Barry Hurley

Proteus: A Hierarchical Portfolio of Solvers and Transformations

In recent years, portfolio approaches to solving SAT problems and CSPs have become increasingly common. There are also a number of different encodings for representing CSPs as SAT instances. In this paper, we leverage advances in both SAT and CSP solving to present a novel hierarchical portfolio-based approach to CSP solving, which we call Proteus, that does not rely purely on CSP solvers. Instead, it may decide that it is best to encode a CSP problem instance into SAT, selecting an appropriate encoding and a corresponding SAT solver. Our experimental evaluation used an instance of Proteus that involved four CSP solvers, three SAT encodings, and six SAT solvers, evaluated on the most challenging problem instances from the CSP solver competitions, involving global and intensional constraints. We show that significant performance improvements can be achieved by Proteus obtained by exploiting alternative view-points and solvers for combinatorial problem-solving.

Multi-language evaluation of exact solvers in graphical model discrete optimization

By representing the constraints and objective function in factorized form, graphical models can concisely define various NP-hard optimization problems. They are therefore extensively used in several areas of computer science and artificial intelligence. Graphical models can be deterministic or stochastic, optimize a sum or product of local functions, defining a joint cost or probability distribution. Simple transformations exist between these two types of models, but also with MaxSAT or linear programming. In this paper, we report on a large comparison of exact solvers which are all state-of-the-art for their own target language. These solvers are all evaluated on deterministic and probabilistic graphical models coming from the Probabilistic Inference Challenge 2011, the Computer Vision and Pattern Recognition OpenGM2 benchmark, the Weighted Partial MaxSAT Evaluation 2013, the MaxCSP 2008 Competition, the MiniZinc Challenge 2012 & 2013, and the CFLib (a library of Cost Function Networks). All 3026 instances are made publicly available in five different formats and seven formulations. To our knowledge, this is the first evaluation that encompasses such a large set of related NP-complete optimization frameworks, despite their tight connections. The results show that a small number of evaluated solvers are able to perform well on multiple areas. By exploiting the variability and complementarity of solver performances, we show that a simple portfolio approach can be very effective. This portfolio won the last UAI Evaluation 2014 (MAP task).

The ICON Challenge on Algorithm Selection

Algorithm selection is of increasing practical relevance in a variety of applications. Many approaches have been proposed in the literature, but their evaluations are often not comparable, making it hard to judge which approaches work best. The ICON Challenge on Algorithm Selection objectively evaluated many prominent approaches from the literature, making them directly comparable for the first time. The results show that there is still room for improvement, even for the very best approaches.

Adaptation in a CBR-Based Solver Portfolio for the Satisfiability Problem

The satisfiability problem was amongst the very first problems proven to be NP-Complete. It arises in many real world domains such as hardware verification, planning, scheduling, configuration and telecommunications. Recently, there has been growing interest in using portfolios of solvers for this problem. In this paper we present a case-based reasoning approach to SAT solving. A key challenge is the adaptation phase, which we focus on in some depth. We present a variety of adaptation approaches, some heuristic, and one that computes an optimal Kemeny ranking over solvers in our portfolio. Our evaluation over three large case bases of problem instances from artificial, hand-crafted and industrial domains, shows the power of a CBR approach, and the importance of the adaptation schemes used.

Introduction to Combinatorial Optimisation in Numberjack

This chapter presents an introduction to combinatorial optimisation in the context of the high-level modelling platform, Numberjack. The process of developing an effective model for a combinatorial problem is presented, along with details on how such problems can be solved using three of the most prominent solution paradigms.

ICON Loop Energy Show Case

This chapter demonstrates the effectiveness of the ICON loop when applied to energy cost optimization in a data centre. The objective is to schedule the execution of customer tasks such that the overall energy cost is minimised. This is complicated by the fact that the real-time energy price is not known a-priori, therefore machine learning techniques are employed to produce a forecast price vector ahead of time. In practice such a forecast needs to adapt to changes in the world affecting the pricing model over time. Therefore, the model needs to adapt in an iterative process, realised by employing the ICON loop approach.

Advanced Portfolio Techniques

There exists a proliferation of different approaches to using portfolios and algorithm selection to make solving combinatorial search and optimisation problems more efficient, as surveyed in the previous chapter. In this chapter, we take a look at a detailed case study that leverages transformations between problem representations to make portfolios more effective, followed by extensions to the state of the art that make algorithm selection more robust in practice.

ICON Loop Health Show Case

In this document we describe the health show case for the ICON project. This corresponds to Task 6.2 in WP 6 of the Description of Work for the project. The description provides a high-level abstraction, detailed description of the interfaces between modules, and a description of sample data.

Large Neighbourhood Search for Energy-Efficient Train Timetabling

The electric rail sector, like many sectors, is looking for means to reduce its energy consumption and energy cost. In this work we consider the scenario where the utility provider charges based on the maximum consumption over a period. Therefore one wishes to schedule the departure of trains such that the aggregate load is balanced across time periods while satisfying timetabling and resource restrictions. We present an approach which combines the strengths of a number of research areas such as constraint programming, linear programming, mixed-integer programming, and large neighbourhood search. The empirical performance on instances from an ongoing research challenge demonstrates the approachs ability to dramatically reduce the overall energy cost. In addition, we are able to close a number of the instances for which we prove optimality.