The problem of optimizing unknown costly-to-evaluate functions has been studied for a long time in the context of Bayesian Optimization. Algorithms in this field aim to find the optimizer of the function by asking only a few function evaluations at locations carefully selected based on a posterior model. In this paper, we assume the unknown function is Lipschitz continuous. Leveraging the Lipschitz property, we propose an algorithm with a distinct exploration phase followed by an exploitation phase. The exploration phase aims to select samples that shrink the search space as much as possible. The exploitation phase then focuses on the reduced search space and selects samples closest to the optimizer. Considering the Expected Improvement (EI) as a baseline, we empirically show that the proposed algorithm significantly outperforms EI.

Getting an embodied cognitive system to form a mental model of its world is a challenging prospect. Most AI systems leverage domains defined entirely by the system designers—initial objects, relations, operations, and even search control knowledge are often pre-specified. Building autonomous systems that can bootstrap themselves using a minimal domain definition is a critical research objective of Developmental AI. PAM-P2, a domain agnostic cognitive system, builds a world model using an initial set of user specified primitive actions and homeostatic needs. As sensory datasets are received, an ontology is formed and used to derive situations, events, episodes, solutions, problems, and predictions. An overview of the PAM-P2 architecture and knowledge representation is presented.

We consider the problem of how to plan efficiently in low-level, continuous state spaces with temporally abstract actions (or skills), by constructing abstract representations of the problem suitable for task-level planning.The central question this effort poses is which abstract representations are required to express and evaluate plans composed of sequences of skills. We show that classifiers can be used as a symbolic representation system, and that the ability to represent the preconditions and effects of an agent's skills is both necessary and sufficient for task-level planning.The resulting representations allow a reinforcement learning agent to acquire a symbolic representation appropriate for planning from experience.

We analyze several of the key technical and practical challenges involved in representing activity context across a large variety of knowledge, components, and applications. We present two novel broad methods that enable semantic knowledge capture and interchange, and suggest how they can be used for activity context-awareness. The first is knowledge representation and reasoning (KRR) in Rulelog, an expressively extended form of declarative logic programs that features defeasible higher-order logic formulas yet is computationally tractable, and is a draft dialect of W3C RIF. Rulelog's expressiveness enables representation of exceptions and change, and thus processes, agreements, and policies, e.g., for confidentiality. The second broad method is Textual Logic, an approach to mapping between natural language (text) and logic, where the mapping itself is logic-based. Textual Logic leverages Rulelog's expressiveness to enable relatively rapid text-based authoring of rich knowledge, reducing the knowledge acquisition bottleneck. Together, Rulelog and Textual Logic help address the potential for ontological and KRR Babel that lurks when representing activity context using previous semantic technologies.

CSP solvers usually search a partial assignment search tree.We present a new formalization for CSP solvers, which spansa conceptually different search tree, where each node representssubsets of the original domains for the variables. We experimentwith a simple backtracking algorithm for this searchtree and show that it outperforms a simple backtracking algorithmon the traditional search tree in many cases.

Many AI courses include design and programming projects that provide students with opportunities for experiential learning. Design and programming projects in courses on knowledge-based AI typically explore topics in knowledge, memory, reasoning, and learning. Traditional AI curricula, however, seldom highlight issues of modality of representations, often focusing solely on propositional representations. In this paper, we report on an investigation into learning about representational modality through a series of projects based around geometric analogy problems similar to the Raven’s Progressive Matrices test of intelligence. We conducted this experiment over three years, from Fall 2010 through Fall 2012, in a class on knowledge-based AI. We used the methodology of action research in which the teacher is also the researcher. We discovered that students found these projects motivating, engaging, and challenging, in several cases investing significant time and posting their work online. From our perspective, the projects accomplished the goal of learning about representational modality in addition to knowledge representation and reasoning.

Since their popularity began to rise in the mid-2000s there has been significant growth in the number of multi-core and multi-processor computers available. Knowledge representation systems using logical inference have been slow to embrace this new technology. We present the concept of inference graphs, a natural deduction inference system which scales well on multi-core and multi-processor machines. Inference graphs enhance propositional graphs by treating propositional nodes as tasks which can be scheduled to operate upon messages sent between nodes via the arcs that already exist as part of the propositional graph representation. The use of scheduling heuristics within a prioritized message passing architecture allows inference graphs to perform very well in forward, backward, bi-directional, and focused reasoning. Tests demonstrate the usefulness of our scheduling heuristics, and show significant speedup in both best case and worst case inference scenarios as the number of processors increases.

Story generation is the problem of automatically selecting a sequence of events that meet a set of criteria and can be told as a story. Story generation is knowledge-intensive; traditional story generators rely on a priori defined domain models about fictional worlds, including characters, places, and actions that can be performed. Manually authoring the domain models is costly and thus not scalable. We present a novel class of story generation system that can generate stories in an unknown domain. Our system (a) automatically learns a domain model by crowdsourcing a corpus of narrative examples and (b) generates stories by sampling from the space defined by the domain model. A large-scale evaluation shows that stories generated by our system for a previously unknown topic are comparable in quality to simple stories authored by untrained humans

Answer set programs with external source access may introduce new constants that are not present in the program, which is known as value invention. As naive value invention leads to programs with infinite grounding and answer sets, syntactic safety criteria are imposed on programs. However, traditional criteria are in many cases unnecessarily strong and limit expressiveness. We present liberal domain-expansion (de-) safe programs, a novel generic class of answer set programs with external source access that has a finite grounding and allows for value invention. De-safe programs use so-called term bounding functions as a parameter for modular instantiation with concrete—e.g., syntactic or semantic or both—safety criteria. This ensures extensibility of the approach in the future. We provide concrete instances of the framework and develop an operator that can be used for computing a finite grounding. Finally, we discuss related notions of safety from the literature, and show that our approach is strictly more expressive.

In many logic-based BDI agent programming languages, plan selection involves inferencing over some underlying knowledge representation. While context-sensitive plan selection facilitates the development of flexible, declarative programs, the overhead of evaluating repeated queries to the agent's beliefs and goals can result in poor run time performance. In this paper we present an approach to multi-cycle query caching for logic-based BDI agent programming languages. We extend the abstract performance model presented in (Alechina et al. 2012) to quantify the costs and benefits of caching query results over multiple deliberation cycles. We also present results of experiments with prototype implementations of both single- and multi-cycle caching in three logic-based BDI agent platforms, which demonstrate that significant performance improvements are achievable in practice.