Artificial Intelligence最新文献

Much of the social choice literature examines direct voting systems, in which voters submit their ranked preferences over candidates and a voting rule picks a winner. Real-world elections and decision-making processes are often more complex and involve multiple stages. For instance, one popular voting system filters candidates through primaries: first, voters affiliated with each political party vote over candidates of their own party and the voting rule picks a set of candidates, one from each party, who then compete in a general election.

We present a model to analyze such multi-stage elections, and conduct what is, to the best of our knowledge, the first quantitative comparison of the direct and primary voting systems in terms of the quality of the elected candidate, using the metric of distortion, which attempts to quantify how far from the optimal winner is the actual winner of an election. Our main theoretical result is that voting rules (which are independent of party affiliations, of course) are guaranteed to perform in the primary system within a constant factor of the direct, single stage setting. Surprisingly, the converse does not hold: we show settings in which there exist voting rules that perform significantly better under the primary system than under the direct system. Using simulations, we see that plurality benefits significantly from using a primary system over a direct one, while Condorcet-consistent rules do not.

Multi-Agent Path Finding (MAPF) is the problem of planning paths for agents to reach their targets from their start locations, such that the agents do not collide while executing the plan. In many settings, the plan (or a digest thereof) is conveyed to a supervising entity, e.g., for confirmation before execution, for a report, etc. In such cases, we wish to convey that the plan is collision-free with minimal amount of information. To this end, we propose an explanation scheme for MAPF. The scheme decomposes a plan into segments such that within each segment, the agents' paths are disjoint. We can then convey the plan whilst convincing that it is collision-free, using a small number of frames (dubbed an explanation). We can also measure the simplicity of a plan by the number of segments required for the decomposition. We study the complexity of algorithmic problems that arise by the explanation scheme and the tradeoff between the length (makespan) of a plan and its minimal decomposition. We also introduce two centralized (i.e. runs on a single CPU with full knowledge of the multi-agent system) algorithms for planning with explanations. One is based on a coupled search algorithm similar to A^⁎, and the other is a decoupled method based on Conflict-Based Search (CBS). We refer to the latter as Explanation-Guided CBS (XG-CBS), which uses a low-level search for individual agents and maintains a high-level conflict tree to guide the low-level search to avoid collisions as well as increasing the number of segments. We propose four approaches to the low-level search of XG-CBS by modifying A^⁎ for explanations and analyze their effects on the completeness of XG-CBS. Finally, we highlight important aspects of the proposed explanation scheme in various MAPF problems and empirically evaluate the performance of the proposed planning algorithms in a series of benchmark problems.

Lifting probabilistic graphical models and developing lifted inference algorithms aim to use higher level groups of random variables instead of individual instances. In the past, many inference algorithms for discrete probabilistic graphical models have been lifted. Lifting continuous probabilistic graphical models has played a minor role. Since many real-world applications involve continuous random variables, this article turns its focus to lifting approaches for Gaussian Bayesian networks. Specifically, we present algorithms for constructing a lifted joint distribution for scenarios of sequences of overlapping and non-overlapping logical variables. We present operations that work in a fully lifted way including addition, multiplication, and inversion. We present how the operations can be used for lifted query answering algorithms and extend the existing query answering algorithm by a new way of evidence handling. The new way of evidence handling groups evidence that has the same effect on its neighboring variables in cases of partial overlap between the logical-variable sequences. In the theoretical complexity analysis and the experimental evaluation, we show under which conditions the existing lifted approach and the new lifted approach including evidence grouping lead to the most time savings compared to the grounded approach.

In this work, we introduce and analyze an approach to knowledge transfer from one collection of facts to another without the need for entity or relation matching. The method works for both canonicalized knowledge bases and uncanonicalized or open knowledge bases, i.e., knowledge bases where more than one copy of a real-world entity or relation may exist. The main contribution is a method that can make use of large-scale pre-training on facts, which were collected from unstructured text, to improve predictions on structured data from a specific domain. The introduced method is most impactful on small datasets such as ReVerb20k, where a 6% absolute increase of mean reciprocal rank and 65% relative decrease of mean rank over the previously best method was achieved, despite not relying on large pre-trained models like Bert. To understand the obtained pre-trained models better, we then introduce a novel dataset for the analysis of pre-trained models for Open Knowledge Base Completion, called Doge (Diagnostics of Open knowledge Graph Embeddings). It consists of 6 subsets and is designed to measure multiple properties of a pre-trained model: robustness against synonyms, ability to perform deductive reasoning, presence of gender stereotypes, consistency with reverse relations, and coverage of different areas of general knowledge. Using the introduced dataset, we show that the existing OKBC models lack consistency in presence of synonyms and inverse relations and are unable to perform deductive reasoning. Moreover, their predictions often align with gender stereotypes, which persist even when presented with counterevidence. We additionally investigate the role of pre-trained word embeddings and demonstrate that avoiding biased word embeddings is not a sufficient measure to prevent biased behavior of OKBC models.

Despite the great theoretical advancements in the area of Belief Revision, there has been limited success in terms of implementations. One of the hurdles in implementing revision operators is that their specification (let alone their computation), requires substantial resources. On the other hand, implementing a specific revision operator, like Dalal's operator, would be of limited use. In this paper we generalise Dalal's construction, defining a whole family of concrete revision operators, called Parametrised Difference revision operators or PD operators for short. This family is wide enough to cover a wide range of different applications, and at the same time it is easy to represent. In addition to its semantic definition, we characterise the family of PD operators axiomatically (including a characterisation specifically for Dalal's operator), we prove its' compliance with Parikh's relevance-sensitive postulate (P), we study its computational complexity, and discuss its benefits for belief revision implementations.

Multi-agent path finding (MAPF) is a classical NP-hard problem that considers planning collision-free paths for multiple agents simultaneously. A MAPF problem is typically solved via addressing a sequence of single-agent path finding subproblems in which well-studied algorithms such as ${\text{A}}^{⁎}$ are applicable. Existing methods based on this idea, however, rely on an exhaustive search and therefore only have asymptotic performance guarantees. In this article, we provide a modeling paradigm that converts a MAPF problem into a stochastic process and adopts a confidence bound based rule for finding the optimal state transition strategy. A randomized algorithm is proposed to solve this stochastic process, which combines ideas from conflict based search and Monte Carlo tree search. We show that the proposed method is almost surely optimal while enjoying non-asymptotic performance guarantees. In particular, the proposed method can, after solving N single-agent subproblems, produce a feasible solution with suboptimality bounded by $O(1/\sqrt{N})$. The theoretical results are verified by several numerical experiments based on grid maps.

Temporal Knowledge Graph (TKG) is an extension of traditional Knowledge Graph (KG) that incorporates the dimension of time. Reasoning on TKGs is a crucial task that aims to predict future facts based on historical occurrences. The key challenge lies in uncovering structural dependencies within historical subgraphs and temporal patterns. Most existing approaches model TKGs relying on entity modeling, as nodes in the graph play a crucial role in knowledge representation. However, the real-world scenario often involves an extensive number of entities, with new entities emerging over time. This makes it challenging for entity-dependent methods to cope with extensive volumes of entities, and effectively handling newly emerging entities also becomes a significant challenge. Therefore, we propose Temporal Inductive Path Neural Network (TiPNN), which models historical information in an entity-independent perspective. Specifically, TiPNN adopts a unified graph, namely history temporal graph, to comprehensively capture and encapsulate information from history. Subsequently, we utilize the defined query-aware temporal paths on a history temporal graph to model historical path information related to queries for reasoning. Extensive experiments illustrate that the proposed model not only attains significant performance enhancements but also handles inductive settings, while additionally facilitating the provision of reasoning evidence through history temporal graphs.

Discrete random variables are essential ingredients in various artificial intelligence problems. These include the estimation of the probability of missing the deadline in a series-parallel schedule and the assignment of suppliers to tasks in a project in a manner that maximizes the probability of meeting the overall project deadline. The solving of such problems involves repetitive operations, such as summation, over random variables. However, these computations are NP-hard. Therefore, we explore techniques and methods for approximating random variables with a given support size and minimal Kolmogorov distance. We examine both the general problem of approximating a random variable and a one-sided version in which over-approximation is allowed but not under-approximation. We propose several algorithms and evaluate their performance through computational complexity analysis and empirical evaluation. All the presented algorithms are optimal in the sense that given an input random variable and a requested support size, they return a new approximated random variable with the requested support size and minimal Kolmogorov distance from the input random variable. Our approximation algorithms offer useful estimations of probabilities in situations where exact computations are not feasible due to NP-hardness complexity.

The Reinforcement Learning (RL) framework offers a general paradigm for constructing autonomous agents that can make effective decisions when solving tasks. An important area of study within the field of RL is transfer learning, where an agent utilizes knowledge gained from solving previous tasks to solve a new task more efficiently. While the notion of transfer learning is conceptually appealing, in practice, not all RL representations are amenable to transfer learning. Moreover, much of the research on transfer learning in RL is purely empirical. Previous research has shown that object-oriented representations are suitable for the purposes of transfer learning with theoretical efficiency guarantees. Such representations leverage the notion of object classes to learn lifted rules that apply to grounded object instantiations. In this paper, we extend previous research on object-oriented representations and introduce two formalisms: the first is based on deictic predicates, and is used to learn a transferable transition dynamics model; the second is based on propositions, and is used to learn a transferable reward dynamics model. In addition, we extend previously introduced efficient learning algorithms for object-oriented representations to our proposed formalisms. Our frameworks are then combined into a single efficient algorithm that learns transferable transition and reward dynamics models across a domain of related tasks. We illustrate our proposed algorithm empirically on an extended version of the Taxi domain, as well as the more difficult Sokoban domain, showing the benefits of our approach with regards to efficient learning and transfer.

A distributed constraint optimization problem (DCOP) is a framework to model multi-agent coordination problems. Asynchronous distributed optimization (ADOPT) is a well-known complete DCOP algorithm, and many of its variants have been proposed over the last decade. It is considered proven that ADOPT-based algorithms have the key properties of termination and optimality, which guarantee that the algorithms terminate in a finite time and obtain an optimal solution, respectively. In this paper, we present counterexamples to the termination and optimality of ADOPT-based algorithms. They are classified into three types, at least one of which exists in each of ADOPT and eight of its variants that we analyzed. In other words, the algorithms may potentially not terminate or terminate with a suboptimal solution. Furthermore, we show that the bounded-error approximation of ADOPT, which enables the algorithm to terminate faster with the quality of the solution guaranteed within a predefined error bound, also suffers from flaws. Additionally, we propose an amended version of ADOPT that avoids the flaws in existing algorithms and prove that it has the properties of termination and optimality.