Category Archives: Artificial Intelligence

Reinforcement learning to learn the model of the world intrinsically motivated

Todd Hester, Peter Stone, Intrinsically motivated model learning for developing curious robots, Artificial Intelligence, Volume 247, June 2017, Pages 170-186, ISSN 0004-3702, DOI: 10.1016/j.artint.2015.05.002.

Reinforcement Learning (RL) agents are typically deployed to learn a specific, concrete task based on a pre-defined reward function. However, in some cases an agent may be able to gain experience in the domain prior to being given a task. In such cases, intrinsic motivation can be used to enable the agent to learn a useful model of the environment that is likely to help it learn its eventual tasks more efficiently. This paradigm fits robots particularly well, as they need to learn about their own dynamics and affordances which can be applied to many different tasks. This article presents the texplore with Variance-And-Novelty-Intrinsic-Rewards algorithm (texplore-vanir), an intrinsically motivated model-based RL algorithm. The algorithm learns models of the transition dynamics of a domain using random forests. It calculates two different intrinsic motivations from this model: one to explore where the model is uncertain, and one to acquire novel experiences that the model has not yet been trained on. This article presents experiments demonstrating that the combination of these two intrinsic rewards enables the algorithm to learn an accurate model of a domain with no external rewards and that the learned model can be used afterward to perform tasks in the domain. While learning the model, the agent explores the domain in a developing and curious way, progressively learning more complex skills. In addition, the experiments show that combining the agent’s intrinsic rewards with external task rewards enables the agent to learn faster than using external rewards alone. We also present results demonstrating the applicability of this approach to learning on robots.

State of the art and historical background of the classical divergence between AI and robotics

Kanna Rajan, Alessandro Saffiotti, Towards a science of integrated AI and Robotics, Artificial Intelligence, Volume 247, June 2017, Pages 1-9, ISSN 0004-3702, DOI: 10.1016/j.artint.2017.03.003.

The early promise of the impact of machine intelligence did not involve the partitioning of the nascent field of Artificial Intelligence. The founders of AI envisioned the notion of embedded intelligence as being conjoined between perception, reasoning and actuation. Yet over the years the fields of AI and Robotics drifted apart. Practitioners of AI focused on problems and algorithms abstracted from the real world. Roboticists, generally with a background in mechanical and electrical engineering, concentrated on sensori-motor functions. That divergence is slowly being bridged with the maturity of both fields and with the growing interest in autonomous systems. This special issue brings together the state of the art and practice of the emergent field of integrated AI and Robotics, and highlights the key areas along which this current evolution of machine intelligence is heading.

Modelling hierarchical stochastic signals (i.e., decomposable into sub-signals hierarchichally)

Truyen Tran, Dinh Phung, Hung Bui, Svetha Venkatesh, Hierarchical semi-Markov conditional random fields for deep recursive sequential data, Artificial Intelligence, Volume 246, May 2017, Pages 53-85, ISSN 0004-3702, DOI: 10.1016/j.artint.2017.02.003.

We present the hierarchical semi-Markov conditional random field (HSCRF), a generalisation of linear-chain conditional random fields to model deep nested Markov processes. It is parameterised as a conditional log-linear model and has polynomial time algorithms for learning and inference. We derive algorithms for partially-supervised learning and constrained inference. We develop numerical scaling procedures that handle the overflow problem. We show that when depth is two, the HSCRF can be reduced to the semi-Markov conditional random fields. Finally, we demonstrate the HSCRF on two applications: (i) recognising human activities of daily living (ADLs) from indoor surveillance cameras, and (ii) noun-phrase chunking. The HSCRF is capable of learning rich hierarchical models with reasonable accuracy in both fully and partially observed data cases.

Subgraph matching (isomorphism) using GPUs for managing commonsense knowledge, and a short list of other graph problems that have had benefit from multiprocessing

Ha-Nguyen Tran, Erik Cambria, Amir Hussain, Towards GPU-Based Common-Sense Reasoning: Using Fast Subgraph Matching, Cognitive Computation, December 2016, Volume 8, Issue 6, pp 1074–1086, DOI: 10.1007/s12559-016-9418-4.

Common-sense reasoning is concerned with simulating cognitive human ability to make presumptions about the type and essence of ordinary situations encountered every day. The most popular way to represent common-sense knowledge is in the form of a semantic graph. Such type of knowledge, however, is known to be rather extensive: the more concepts added in the graph, the harder and slower it becomes to apply standard graph mining techniques.In this work, we propose a new fast subgraph matching approach to overcome these issues. Subgraph matching is the task of finding all matches of a query graph in a large data graph, which is known to be a non-deterministic polynomial time-complete problem. Many algorithms have been previously proposed to solve this problem using central processing units. Here, we present a new graphics processing unit-friendly method for common-sense subgraph matching, termed GpSense, which is designed for scalable massively parallel architectures, to enable next-generation Big Data sentiment analysis and natural language processing applications.We show that GpSense outperforms state-of-the-art algorithms and efficiently answers subgraph queries on large common-sense graphs.

State of the art of symbolic planning, particularly the one that optimizes some cost, and a novel approach

Álvaro Torralba, Vidal Alcázar, Peter Kissmann, Stefan Edelkamp, Efficient symbolic search for cost-optimal planning, Artificial Intelligence, Volume 242, January 2017, Pages 52-79, ISSN 0004-3702, DOI: 10.1016/j.artint.2016.10.001.

In cost-optimal planning we aim to find a sequence of operators that achieve a set of goals with minimum cost. Symbolic search with Binary Decision Diagrams (BDDs) performs efficient state space exploration in terms of time and memory. This is crucial in optimal settings, in which large parts of the state space must be explored in order to prove optimality. However, the development of accurate heuristics for explicit-state search in recent years have left symbolic search techniques in a secondary place. In this article we propose two orthogonal improvements for symbolic search planning. On the one hand, we analyze and compare different methods for image computation in order to efficiently perform the successor generation on symbolic search. Image computation is the main bottleneck of symbolic search algorithms so an efficient computation is paramount for efficient symbolic search planning. On the other hand, we study how to use state-invariant constraints to prune states in symbolic search. This is essential in regression search but it is yet to be exploited in symbolic search planners. Experiments with symbolic bidirectional uniform-cost search and symbolic A ⁎ search with PDBs show remarkable performance improvements on most IPC benchmark domains. Overall, with the help of our improvements, symbolic bidirectional search outperforms explicit-state search with state-of-the-art heuristics such as LM-cut across many different domains.

Interesting mixture of automated planning with reinforcement learning

Matteo Leonetti, Luca Iocchi, Peter Stone, A synthesis of automated planning and reinforcement learning for efficient, robust decision-making, Artificial Intelligence, Volume 241, 2016, Pages 103-130, ISSN 0004-3702, DOI: 10.1016/j.artint.2016.07.004.

Automated planning and reinforcement learning are characterized by complementary views on decision making: the former relies on previous knowledge and computation, while the latter on interaction with the world, and experience. Planning allows robots to carry out different tasks in the same domain, without the need to acquire knowledge about each one of them, but relies strongly on the accuracy of the model. Reinforcement learning, on the other hand, does not require previous knowledge, and allows robots to robustly adapt to the environment, but often necessitates an infeasible amount of experience. We present Domain Approximation for Reinforcement LearnING (DARLING), a method that takes advantage of planning to constrain the behavior of the agent to reasonable choices, and of reinforcement learning to adapt to the environment, and increase the reliability of the decision making process. We demonstrate the effectiveness of the proposed method on a service robot, carrying out a variety of tasks in an office building. We find that when the robot makes decisions by planning alone on a given model it often fails, and when it makes decisions by reinforcement learning alone it often cannot complete its tasks in a reasonable amount of time. When employing DARLING, even when seeded with the same model that was used for planning alone, however, the robot can quickly learn a behavior to carry out all the tasks, improves over time, and adapts to the environment as it changes.

Learning concepts from graphs in robotics, through first-order logic and discovery of subgraphs, forming arbitrary hierarchies

Ana C. Tenorio-González, Eduardo F. Morales, Automatic discovery of relational concepts by an incremental graph-based representation, Robotics and Autonomous Systems, Volume 83, 2016, Pages 1-14, ISSN 0921-8890, DOI: 10.1016/j.robot.2016.06.012.

Automatic discovery of concepts has been an elusive area in machine learning. In this paper, we describe a system, called ADC, that automatically discovers concepts in a robotics domain, performing predicate invention. Unlike traditional approaches of concept discovery, our approach automatically finds and collects instances of potential relational concepts. An agent, using ADC, creates an incremental graph-based representation with the information it gathers while exploring its environment, from which common sub-graphs are identified. The subgraphs discovered are instances of potential relational concepts which are induced with Inductive Logic Programming and predicate invention. Several concepts can be induced concurrently and the learned concepts can form arbitrarily hierarchies. The approach was tested for learning concepts of polygons, furniture, and floors of buildings with a simulated robot and compared with concepts suggested by users.

A computational cognitive architecture that models emotion

Ron Sun, Nick Wilson, Michael Lynch, Emotion: A Unified Mechanistic Interpretation from a Cognitive Architecture, Cognitive Computation, February 2016, Volume 8, Issue 1, pp 1–14, DOI: 10.1007/s12559-015-9374-4.

This paper reviews a project that attempts to interpret emotion, a complex and multifaceted phenomenon, from a mechanistic point of view, facilitated by an existing comprehensive computational cognitive architecture—CLARION. This cognitive architecture consists of a number of subsystems: the action-centered, non-action-centered, motivational, and metacognitive subsystems. From this perspective, emotion is, first and foremost, motivationally based. It is also action-oriented. It involves many other identifiable cognitive functionalities within these subsystems. Based on these functionalities, we fit the pieces together mechanistically (computationally) within the CLARION framework and capture a variety of important aspects of emotion as documented in the literature.

A formal study of the guarantees that deep neural network offer for classification

R. Giryes, G. Sapiro and A. M. Bronstein, “Deep Neural Networks with Random Gaussian Weights: A Universal Classification Strategy?,” in IEEE Transactions on Signal Processing, vol. 64, no. 13, pp. 3444-3457, July1, 1 2016. DOI: 10.1109/TSP.2016.2546221.

Three important properties of a classification machinery are i) the system preserves the core information of the input data; ii) the training examples convey information about unseen data; and iii) the system is able to treat differently points from different classes. In this paper, we show that these fundamental properties are satisfied by the architecture of deep neural networks. We formally prove that these networks with random Gaussian weights perform a distance-preserving embedding of the data, with a special treatment for in-class and out-of-class data. Similar points at the input of the network are likely to have a similar output. The theoretical analysis of deep networks here presented exploits tools used in the compressed sensing and dictionary learning literature, thereby making a formal connection between these important topics. The derived results allow drawing conclusions on the metric learning properties of the network and their relation to its structure, as well as providing bounds on the required size of the training set such that the training examples would represent faithfully the unseen data. The results are validated with state-of-the-art trained networks.

A new theoretical framework for modeling concepts that allows them to combine reflecting the way humans do, with a good related-work on other concept frameworks in AI

Martha Lewis, Jonathan Lawry, Hierarchical conceptual spaces for concept combination, Artificial Intelligence, Volume 237, August 2016, Pages 204-227, ISSN 0004-3702, DOI: 10.1016/j.artint.2016.04.008.

We introduce a hierarchical framework for conjunctive concept combination based on conceptual spaces and random set theory. The model has the flexibility to account for composition of concepts at various levels of complexity. We show that the conjunctive model includes linear combination as a special case, and that the more general model can account for non-compositional behaviours such as overextension, non-commutativity, preservation of necessity and impossibility of attributes and to some extent, attribute loss or emergence. We investigate two further aspects of human concept use, the conjunction fallacy and the “guppy effect”.