Reinforcement learning for parameter estimation in statistical spoken dialogue systems

Reinforcement techniques have been successfully used to maximise the expected cumulative reward of statistical dialogue systems. Typically, reinforcement learning is used to estimate the parameters of a dialogue policy which selects the system's responses based on the inferred dialogue state. However, the inference of the dialogue state itself depends on a dialogue model which describes the expected behaviour of a user when interacting with the system. Ideally the parameters of this dialogue model should be also optimised to maximise the expected cumulative reward. This article presents two novel reinforcement algorithms for learning the parameters of a dialogue model. First, the Natural Belief Critic algorithm is designed to optimise the model parameters while the policy is kept fixed. This algorithm is suitable, for example, in systems using a handcrafted policy, perhaps prescribed by other design considerations. Second, the Natural Actor and Belief Critic algorithm jointly optimises both the model and the policy parameters. The algorithms are evaluated on a statistical dialogue system modelled as a Partially Observable Markov Decision Process in a tourist information domain. The evaluation is performed with a user simulator and with real users. The experiments indicate that model parameters estimated to maximise the expected reward function provide improved performance compared to the baseline handcrafted parameters. © 2011 Elsevier Ltd. All rights reserved.

Policy optimisation of POMDP-based dialogue systems without state space compression

The partially observable Markov decision process (POMDP) has been proposed as a dialogue model that enables automatic improvement of the dialogue policy and robustness to speech understanding errors. It requires, however, a large number of dialogues to train the dialogue policy. Gaussian processes (GP) have recently been applied to POMDP dialogue management optimisation showing an ability to substantially increase the speed of learning. Here, we investigate this further using the Bayesian Update of Dialogue State dialogue manager. We show that it is possible to apply Gaussian processes directly to the belief state, removing the need for a parametric policy representation. In addition, the resulting policy learns significantly faster while maintaining operational performance. © 2012 IEEE.

Truth Maintenance Systems for Problem Solving

The thesis developed here is that reasoning programs which take care to record the logical justifications for program beliefs can apply several powerful, but simple, domain-independent algorithms to (1) maintain the consistency of program beliefs, (2) realize substantial search efficiencies, and (3) automatically summarize explanations of program beliefs. These algorithms are the recorded justifications to maintain the consistency and well founded basis of the set of beliefs. The set of beliefs can be efficiently updated in an incremental manner when hypotheses are retracted and when new information is discovered. The recorded justifications also enable the pinpointing of exactly whose assumptions which support any particular belief. The ability to pinpoint the underlying assumptions is the basis for an extremely powerful domain-independent backtracking method. This method, called Dependency-Directed Backtracking, offers vastly improved performance over traditional backtracking algorithms.

Fault identification through the combination of symbolic conflict recognition and Markov Chain-aided belief revision

F. Smith and Q. Shen. Fault identification through the combination of symbolic conflict recognition and Markov Chain-aided belief revision. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 34(5):649-663, 2004.

PROSTATIC INTRAEPITHELIAL NEOPLASIA (PIN) - PERFORMANCE OF BAYESIAN BELIEF NETWORK FOR DIAGNOSIS AND GRADING

Prostatic intraepithelial neoplasia (PIN) diagnosis and grading are affected by uncertainties which arise from the fact that almost all knowledge of PIN histopathology is expressed in concepts, descriptive linguistic terms, and words. A Bayesian belief network (BBN) was therefore used to reduce the problem of uncertainty in diagnostic clue assessment, while still considering the dependences between elements in the reasoning sequence. A shallow network was used with an open-tree topology, with eight first-level descendant nodes for the diagnostic clues (evidence nodes), each independently linked by a conditional probability matrix to a root node containing the diagnostic alternatives (decision node). One of the evidence nodes was based on the tissue architecture and the others were based on cell features. The system was designed to be interactive, in that the histopathologist entered evidence into the network in the form of likelihood ratios for outcomes at each evidence node. The efficiency of the network was tested on a series of 110 prostate specimens, subdivided as follows: 22 cases of non-neoplastic prostate or benign prostatic tissue (NP), 22 PINs of low grade (PINlow), 22 PINs of high grade (PINhigh), 22 prostatic adenocarcinomas with cribriform pattern (PACcri), and 22 prostatic adenocarcinomas with large acinar pattern (PAClgac). The results obtained in the benign and malignant categories showed that the belief for the diagnostic alternatives is very high, the values being in general more than 0.8 and often close to 1.0. When considering the PIN lesions, the network classified and graded most of the cases with high certainty. However, there were some cases which showed values less than 0.8 (13 cases out of 44), thus indicating that there are situations in which the feature changes are intermediate between contiguous categories or grades. Discrepancy between morphological grading and the BBN results was observed in four out of 44 PIN cases: one PINlow was classified as PINhigh and three PINhigh were classified as PINlow. In conclusion, the network can grade PlN lesions and differentiate them from other prostate lesions with certainty. In particular, it offers a descriptive classifier which is readily implemented and which allows the use of linguistic, fuzzy variables.

Belief change with noisy sensing in the situation calculus

Situation calculus has been applied widely in arti?cial intelligence to model and reason about actions and changes in dynamic systems. Since actions carried out by agents will cause constant changes of the agents’ beliefs, how to manage
these changes is a very important issue. Shapiro et al. [22] is one of the studies that considered this issue. However, in this framework, the problem of noisy sensing, which often presents in real-world applications, is not considered. As a
consequence, noisy sensing actions in this framework will lead to an agent facing inconsistent situation and subsequently the agent cannot proceed further. In this paper, we investigate how noisy sensing actions can be handled in iterated
belief change within the situation calculus formalism. We extend the framework proposed in [22] with the capability of managing noisy sensings. We demonstrate that an agent can still detect the actual situation when the ratio of noisy sensing actions vs. accurate sensing actions is limited. We prove that our framework subsumes the iterated belief change strategy in [22] when all sensing actions are accurate. Furthermore, we prove that our framework can adequately handle belief introspection, mistaken beliefs, belief revision and belief update even with noisy sensing, as done in [22] with accurate sensing actions only.

Deep Salience: Visual Salience Modeling via Deep Belief Propagation

Visual salience is an intriguing phenomenon observed in biological neural systems. Numerous attempts have been made to model visual salience mathematically using various feature contrasts, either locally or globally. However, these algorithmic models tend to ignore the problem’s biological solutions, in which visual salience appears to arise during the propagation of visual stimuli along the visual cortex. In this paper, inspired by the conjecture that salience arises from deep propagation along the visual cortex, we present a Deep Salience model where a multi-layer model based on successive Markov random fields (sMRF) is proposed to analyze the input image successively through its deep belief propagation. As a result, the foreground object can be automatically separated from the background in a fully unsupervised way. Experimental evaluation on the benchmark dataset validated that our Deep Salience model can consistently outperform eleven state-of-the-art salience models, yielding the higher rates in the precision-recall tests and attaining the best F-measure and mean-square error in the experiments.

Patterns and quality of object-oriented software systems

Lors de ces dix dernières années, le coût de la maintenance des systèmes orientés objets s'est accru jusqu' à compter pour plus de 70% du coût total des systèmes. Cette situation est due à plusieurs facteurs, parmi lesquels les plus importants sont: l'imprécision des spécifications des utilisateurs, l'environnement d'exécution changeant rapidement et la mauvaise qualité interne des systèmes. Parmi tous ces facteurs, le seul sur lequel nous ayons un réel contrôle est la qualité interne des systèmes. De nombreux modèles de qualité ont été proposés dans la littérature pour contribuer à contrôler la qualité. Cependant, la plupart de ces modèles utilisent des métriques de classes (nombre de méthodes d'une classe par exemple) ou des métriques de relations entre classes (couplage entre deux classes par exemple) pour mesurer les attributs internes des systèmes. Pourtant, la qualité des systèmes par objets ne dépend pas uniquement de la structure de leurs classes et que mesurent les métriques, mais aussi de la façon dont celles-ci sont organisées, c'est-à-dire de leur conception, qui se manifeste généralement à travers les patrons de conception et les anti-patrons. Dans cette thèse nous proposons la méthode DEQUALITE, qui permet de construire systématiquement des modèles de qualité prenant en compte non seulement les attributs internes des systèmes (grâce aux métriques), mais aussi leur conception (grâce aux patrons de conception et anti-patrons). Cette méthode utilise une approche par apprentissage basée sur les réseaux bayésiens et s'appuie sur les résultats d'une série d'expériences portant sur l'évaluation de l'impact des patrons de conception et des anti-patrons sur la qualité des systèmes. Ces expériences réalisées sur 9 grands systèmes libres orientés objet nous permettent de formuler les conclusions suivantes: • Contre l'intuition, les patrons de conception n'améliorent pas toujours la qualité des systèmes; les implantations très couplées de patrons de conception par exemple affectent la structure des classes et ont un impact négatif sur leur propension aux changements et aux fautes. • Les classes participantes dans des anti-atrons sont beaucoup plus susceptibles de changer et d'être impliquées dans des corrections de fautes que les autres classes d'un système. • Un pourcentage non négligeable de classes sont impliquées simultanément dans des patrons de conception et dans des anti-patrons. Les patrons de conception ont un effet positif en ce sens qu'ils atténuent les anti-patrons. Nous appliquons et validons notre méthode sur trois systèmes libres orientés objet afin de démontrer l'apport de la conception des systèmes dans l'évaluation de la qualité.

Uptake pathways: the potential of Bayesian belief networks to assist the management, monitoring and evaluation of development-orientated research

The effectiveness of development assistance has come under renewed scrutiny in recent years. In an era of growing economic liberalisation, research organisations are increasingly being asked to account for the use of public funds by demonstrating achievements. However, in the natural resources (NR) research field, conventional economic assessment techniques have focused on quantifying the impact achieved rather understanding the process that delivered it. As a result, they provide limited guidance for planners and researchers charged with selecting and implementing future research. In response, “pathways” or logic models have attracted increased interest in recent years as a remedy to this shortcoming. However, as commonly applied these suffer from two key limitations in their ability to incorporate risk and assess variance from plan. The paper reports the results of a case study that used a Bayesian belief network approach to address these limitations and outlines its potential value as a tool to assist the planning, monitoring and evaluation of development-orientated research.

Argumentative systems in robots

Nowadays, the development of intelligent agents intends to be more refined, using improved architectures and reasoning mechanisms. Revise the beliefs of an agent is also an important subject, due to the consistency that agents should have about their knowledge. In this work we propose deliberative and argumentative agents using Lego Mindstorms robots, Argumentative NXT BDI-like Agents. These agents are built using the notions of the BDI model and they are capable to reason using the DeLP formalism. They update their knowledge base with their perceptions and revise it when necessary. Two variations are presented: the Single Argumentative NXT BDI-like Agent and the MAS Argumentative NXT BDI-like Agent.

Update as Evidence: Belief Expansion

We introduce a justification logic with a novel constructor for evidence terms, according to which the new information itself serves as evidence for believing it. We provide a sound and complete axiomatization for belief expansion and minimal change and explain how the minimality can be graded according to the strength of reasoning. We also provide an evidential analog of the Ramsey axiom.

Intuitionistic common knowledge or belief

Starting off from the usual language of modal logic for multi-agent systems dealing with the agents’ knowledge/belief and common knowledge/belief we define so-called epistemic Kripke structures for intu- itionistic (common) knowledge/belief. Then we introduce corresponding deductive systems and show that they are sound and complete with respect to these semantics.

Framework for risk assessment for maritime transportation systems

Maritime accidents involving ships carrying passengers may pose a high risk with respect to human casualties. For effective risk mitigation, an insight into the process of risk escalation is needed. This requires a proactive approach when it comes to risk modelling for maritime transportation systems. Most of the existing models are based on historical data on maritime accidents, and thus they can be considered reactive instead of proactive. This paper introduces a systematic, transferable and proactive framework estimating the risk for maritime transportation systems, meeting the requirements stemming from the adopted formal definition of risk. The framework focuses on ship-ship collisions in the open sea, with a RoRo/Passenger ship (RoPax) being considered as the struck ship. First, it covers an identification of the events that follow a collision between two ships in the open sea, and, second, it evaluates the probabilities of these events, concluding by determining the severity of a collision. The risk framework is developed with the use of Bayesian Belief Networks and utilizes a set of analytical methods for the estimation of the risk model parameters. The model can be run with the use of GeNIe software package. Finally, a case study is presented, in which the risk framework developed here is applied to a maritime transportation system operating in the Gulf of Finland (GoF). The results obtained are compared to the historical data and available models, in which a RoPax was involved in a collision, and good agreement with the available records is found.