Our technique presents a way of flexibly learning an allocation strategy by only requiring forward simulators for the physics of interest and the measurement process. We anticipate that our technique will also find applications in a range of resource allocation problems.
Comments: Accepted to PMLR/contributed oral at NeurIPS 2020 Pre-registration Workshop. Code at this https URL
Subjects: Machine Learning (cs.LG); Cosmology and Nongalactic Astrophysics (astro-ph.CO); Instrumentation and Methods for Astrophysics (astro-ph.IM); Artificial Intelligence (cs.AI)
Cite as: arXiv:2106.09761 [cs.LG]
  (or arXiv:2106.09761v1 [cs.LG] for this version)

Submission history

From: Miles Cranmer [view email]
[v1] Thu, 17 Jun 2021 18:44:04 GMT (2937kb,D)

Link back to: arXiv, form interface, contact.