Meta-Algorithms Research Group

The goals of the meta algorithms research group is to develop automatic algorithm/model selection for optimization problems or for machine learning tasks. In scenarios where multiple algorithms are available that differ in their properties and the running time they require, the goal is to select, for a given data set, an algorithm that will deliver the best possible solution, in terms of the defined evaluation criteria, and within computational constraints on time limit and computational resources. One of the challenges addressed by the group is identifying compact fast-to-compute representations (characterizations) of data sets and problem instances, particularly in the presence of massive data sets. We plan to devise a suitable machine learning technique for each family of problems, that will be trained to recognize the problem set properties and to select an algorithm that achieves best performance according to the user's objectives and restrictions. Our plan is to focus on broadly studied problems with strong links to real world problems including Vehicle Routing, Satisfiability (SAT), Travelling Salesperson Problem (TSP), Planning, Multi Agent Path Finding and Classification.

dorit_0_square

Leader

Dorit S. Hochbaum

University of California, Berkeley IEOR department
hochbaum@ieor.berkeley.edu

Roberto-Asin

Co-Leader

Roberto Javier Asín Achá

University of Concepcion Faculty
rasin@inf.udec.cl

ai4opt
UCBEngineering_logo
2ColorPrimary_IEOR_WEB_RGB

The Meta-Algorithms Research Group is part of the NSF AI Institute for Advances in Optimization in Berkeley Engineering's Department of Industrial Engineering & Operations Research.  Learn more