Foundation models (FMs) are achieving remarkable successes to realize complex downstream tasks in domains including natural language and visions. In this paper, we propose building an FM for material science, which is trained with massive data across a wide variety of material domains and data modalities. Nowadays machine learning models play key roles in material discovery, particularly for property prediction and structure generation. However, those models have been independently developed to address only specific tasks without sharing more global knowledge. Development of an FM for material science will enable overarching modeling across material domains and data modalities by sharing their feature representations. We discuss fundamental challenges and required technologies to build an FM from the aspects of data preparation, model development, and downstream tasks.

Using reinforcement learning for automated theorem proving has recently received much attention. Current approaches use representations of logical statements that often rely on the names used in these statements and, as a result, the models are generally not transferable from one domain to another. The size of these representations and whether to include the whole theory or part of it are other important decisions that affect the performance of these approaches as well as their runtime efficiency. In this paper, we present NIAGRA; an ensemble Name InvAriant Graph RepresentAtion. NIAGRA addresses this problem by using 1) improved Graph Neural Networks for learning name-invariant formula representations that is tailored for their unique characteristics and 2) an efficient ensemble approach for automated theorem proving. Our experimental evaluation shows state-of-the-art performance on multiple datasets from different domains with improvements up to 10% compared to the best learning-based approaches. Furthermore, transfer learning experiments show that our approach significantly outperforms other learning-based approaches by up to 28%.

Optimizing a machine learning (ML) pipeline has been an important topic of AI and ML. Despite recent progress, pipeline optimization remains a challenging problem, due to potentially many combinations to consider as well as slow training and validation. We present the BLDS algorithm for optimized algorithm selection (ML operations) in a fixed ML pipeline structure. BLDS performs multi-fidelity optimization for selecting ML algorithms trained with smaller computational overhead, while controlling its pipeline search based on multi-armed bandit and limited discrepancy search. Our experiments on well-known classification benchmarks show that BLDS is superior to competing algorithms. We also combine BLDS with hyperparameter optimization, empirically showing the advantage of BLDS.

AutoML automatically selects, composes and parameterizes machine learning algorithms into a workflow or pipeline of operations that aims at maximizing performance on a given dataset. Although current methods for AutoML achieved impressive results they mostly concentrate on optimizing fixed linear workflows. In this paper, we take a different approach and focus on generating and optimizing pipelines of complex directed acyclic graph shapes. These complex pipeline structure may lead to discovering hidden features and thus boost performance considerably. We explore the power of heuristic search and context-free grammars to search and optimize these kinds of pipelines. Experiments on various benchmark datasets show that our approach is highly competitive and often outperforms existing AutoML systems.

Finding a shortest path in a graph is at the core of many combinatorial search problems. A closely related problem refers to counting the number of shortest paths between two nodes. Such problems are solvable in polynomial time in the size of the graph. However, more realistic problem formulations could additionally specify constraints to satisfy. We study the problem of counting the shortest paths that are vertex disjoint and can satisfy additional constraints. Specifically, we look at the problems of counting vertex-disjoint shortest paths in edge-colored graphs, counting vertex-disjoint shortest paths with directional constraints, and counting vertex-disjoint shortest paths between multiple source-target pairs. We give a detailed theoretical analysis, and show formally that all of these three counting problems are NP-complete in general.

Computing cycle-free solutions in cyclic AND/OR search spaces is an important AI problem. Previous work on optimal depth-first search strongly assumes the use of consistent heuristics, the need to keep all examined states in a transposition table, and the existence of solutions. We give a new theoretical analysis under relaxed assumptions where previous results no longer hold. We then present a generic approachto proving unsolvability, and apply it to RBFAOO and BLDFS, two state-of-the-art algorithms. We demonstrate the performance in domain-independent nondeterministic planning

Search techniques, such as Monte Carlo Tree Search (MCTS) and Proof-Number Search (PNS), are effective in playing and solving games. However, the understanding of their performance in industrial applications is still limited. We investigate MCTS and Depth-First Proof-Number (DFPN) Search, a PNS variant, in the domain of Retrosynthetic Analysis (RA). We find that DFPN's strengths, that justify its success in games, have limited value in RA, and that an enhanced MCTS variant by Segler et al. significantly outperforms DFPN. We address this disadvantage of DFPN in RA with a novel approach to combine DFPN with Heuristic Edge Initialization. Our new search algorithm DFPN-E outperforms the enhanced MCTS in search time by a factor of 3 on average, with comparable success rates.

Multi-modal journey planning, which allows multiple types of transport within a single trip, is becoming increasingly popular, due to a strong practical interest and an increasing availability of data. In real life, transport networks feature uncertainty. Yet, most approaches assume a deterministic environment, making plans more prone to failures such as missed connections and major delays in the arrival. This paper presents an approach to computing optimal contingent plans in multi-modal journey planning. The problem is modeled as a search in an and/or state space. We describe search enhancements used on top of the AO* algorithm. Enhancements include admissible heuristics, multiple types of pruning that preserve the completeness and the optimality, and a hybrid search approach with a deterministic and a nondeterministic search. We demonstrate an NP-hardness result, with the hardness stemming from the dynamically changing distributions of the travel time random variables. We perform a detailed empirical analysis on realistic transport networks from cities such as Montpellier, Rome and Dublin. The results demonstrate the effectiveness of our algorithmic contributions, and the benefits of contingent plans as compared to standard sequential plans, when the arrival and departure times of buses are characterized by uncertainty.

We argue that chemistry should be the next grand challenge for Artificial Intelligence. The AI research community and humanity would benefit tremendously from focusing AI research on chemistry on a regular basis, as a benchmark as well as a real-world application domain. To support our position, we review the importance of chemical compound discovery and synthesis planning and discuss the properties of search spaces in a chemistry problem. Knowledge acquired in domains such as two-player board games or single-player puzzles places the AI community in a good position to solve critical problems in the chemistry domain. Yet, we show that searching in chemistry problems poses significant additional challenges that will have to be addressed. Finally, we envision how several AI areas like Natural Language Processing, Machine Learning, planning and search, are relevant for chemistry.

Quantum circuit compilation (QCC) is an important problem in the emerging field of quantum computing. The problem has a strong relevance to combinatorial search, as solving approaches recently published include constraint programming and temporal planning. In this paper, we focus on a complexity analysis of quantum circuit compilation. We formulate a makespan optimization problem based on QCC, and prove that the problem is NP-complete. To the best of our knowledge, this is the first study on the theoretical complexity of QCC.

Optimal heuristic search has been successful in many domains, including journey planning, route planning and puzzle solving. Existing work typically assumes that the cost of each action can easily be obtained. However, in many problems, the exact edge cost is expensive to compute. Existing search algorithms face a significant performance bottleneck, due to an excessive overhead associated with dynamically calculating exact edge costs. We present DEA*, an algorithm for problems with expensive edge cost computations. DEA* combines heuristic edge cost evaluations with delayed node expansions, reducing the number of exact edge computations. We formally prove that DEA* is optimal and it is efficient with respect to the number of exact edge cost computations. We empirically evaluate DEA* on multiple-worker routing problems where the exact edge cost is calculated by invoking an external multi-modal journey planning engine. The results demonstrate the effectiveness of our ideas in reducing the computational time and improving the solving ability. In addition, we show the advantages of DEA* in domain-independent planning, where we simulate that accurate edge costs are expensive to compute.

Chasing multiple mobile targets with multiple agents is important in several applications, such as computer games and police chasing scenarios. Existing approaches can compute optimal policies. However, they have a limited scalability, as they implement expensive minimax searches. We introduce a sub-optimal but scalable approach that assigns individual agents to individual targets and that can dynamically re-compute such assignments. We provide a theoretical analysis, including upper bounds on the number of time steps required to solve an instance. In a detailed empirical evaluation on grid maps, our algorithm scales up very convincingly beyond the limits of previous methods. On small problems, where a comparison to a minimax approach is possible, the results demonstrate a good solution quality for our method.

This paper focuses on the overall task of recommending to the chemist candidate molecules (reactants) necessary to synthesize a given target molecule (product), which is a novel application as well as an important step for the chemist to find a synthesis route to generate the product. We formulate this task as a link-prediction problem over a so-called Network of Organic Chemistry (NOC) that we have constructed from 8 million chemical reactions described in the US patent literature between 1976 and 2013. We leverage state-of-the-art factorization algorithms for recommender systems to solve this task. Our empirical evaluation demonstrates that Factorization Machines, trained with chemistry-specific knowledge, outperforms current methods based on similarity of chemical structures.

This document gathers the abstracts for the papers that were presented as part of the Previously Published Paper Track.

This paper presents parallel dovetailing in a distributed-memory environment for exact MAP inference in graphical models. Parallel dovetailing is a simple procedure which performs multiple searches in parallel with different parameter configurations. We evaluate empirically the performance of parallel dovetailing with three state-of-the-art AND/OR search algorithms in solving various MAP inference benchmarks. Our results clearly show that parallel dovetailing is effective, yielding considerable speedups and improving the solving abilities of these state-of-the-art baseline methods.

Optimal multi-modal journey planning under uncertainty is a challenging problem, due in part to an increased branching factor generated by nondeterministic actions. Deterministic search, which ignores all uncertainty, can be much faster, but deterministic plans lack correctness and optimality guarantees in the uncertainty-aware domain.

The paper presents and evaluates the power of parallel search for exact MAP inference in graphical models. We introduce a new parallel shared-memory recursive best-first AND/OR search algorithm, called SPRBFAOO, that explores the search space in a best-first manner while operating with restricted memory. Our experiments show that SPRBFAOO is often superior to the current state-of-the-art sequential AND/OR search approaches, leading to considerable speed-ups (up to 7-fold with 12 threads), especially on hard problem instances.

Knowledge on the Web relies heavily on multi-relational representations, such as RDF and Schema.org. Automatically extracting knowledge from documents and linking existing databases are common approaches to construct multi-relational data. Complementary to such approaches, there is still a strong demand for manually encoding human expert knowledge. For example, human annotation is necessary for constructing a common-sense knowledge base, which stores facts implicitly shared in a community, because such knowledge rarely appears in documents. As human annotation is both tedious and costly, an important research challenge is how to best use limited human resources, whiles maximizing the quality of the resulting dataset. In this paper, we formalize the problem of dataset construction as active learning problems and present the Active Multi-relational Data Construction (AMDC) method. AMDC repeatedly interleaves multi-relational learning and expert input acquisition, allowing us to acquire helpful labels for data construction. Experiments on real datasets demonstrate that our solution increases the number of positive triples by a factor of 2.28 to 17.0, and that the predictive performance of the multi-relational model in AMDC achieves the highest or comparable to the best performance throughout the data construction process.

Recommender systems can be employed to assist users in complex decision making processes. This paper presents a multi-criteria housing recommender system which takes into account not just features of a home, such as rent, but also the transportation links to user specified locations. First, we describe an efficient multi-hop journey time calculator. Second, we introduce a mechanism to find the optimal solutions for multi-criteria evaluation, where a balanced trade-off between the target goals is found. Finally, we present a user study to demonstrate the potential of such a system.

Analyzing functional interactions between small compounds and proteins is indispensable in genomic drug discovery. Since rich information on various compound-protein inter- actions is available in recent molecular databases, strong demands for making best use of such databases require to in- vent powerful methods to help us find new functional compound-protein pairs on a large scale. We present the succinct interval-splitting tree algorithm (SITA) that efficiently per- forms similarity search in databases for compound-protein pairs with respect to both binary fingerprints and real-valued properties. SITA achieves both time and space efficiency by developing the data structure called interval-splitting trees, which enables to efficiently prune the useless portions of search space, and by incorporating the ideas behind wavelet tree, a succinct data structure to compactly represent trees. We experimentally test SITA on the ability to retrieve similar compound-protein pairs/substrate-product pairs for a query from large databases with over 200 million compound- protein pairs/substrate-product pairs and show that SITA performs better than other possible approaches.

The increasing availability of “utility computing” resources such as clouds, grids, and massively parallel shared clusters can provide practically unlimited processing and memory capacity on demand, at some cost per unit of resource usage. This requires a new perspective in the design and evaluation of parallel search algorithms. Previous work in parallel search implicitly assumed ownership of a cluster with a static amount of CPU cores and RAM, and emphasized wallclock runtime. With utility computing resources, trade-offs between performance and monetary costs must be considered. This paper considers dynamically increasing the usage of utility computing resources until a problem is solved. Efficient resource allocation policies are analyzed in comparison with an optimal allocation strategy. We evaluate our iterative allocation strategy by applying it to the HDA* parallel search algorithm. The experimental results validate our theoretical predictions. They show that, in practice, the costs incurred by iterative allocation are reasonably close to an optimal (but a priori unknown) policy, and are significantly better than the worst-case analytical bounds.

The increasing availability of “utility computing” resources such as clouds, grids, and massively parallel shared clusters can provide practically unlimited processing and memory capacity on demand, at some cost per unit of resource usage. This requires a new perspective in the design and evaluation of parallel search algorithms. Previous work in parallel search implicitly assumed ownership of a cluster with a static amount of CPU cores and RAM, and emphasized wallclock runtime. With utility computing resources, trade-offs between performance and monetary costs must be considered. This paper considers dynamically increasing the usage of utility computing resources until a problem is solved. Efficient resource allocation policies are analyzed in comparison with an optimal allocation strategy. We evaluate our iterative allocation strategy by applying it to the HDA* parallel search algorithm. The experimental results validate our theoretical predictions. They show that, in practice, the costs incurred by iterative allocation are reasonably close to an optimal (but a priori unknown) policy, and are significantly better than the worst-case analytical bounds.

Diverse best-first search (DBFS) is a successful algorithm for satisficing planning that is built on top of greedy best-first search and uses stochastic search techniques to select the next ``best'' state that is not always in agreement with the heuristic function. DBFS thus considers a diversity of search directions and avoids search plateaus caused by inaccurate heuristic estimation. However, the scalability of DBFS is still limited by the available memory resource, as in any planning system that uses best-first search at its core. This paper presents the diverse depth-first search (DDFS) algorithm that overcomes the memory bottleneck of DBFS. DDFS requires much less memory than DBFS yet it successfully incorporates DBFS' randomized node selection scheme for effective search diversification. Additionally, DDFS is enhanced with a transposition table to avoid duplicate search effort commonly found in depth-first search. Experimental results in solving problems used in previous International Planning Competitions show that DDFS can solve instances that remain unsolvable by DBFS due to its excessive memory requirements.

Greedy best-first search (GBFS) is a popular and effective algorithm in satisficing planning and is incorporated into high-performance planners. GBFS in planning decides its search direction with automatically generated heuristic functions. However, if the heuristic functions evaluate nodes inaccurately, GBFS may be misled into a valueless search direction, thus resulting in performance degradation. This paper presents a simple but effective algorithm considering a diversity of search directions to avoid the errors of heuristic information. Experimental results in solving a variety of planning problems show that our approach is successful.

Hash Distributed A* (HDA*) is a parallel A* algorithm that is proven to be effective in optimal sequential planning with unit edge costs. HDA* leverages the Zobrist function to almost uniformly distribute and schedule work among processors. This paper evaluates the performance of HDA * in optimal sequence alignment. We observe that with a large number of CPU cores HDA* suffers from an increase of search overhead caused by reexpansions of states in the closed list due to nonuniform edge costs in this domain. We therefore present a new work distribution strategy limiting processors to distribute work, thus increasing the possibility of detecting such duplicate search effort. We evaluate the performance of this approach on a cluster of multi-core machines and show that the approach scales well up to 384 CPU cores.

Heuristic functions play an important role in drastically improving performance of satisficing planners based on greedy best-first search (GBFS). While automatic generation of heuristic functions (e.g., (Hoffmann and Nebel 2001; Helmert 2006)) enables state-of-the-art satisficing planners to solve very complicated planning problems including benchmarks in the International Planning Competitions, accurate evaluations of nodes still remain as a challenging task. Although GBFS is fundamental and powerful in planning, it has an essential drawback when heuristic functions return inaccurate estimates. Assume that a heuristic function underestimates the difficulties of unpromising nodes. Then, since GBFS must expand nodes with small heuristic values first, it spends most of time in searching only unpromising areas and delays moving to the promising part.Previous work tackles this issue by adding a diversity to search, which is an ability in simultaneously exploring different parts of the search space to bypass large errors in heuristic functions. Several algorithms combined with diversity (e.g., K-best-first search (KBFS) in (Felner, Kraus, and Korf 2003)) are empirically shown to be superior to naive best-first search algorithms. However, they still have limited diversity, since they do not immediately expand nodes mistakenly evaluated as very unpromising ones.This paper presents a new technique called diverse best-first search (DBFS), which incorporates a diversity into search in a different way than previous search-based approaches. We show empirical results clearly showing that DBFS is effective in satisficing planning.

Monte-Carlo Tree Search (MCTS) is remarkably successful in two-player games, but parallelizing MCTS has been notoriously difficult to scale well, especially in distributed environments. For a distributed parallel search, transposition-table driven scheduling (TDS) is known to be efficient in several domains. We present a massively parallel MCTS algorithm, that applies the TDS parallelism to the Upper Confidence bound Applied to Trees (UCT) algorithm, which is the most representative MCTS algorithm. To drastically decrease communication overhead, we introduce a reformulation of UCT called Depth-First UCT. The parallel performance of the algorithm is evaluated on clusters using up to 1,200 cores in artificial game-trees. We show that this approach scales well, achieving 740-fold speedups in the best case.

Depth-first proof-number search (df-pn) is powerful AND/OR tree search to solve positions in games. However, df-pn has a notorious problem of infinite loops when applied to domains with repetitions. Df-pn(r) cures it by ignoring proof and disproof numbers that may lead to infinite loops. This paper points out that df-pn(r) has a serious issue of underestimating proof and disproof numbers, while it also suffers from the overestimation problem occurring in directed acyclic graph. It then presents two practical solutions to these problems. While bypassing infinite loops, the threshold controlling algorithm (TCA) solves the underestimation problem by increasing the thresholds of df-pn. The source node detection algorithm (SNDA) detects the cause of overestimation and modifies the computation of proof and disproof numbers. Both TCA and SNDA are implemented on top of df-pn to solve tsume-shogi (checkmating problem in Japanese chess). Results show that df-pn with TCA and SNDA is far superior to df-pn(r). Our tsume-shogi solver is able to solve several difficult positions previously unsolved by any other solvers.

Transposition tables are a well-known method for pruning duplicates in heuristic search. This paper presents a detailed analysis of transposition tables for IDA*. We show that some straightforward implementations of IDA* with transposition tables (IDA*+TT) can result in suboptimal solutions being returned. Furthermore, straightforward implementations of IDA*+TT are not complete. We identify several variants of IDA*+TT which are guaranteed to return the optimal solution, as well as a complete variant. An empirical study shows that IDA*+TT can significantly improve upon the performance of A* in domain-independent planning.

HDA* is a simple, parallelization of A* where work is asynchronously distributed among the nodes by a global hash function. Using up to 1024 cores on a large distributed memory cluster, we evaluate HDA* for a domain-independent planner as well an application-specific 24-puzzle solver. We show that HDA* scales fairly well on a large cluster using up to 1024 cores. Our analysis of the scaling behavior shows that on a cluster of multicore nodes, using only a subset of the available cores and leaving some cores idle can, surprisingly, lead to better results.

Many search algorithms have parameters that need to be tuned to get the best performance. Typically, the parameters are tuned offline, resulting in a generic setting that is supposed to be effective on all problem instances. For suboptimal single-agent search, problem-instance-specific parameter settings can result in substantially reduced search effort. We consider the use of dovetailing as a way to take advantage of this fact. Dovetailing is a procedure that performs search with multiple parameter settings simultaneously. Dovetailing is shown to improve the search speed of weighted IDA* by several orders of magnitude and to generally enhance the performance of weighted RBFS. This procedure is trivially parallelizable and is shown to be an effective form of parallelization for WA* and BULB. In particular, using WA* with parallel dovetailing yields good speedups in the sliding-tile puzzle domain, and increases the number of problems solved when used in an automated planning system.

Parallelizing Monte Carlo tree search (MCTS) has been considered to be a way to improve the strength of Computer Go programs. In this paper, we analyze the performance of two root parallelization methods: the standard strategy based on average selection and our new strategy based on majority voting. As a starting code base, we used Fuego, which is one of the best programs available. Our experimental results with 64 central processing unit (CPU) cores show that majority voting outperforms average selection. Additionally, we show through an extensive analysis that root parallelization has limitations.

The /spl alpha//spl beta/ algorithm for two-player game-tree search has a notorious reputation as being a challenging algorithm for achieving reasonable parallel performance. MTD(f), a new /spl alpha//spl beta/ variant, has become the sequential algorithm of choice for practitioners. Unfortunately, MTD(f) inherits most of the parallel obstacles of /spl alpha//spl beta/, as well as creating new performance hurdles. Transposition-table-driven scheduling (TDS) is a new parallel search algorithm that has proven to be effective in the single-agent (one-player) domain. This paper presents TDSAB, the first time TDS parallelism has been applied to two-player search (the MTD(f) algorithm). Results show that TDSAB gives comparable speedups to that achieved by conventional parallel /spl alpha//spl beta/ algorithms. However, since this is a parallelization of a superior sequential algorithm the results in fact are better. This paper shows that the TDS idea can be extended to more challenging search domains.