Classificação de padrões através de um comitê de máquinas aprimorado por aprendizagem por reforço

Reinforcement learning is a machine learning technique that, although finding a large number of applications, maybe is yet to reach its full potential. One of the inadequately tested possibilities is the use of reinforcement learning in combination with other methods for the solution of pattern classification problems. It is well documented in the literature the problems that support vector machine ensembles face in terms of generalization capacity. Algorithms such as Adaboost do not deal appropriately with the imbalances that arise in those situations. Several alternatives have been proposed, with varying degrees of success. This dissertation presents a new approach to building committees of support vector machines. The presented algorithm combines Adaboost algorithm with a layer of reinforcement learning to adjust committee parameters in order to avoid that imbalances on the committee components affect the generalization performance of the final hypothesis. Comparisons were made with ensembles using and not using the reinforcement learning layer, testing benchmark data sets widely known in area of pattern classification

Reinforcement Lernen mit Regularisierungsnetzwerken

Die Arbeit behandelt das Problem der Skalierbarkeit von Reinforcement Lernen auf hochdimensionale und komplexe Aufgabenstellungen. Unter Reinforcement Lernen versteht man dabei eine auf approximativem Dynamischen Programmieren basierende Klasse von Lernverfahren, die speziell Anwendung in der Künstlichen Intelligenz findet und zur autonomen Steuerung simulierter Agenten oder realer Hardwareroboter in dynamischen und unwägbaren Umwelten genutzt werden kann. Dazu wird mittels Regression aus Stichproben eine Funktion bestimmt, die die Lösung einer "Optimalitätsgleichung" (Bellman) ist und aus der sich näherungsweise optimale Entscheidungen ableiten lassen. Eine große Hürde stellt dabei die Dimensionalität des Zustandsraums dar, die häufig hoch und daher traditionellen gitterbasierten Approximationsverfahren wenig zugänglich ist. Das Ziel dieser Arbeit ist es, Reinforcement Lernen durch nichtparametrisierte Funktionsapproximation (genauer, Regularisierungsnetze) auf -- im Prinzip beliebig -- hochdimensionale Probleme anwendbar zu machen. Regularisierungsnetze sind eine Verallgemeinerung von gewöhnlichen Basisfunktionsnetzen, die die gesuchte Lösung durch die Daten parametrisieren, wodurch die explizite Wahl von Knoten/Basisfunktionen entfällt und so bei hochdimensionalen Eingaben der "Fluch der Dimension" umgangen werden kann. Gleichzeitig sind Regularisierungsnetze aber auch lineare Approximatoren, die technisch einfach handhabbar sind und für die die bestehenden Konvergenzaussagen von Reinforcement Lernen Gültigkeit behalten (anders als etwa bei Feed-Forward Neuronalen Netzen). Allen diesen theoretischen Vorteilen gegenüber steht allerdings ein sehr praktisches Problem: der Rechenaufwand bei der Verwendung von Regularisierungsnetzen skaliert von Natur aus wie O(n**3), wobei n die Anzahl der Daten ist. Das ist besonders deswegen problematisch, weil bei Reinforcement Lernen der Lernprozeß online erfolgt -- die Stichproben werden von einem Agenten/Roboter erzeugt, während er mit der Umwelt interagiert. Anpassungen an der Lösung müssen daher sofort und mit wenig Rechenaufwand vorgenommen werden. Der Beitrag dieser Arbeit gliedert sich daher in zwei Teile: Im ersten Teil der Arbeit formulieren wir für Regularisierungsnetze einen effizienten Lernalgorithmus zum Lösen allgemeiner Regressionsaufgaben, der speziell auf die Anforderungen von Online-Lernen zugeschnitten ist. Unser Ansatz basiert auf der Vorgehensweise von Recursive Least-Squares, kann aber mit konstantem Zeitaufwand nicht nur neue Daten sondern auch neue Basisfunktionen in das bestehende Modell einfügen. Ermöglicht wird das durch die "Subset of Regressors" Approximation, wodurch der Kern durch eine stark reduzierte Auswahl von Trainingsdaten approximiert wird, und einer gierigen Auswahlwahlprozedur, die diese Basiselemente direkt aus dem Datenstrom zur Laufzeit selektiert. Im zweiten Teil übertragen wir diesen Algorithmus auf approximative Politik-Evaluation mittels Least-Squares basiertem Temporal-Difference Lernen, und integrieren diesen Baustein in ein Gesamtsystem zum autonomen Lernen von optimalem Verhalten. Insgesamt entwickeln wir ein in hohem Maße dateneffizientes Verfahren, das insbesondere für Lernprobleme aus der Robotik mit kontinuierlichen und hochdimensionalen Zustandsräumen sowie stochastischen Zustandsübergängen geeignet ist. Dabei sind wir nicht auf ein Modell der Umwelt angewiesen, arbeiten weitestgehend unabhängig von der Dimension des Zustandsraums, erzielen Konvergenz bereits mit relativ wenigen Agent-Umwelt Interaktionen, und können dank des effizienten Online-Algorithmus auch im Kontext zeitkritischer Echtzeitanwendungen operieren. Wir demonstrieren die Leistungsfähigkeit unseres Ansatzes anhand von zwei realistischen und komplexen Anwendungsbeispielen: dem Problem RoboCup-Keepaway, sowie der Steuerung eines (simulierten) Oktopus-Tentakels.

Lernen mit einer Hybride aus Lernendem Klassifizierendem System und Selbstorganisierender Karte

Im Forschungsgebiet der Künstlichen Intelligenz, insbesondere im Bereich des maschinellen Lernens, hat sich eine ganze Reihe von Verfahren etabliert, die von biologischen Vorbildern inspiriert sind. Die prominentesten Vertreter derartiger Verfahren sind zum einen Evolutionäre Algorithmen, zum anderen Künstliche Neuronale Netze. Die vorliegende Arbeit befasst sich mit der Entwicklung eines Systems zum maschinellen Lernen, das Charakteristika beider Paradigmen in sich vereint: Das Hybride Lernende Klassifizierende System (HCS) wird basierend auf dem reellwertig kodierten eXtended Learning Classifier System (XCS), das als Lernmechanismus einen Genetischen Algorithmus enthält, und dem Wachsenden Neuralen Gas (GNG) entwickelt. Wie das XCS evolviert auch das HCS mit Hilfe eines Genetischen Algorithmus eine Population von Klassifizierern - das sind Regeln der Form [WENN Bedingung DANN Aktion], wobei die Bedingung angibt, in welchem Bereich des Zustandsraumes eines Lernproblems ein Klassifizierer anwendbar ist. Beim XCS spezifiziert die Bedingung in der Regel einen achsenparallelen Hyperquader, was oftmals keine angemessene Unterteilung des Zustandsraumes erlaubt. Beim HCS hingegen werden die Bedingungen der Klassifizierer durch Gewichtsvektoren beschrieben, wie die Neuronen des GNG sie besitzen. Jeder Klassifizierer ist anwendbar in seiner Zelle der durch die Population des HCS induzierten Voronoizerlegung des Zustandsraumes, dieser kann also flexibler unterteilt werden als beim XCS. Die Verwendung von Gewichtsvektoren ermöglicht ferner, einen vom Neuronenadaptationsverfahren des GNG abgeleiteten Mechanismus als zweites Lernverfahren neben dem Genetischen Algorithmus einzusetzen. Während das Lernen beim XCS rein evolutionär erfolgt, also nur durch Erzeugen neuer Klassifizierer, ermöglicht dies dem HCS, bereits vorhandene Klassifizierer anzupassen und zu verbessern. Zur Evaluation des HCS werden mit diesem verschiedene Lern-Experimente durchgeführt. Die Leistungsfähigkeit des Ansatzes wird in einer Reihe von Lernproblemen aus den Bereichen der Klassifikation, der Funktionsapproximation und des Lernens von Aktionen in einer interaktiven Lernumgebung unter Beweis gestellt.

Speech technology in the year 2001.

This paper introduces the session "Technology in the Year 2001" and is the first of four papers dealing with the future of human-machine communication by voice. In looking to the future it is important to recognize both the difficulties of technological forecasting and the frailties of the technology as it exists today--frailties that are manifestations of our limited scientific understanding of human cognition. The technology to realize truly advanced applications does not yet exist and cannot be supported by our presently incomplete science of speech. To achieve this long-term goal, the authors advocate a fundamental research program using a cybernetic approach substantially different from more conventional synthetic approaches. In a cybernetic approach, feedback control systems will allow a machine to adapt to a linguistically rich environment using reinforcement learning.

Statistical dialog management for health interventions

Research endeavors on spoken dialogue systems in the 1990s and 2000s have led to the deployment of commercial spoken dialogue systems (SDS) in microdomains such as customer service automation, reservation/booking and question answering systems. Recent research in SDS has been focused on the development of applications in different domains (e.g. virtual counseling, personal coaches, social companions) which requires more sophistication than the previous generation of commercial SDS. The focus of this research project is the delivery of behavior change interventions based on the brief intervention counseling style via spoken dialogue systems. ^ Brief interventions (BI) are evidence-based, short, well structured, one-on-one counseling sessions. Many challenges are involved in delivering BIs to people in need, such as finding the time to administer them in busy doctors' offices, obtaining the extra training that helps staff become comfortable providing these interventions, and managing the cost of delivering the interventions. Fortunately, recent developments in spoken dialogue systems make the development of systems that can deliver brief interventions possible. ^ The overall objective of this research is to develop a data-driven, adaptable dialogue system for brief interventions for problematic drinking behavior, based on reinforcement learning methods. The implications of this research project includes, but are not limited to, assessing the feasibility of delivering structured brief health interventions with a data-driven spoken dialogue system. Furthermore, while the experimental system focuses on harmful alcohol drinking as a target behavior in this project, the produced knowledge and experience may also lead to implementation of similarly structured health interventions and assessments other than the alcohol domain (e.g. obesity, drug use, lack of exercise), using statistical machine learning approaches. ^ In addition to designing a dialog system, the semantic and emotional meanings of user utterances have high impact on interaction. To perform domain specific reasoning and recognize concepts in user utterances, a named-entity recognizer and an ontology are designed and evaluated. To understand affective information conveyed through text, lexicons and sentiment analysis module are developed and tested.^

Statistical Dialog Management for Health Interventions

Research endeavors on spoken dialogue systems in the 1990s and 2000s have led to the deployment of commercial spoken dialogue systems (SDS) in microdomains such as customer service automation, reservation/booking and question answering systems. Recent research in SDS has been focused on the development of applications in different domains (e.g. virtual counseling, personal coaches, social companions) which requires more sophistication than the previous generation of commercial SDS. The focus of this research project is the delivery of behavior change interventions based on the brief intervention counseling style via spoken dialogue systems. Brief interventions (BI) are evidence-based, short, well structured, one-on-one counseling sessions. Many challenges are involved in delivering BIs to people in need, such as finding the time to administer them in busy doctors' offices, obtaining the extra training that helps staff become comfortable providing these interventions, and managing the cost of delivering the interventions. Fortunately, recent developments in spoken dialogue systems make the development of systems that can deliver brief interventions possible. The overall objective of this research is to develop a data-driven, adaptable dialogue system for brief interventions for problematic drinking behavior, based on reinforcement learning methods. The implications of this research project includes, but are not limited to, assessing the feasibility of delivering structured brief health interventions with a data-driven spoken dialogue system. Furthermore, while the experimental system focuses on harmful alcohol drinking as a target behavior in this project, the produced knowledge and experience may also lead to implementation of similarly structured health interventions and assessments other than the alcohol domain (e.g. obesity, drug use, lack of exercise), using statistical machine learning approaches. In addition to designing a dialog system, the semantic and emotional meanings of user utterances have high impact on interaction. To perform domain specific reasoning and recognize concepts in user utterances, a named-entity recognizer and an ontology are designed and evaluated. To understand affective information conveyed through text, lexicons and sentiment analysis module are developed and tested.

SEQUENTIAL DECISIONS AND PREDICTIONS IN NATURAL LANGUAGE PROCESSING

Natural language processing has achieved great success in a wide range of ap- plications, producing both commercial language services and open-source language tools. However, most methods take a static or batch approach, assuming that the model has all information it needs and makes a one-time prediction. In this disser- tation, we study dynamic problems where the input comes in a sequence instead of all at once, and the output must be produced while the input is arriving. In these problems, predictions are often made based only on partial information. We see this dynamic setting in many real-time, interactive applications. These problems usually involve a trade-off between the amount of input received (cost) and the quality of the output prediction (accuracy). Therefore, the evaluation considers both objectives (e.g., plotting a Pareto curve). Our goal is to develop a formal understanding of sequential prediction and decision-making problems in natural language processing and to propose efficient solutions. Toward this end, we present meta-algorithms that take an existent batch model and produce a dynamic model to handle sequential inputs and outputs. Webuild our framework upon theories of Markov Decision Process (MDP), which allows learning to trade off competing objectives in a principled way. The main machine learning techniques we use are from imitation learning and reinforcement learning, and we advance current techniques to tackle problems arising in our settings. We evaluate our algorithm on a variety of applications, including dependency parsing, machine translation, and question answering. We show that our approach achieves a better cost-accuracy trade-off than the batch approach and heuristic-based decision- making approaches. We first propose a general framework for cost-sensitive prediction, where dif- ferent parts of the input come at different costs. We formulate a decision-making process that selects pieces of the input sequentially, and the selection is adaptive to each instance. Our approach is evaluated on both standard classification tasks and a structured prediction task (dependency parsing). We show that it achieves similar prediction quality to methods that use all input, while inducing a much smaller cost. Next, we extend the framework to problems where the input is revealed incremen- tally in a fixed order. We study two applications: simultaneous machine translation and quiz bowl (incremental text classification). We discuss challenges in this set- ting and show that adding domain knowledge eases the decision-making problem. A central theme throughout the chapters is an MDP formulation of a challenging problem with sequential input/output and trade-off decisions, accompanied by a learning algorithm that solves the MDP.

Technologies for urban and rural internet of things

Nowadays, application domains such as smart cities, agriculture or intelligent transportation, require communication technologies that combine long transmission ranges and energy efficiency to fulfill a set of capabilities and constraints to rely on. In addition, in recent years, the interest in Unmanned Aerial Vehicles (UAVs) providing wireless connectivity in such scenarios is substantially increased thanks to their flexible deployment. The first chapters of this thesis deal with LoRaWAN and Narrowband-IoT (NB-IoT), which recent trends identify as the most promising Low Power Wide Area Networks technologies. While LoRaWAN is an open protocol that has gained a lot of interest thanks to its simplicity and energy efficiency, NB-IoT has been introduced from 3GPP as a radio access technology for massive machine-type communications inheriting legacy LTE characteristics. This thesis offers an overview of the two, comparing them in terms of selected performance indicators. In particular, LoRaWAN technology is assessed both via simulations and experiments, considering different network architectures and solutions to improve its performance (e.g., a new Adaptive Data Rate algorithm). NB-IoT is then introduced to identify which technology is more suitable depending on the application considered. The second part of the thesis introduces the use of UAVs as flying Base Stations, denoted as Unmanned Aerial Base Stations, (UABSs), which are considered as one of the key pillars of 6G to offer service for a number of applications. To this end, the performance of an NB-IoT network are assessed considering a UABS following predefined trajectories. Then, machine learning algorithms based on reinforcement learning and meta-learning are considered to optimize the trajectory as well as the radio resource management techniques the UABS may rely on in order to provide service considering both static (IoT sensors) and dynamic (vehicles) users. Finally, some experimental projects based on the technologies mentioned so far are presented.

Análise funcional comparativa de algoritmos de aprendizagem por reforço

De entre todos os paradigmas de aprendizagem actualmente identificados, a Aprendizagem por Reforço revela-se de especial interesse e aplicabilidade nos inúmeros processos que nos rodeiam: desde a solitária sonda que explora o planeta mais remoto, passando pelo programa especialista que aprende a apoiar a decisão médica pela experiencia adquirida, até ao cão de brincar que faz as delícias da criança interagindo com ela e adaptando-se aos seus gostos, e todo um novo mundo que nos rodeia e apela crescentemente a que façamos mais e melhor nesta área. Desde o aparecimento do conceito de aprendizagem por reforço, diferentes métodos tem sido propostos para a sua concretização, cada um deles abordando aspectos específicos. Duas vertentes distintas, mas complementares entre si, apresentam-se como características chave do processo de aprendizagem por reforço: a obtenção de experiência através da exploração do espaço de estados e o aproveitamento do conhecimento obtido através dessa mesma experiência. Esta dissertação propõe-se seleccionar alguns dos métodos propostos mais promissores de ambas as vertentes de exploração e aproveitamento, efectuar uma implementação de cada um destes sobre uma plataforma modular que permita a simulação do uso de agentes inteligentes e, através da sua aplicação na resolução de diferentes configurações de ambientes padrão, gerar estatísticas funcionais que permitam inferir conclusões que retractem entre outros aspectos a sua eficiência e eficácia comparativas em condições específicas.

MASCEM: electricity markets simulation with strategic agents

Electricity markets are complex environments, involving numerous entities trying to obtain the best advantages and profits while limited by power-network characteristics and constraints.1 The restructuring and consequent deregulation of electricity markets introduced a new economic dimension to the power industry. Some observers have criticized the restructuring process, however, because it has failed to improve market efficiency and has complicated the assurance of reliability and fairness of operations. To study and understand this type of market, we developed the Multiagent Simulator of Competitive Electricity Markets (MASCEM) platform based on multiagent simulation. The MASCEM multiagent model includes players with strategies for bid definition, acting in forward, day-ahead, and balancing markets and considering both simple and complex bids. Our goal with MASCEM was to simulate as many market models and player types as possible. This approach makes MASCEM both a short- and mediumterm simulation as well as a tool to support long-term decisions, such as those taken by regulators. This article proposes a new methodology integrated in MASCEM for bid definition in electricity markets. This methodology uses reinforcement learning algorithms to let players perceive changes in the environment, thus helping them react to the dynamic environment and adapt their bids accordingly.

Integração de reacção e deliberação em agentes inteligentes

Trabalho Final de Mestrado para obtenção do grau de Mestre em Engenharia de Redes de Comunicação e Multimédia

Podcast educacional aplicado às Ciências Naturais: o sistema nervoso

O processo ensino-aprendizagem das ciências naturais é desafiado para se adaptar às necessidades e estilo de vida atual dos alunos. Este desafio, para os docentes, é a inovação das estratégias pedagógicas orientadas para a mudança conceptual. As escolas podem estar equipadas com vários recursos tecnológicos, mas a sua integração no processo ensino-aprendizagem, como práticas inovadoras e promotoras de uma utilização efetiva por parte dos discentes, é ainda limitada. Cabe ao professor do século XXI incrementar a utilização das Tecnologias de Inovação e Comunicação nas suas práticas letivas permitindo aos alunos, que temos na sala de aula, e que nasceram na era digital, uma maior motivação e potenciação das suas aprendizagens. O conteúdo “O Sistema Nervoso”, só abordado no 9º ano de escolaridade, temática incluída no tema organizador “Viver Melhor na Terra”, envolve termos científicos complexos, sendo sempre um grande desafio para os docentes a sua abordagem em contexto educativo. Neste contexto enunciou-se o seguinte problema de estudo: o podcast aplicado como auxiliar de aprendizagem do segmento de conteúdo “O Sistema Nervoso”, levará a uma melhoria dos resultados escolares dos alunos do 9º ano? Para este estudo utilizou-se uma metodologia mista, envolvendo 19 alunos, permitindo a utilização de vários métodos e diferentes formas de obter dados e os analisar. É um estudo de caso e os resultados mostram o impacto da utilização de um podcast educacional como auxiliar de aprendizagem de um conteúdo, “O Sistema Nervoso”. O ensino-aprendizagem deixou de estar confinado à sala de aula, podendo ocorrer em qualquer lugar e a todo o momento, de acordo com a escolha dos alunos e respeitando o ritmo de aprendizagem de cada um. A audição do podcast revelou-se um instrumento facilitador do trabalho autónomo, já que é um reforço pedagógico e/ou estratégia de ensino diferenciado, funcionando como um auxiliar esclarecedor, através do qual os alunos clarificam as suas dúvidas, o que não conseguiriam fazer sozinhos. Permite, ainda, recriar o ambiente de sala de aula na explicação dos conteúdos.

Emotional modulation of memory encoding and performance monitoring

Performance monitoring, ERN, CRN, Pe, Memory, Llist learning, Emotion, IAPS, N2, Reinforcement Learning Hypothesis, Conflict Monitoring Hypothesis

Are agent-based simulations robust? The wholesale electricity trading case

Agent-based computational economics is becoming widely used in practice. This paperexplores the consistency of some of its standard techniques. We focus in particular on prevailingwholesale electricity trading simulation methods. We include different supply and demandrepresentations and propose the Experience-Weighted Attractions method to include severalbehavioural algorithms. We compare the results across assumptions and to economic theorypredictions. The match is good under best-response and reinforcement learning but not underfictitious play. The simulations perform well under flat and upward-slopping supply bidding,and also for plausible demand elasticity assumptions. Learning is influenced by the number ofbids per plant and the initial conditions. The overall conclusion is that agent-based simulationassumptions are far from innocuous. We link their performance to underlying features, andidentify those that are better suited to model wholesale electricity markets.

The habenula encodes negative motivational value associated with primary punishment in humans.

Learning what to approach, and what to avoid, involves assigning value to environmental cues that predict positive and negative events. Studies in animals indicate that the lateral habenula encodes the previously learned negative motivational value of stimuli. However, involvement of the habenula in dynamic trial-by-trial aversive learning has not been assessed, and the functional role of this structure in humans remains poorly characterized, in part, due to its small size. Using high-resolution functional neuroimaging and computational modeling of reinforcement learning, we demonstrate positive habenula responses to the dynamically changing values of cues signaling painful electric shocks, which predict behavioral suppression of responses to those cues across individuals. By contrast, negative habenula responses to monetary reward cue values predict behavioral invigoration. Our findings show that the habenula plays a key role in an online aversive learning system and in generating associated motivated behavior in humans.