In this paper we treat the problem of feature selection in unsupervised learning as a state-space search problem. We introduce three different heuristic functions and perform extensive experiments on datasets with tens, hundreds, and thousands of features. Namely, we test different search algorithms using the heuristic functions we introduce. Our results show that the heuristic search approach for feature selection in unsupervised learning problems can be far superior than traditional baselines such as PCA and random projections.

The Heuristic Search community has been concentrating much effort during the last decades in solving more and more efficiently the SHORTEST PATH problem (SPP). As a result, a valuable body of scientific results has been produced, mostly in the form of heuristics and search algorithms. However, not much attention has been given to other problems even if they result from slight variations of the typical problems addressed by the community. Furthermore, other communities attempt at solving hard combinatorial problems which might be well solved with heuristic search. In this paper, an attempt is presented to introduce a preliminary selection of relevant problems that goes well beyond the classical SPP.

The typical objective of path planning is to find the shortest feasible path. Many times, however, there may be no solution given the existence of constraints, such as obstacles. In these cases, the minimum constraint removal problem asks for the minimum set of constraints that need to be removed from the state space to find a solution. Unfortunately, minimum constraint removal paths do not exhibit dynamic programming properties, i.e., subsets of optimum solutions are not necessarily optimal. Thus, searching for such solutions is computationally expensive. This leads to approximate methods, which balance the cost of computing a solution and its quality. This work investigates alternatives in this context and evaluates their performance in terms of such tradeoffs. Solutions that follow a bounded-length approach, i.e., searching for paths up to a certain length, seem to provide a good balance between minimizing constraints, computational cost and path length.

The traditional focus of combinatorial search research is to speed up the search algorithm. An alternative, however, is to create a sparser representation of the search space. This relates to the idea of spanners from graph theory. These are subgraphs which retain paths between any two vertices of the original graph while guaranteeing a maximum stretch in path length. In practice, the path degradation of graph spanners is significantly smaller than the theoretical bound. Even so, expected path degradation of graph spanners is typically not studied. This work focuses on grid path-finding to propose an algorithm that constructs a grid spanner, where analysis for the obstacle-free case shows that significant performance gains can be achieved with a small decrease in expected path quality. This is an important first step towards studying the expected performance of spanners. Experiments on game maps show that expected path quality with obstacles is only sometimes marginally lower than that in the obstacle-free case and that a significant reduction in the size of the search space can be achieved.

In empirical studies we observed that caching can have very little impact in reducing the search effort in Branch and Bound search over context-minimal OR spaces. For example, in one of the problem domains used in our experiments we reduce only by 1% the number of nodes expanded when using caching in context-minimal OR spaces. By contrast, we reduce by 74% the number of nodes expanded when using caching in context-minimal AND/OR spaces on the same instances. In this work we document this unexpected empirical finding and provide explanations for the phenomenon.

Restricting the search space has shown to be an effective approach for improving the performance of automated planning systems. A planner-independent technique for pruning the search space is domain and problem reformulation. Recently, Outer Entanglements, which are relations between planning operators and initial or goal predicates, have been introduced as a reformulation technique for eliminating potential undesirable instances of planning operators, and thus restricting the search space. Reformulation techniques, however, have been mainly applied in classical planning, although many real-world planning applications require to deal with numerical information. In this paper, we investigate the usefulness of reformulation approaches in planning with numerical fluents. In particular, we propose and extension of the notion of outer entanglements for handling numeric fluents. An empirical evaluation, which involves 150 instances from 5 domains, shows promising results.

Monte-Carlo Tree Search (MCTS) algorithms estimate the value of MDP states based on rewards received by performing multiple random simulations. MCTS algorithms can use different strategies to aggregate these rewards and provide an estimation for the states’ values. The most common aggregation method is to store the mean reward of all simulations. Another common approach stores the best observed reward from each state. Both of these methods have complementary benefits and drawbacks. In this paper, we show that both of these methods are biased estimators for the real expected value of MDP states. We propose an hybrid approach that uses the best reward for states with low noise, and otherwise uses the mean. Experimental results on the Sailing MDP domain show that our method has a considerable advantage when the rewards are drawn from a noisy distribution.

We address planning with always preferences in propositional domains, proposing a new compilation schema for translating a STRIPS problem enriched with always preferences (and possibly also soft goals) into a STRIPS problem with action costs. Our method allows many STRIPS planners to effectively address planning with always preferences and soft goals. An experimental analysis indicates that such basic planners are competitive with current planners using techniques specifically developed to handle always preferences.

Meta-reasoning can improve numerous search algorithms, but necessitates collection of statistics to be used as probability distributions, and involves restrictive meta-reasoning assumptions. The recently suggested scheme of type systems in search algorithms is used in this paper for collecting these statistics. The statistics are then used to better estimate the unknown quantity of expected regret of computing a heuristic in Rational Lazy IDA* (RLIDA*), and also facilitate a second improvement due to relaxing one of the unrealistic meta-reasoning assumptions in RLIDA*.

Solving planning problems via translation to propositional satisfiability (SAT) is one of the most successful approaches to automated planning. An important aspect of this approach is the encoding, i.e., the construction of a propositional formula from a given planning problem instance. Numerous encoding schemes have been proposed in the recent years each aiming to outperform the previous encodings on the majority of the benchmark problems. In this paper we take a different approach. Instead of trying to develop a new encoding that is better for all kinds of benchmarks we take recently developed specialized encoding schemes and design a method to automatically select the proper encoding for a given planning problem instance. In the paper we also examine ranking heuristics for the Relaxed Relaxed Exists-Step encoding, which plays an important role in our algorithm. Experiments show that our new approach significantly outperforms the state-of-the-art encoding schemes when compared on the benchmarks of the 2011 International Planning Competition.