Bi-Weekly Talk: Martin Ritzert: Maximum Constraint Satisfaction with Unsupervised NeuralNetworks

Wednesday, May 06, 2020, 10:30am

Location: Zoom meeting

Speaker: Martin Ritzert

 

Abstract: 

In this talk I will present recent work by our chair on aneural network model that works on all binary constraint satisfaction problems,including the four NP-hard problems max-2-SAT, max-cut 3-coloring as well asmaximum independent set.

We show that our model, although highly general, can beata large variety of neural and heuristic baselines and in the case of max-2-SATeven outperforms the state-of-the-art heuristic.