Talkarchive
Removal of Ballistocardiogram Artifacts Using Cyclostationary Source Extraction Method
Cardiff School of Engineering, Cardiff University Foad GhaderiDecoding Spatiotemporal fMRI Activation Patterns via Kernels
Utrecht University S. Sinha, G.J. Brouwer, A.P.J.M. Siebes, R. van EeMachine learning Methods for Protein Secondary Structure Prediction and Speaker Verification system
Research Center of Intelligent Signal Processing Tehran, Iran Sepideh BabaeiHybrid Tracking of a Neurosurgical Robotic Safety Trepanation System
Helmholtz-Institute for Biomedical Engineering, RWTH Aachen University Nils GageikA Trajectory-based Approach for Device Independent Gesture Recognition
DAI-Labor/TU-Berlin Mathias WilhelmImproving a locally researched network flow data analyzer by using MapReduce
Jacobs University Bremen Peter NemethLoad analysis of a Apache webserve on a massive parallel cluster / Rendering-cluster / Smart metering systems
BewerberAutomatic assessment of eye-blinking patters for drivers inattention through statistical shape models
Networking Center on Biomedical Research - CIBER-BBN Federico M. Sukno, PhDSignal classification and estimation : Kernel based methods and Bayesian approaches drivers inattention through statistical shape models
LAGIS (Laboratory of Automatics, Data-processing & Signal) -Central School of Lille - France CIBER-BBN Dr. Asma RabaouiDesign and Analysis of a Nonlinear Controller for a Platoon of Nonholonomic Vehicles
Institute of Automatic Controls (IRT), RWTH AACHEN Keun-wook ChungCurrent Works at the Pragmatic Applied Robot Institute (PARI), Daegu Gyeongbuk Institute of Science and Technology (DGIST), Southkorea
Pragmatic Applied Robot Institute, DGIST Southkorea Dr. Jinung AnExploring the Computational Properties of Neural Systems: Highly Accelerated Neuromorphic Hardware Devices
Kirchhoff Institute of Physics Moritz SchillingArbeitsfeld mobile Robotik - von Konstruktion und Kontrolle bis zu Modularität und Akzeptanz
Michael JahnBio-inspired Motion Detection and Local Navigation:Algorithms and FPGA-based implementations
Tim KöhlerAutonome Fahrerlose Transportsysteme - Navigation auf Basis der simultanen Lokalisierung und Kartenerstellung
Britta GriesbachNovel optical coatings with metallic nanoparticles
1.Physikalisches Institut, Universität Stuttgart Martin HövelMore than just summation: how the visual system uses cues to segregate figure and background
Dr. Sirko StraubeA virtual laboratory for robotics research and education
Informatica Instituut, Universiteit van Amsterdam Arnaud VisserAdapting dynamic maps in a home environment
Neuroinformatics and Cognitive Robotics Lab, TU Ilmenau Christopher GaudigAgent-based automatic generation of 3D computer-animation, using the same principles for real robots and wirelessly connecting them into more complex units
Dr. Jan PaulRoboterdarstellungen im deutschsprachigen Comic von 1945 bis 1990
HNF Heinz Nixdorf MuseumsForum GmbH Dr. Stefan SteinRegelung der physikalischen Interaktion zwischen Mensch und einem fluidischen Soft-Roboter am Beispiel eines end-effektor-basierten Bewegungstherapiegerätes für Schulter-Rehabilitation
Mathias JordanAufbau einer robotischen Rendezvous- und Andock-Simulationsanlage für die Verifikation orbitaler und interplanetarer Missionen
Björn WendtDiplomarbeit: Insekten-inspirierte Detektion des optischen Flusses und visuelle Bestimmung der Eigenbewegung
Simon StrübbeModular software integration with the Orocos Realtime Toolkit
Kuleuven University/OpenSourceWorks Peter SoerensBetween Control Engineering and Robotics - the Automation Experience of a Former Master Student
Lanyue JiRobust Feature Extraction of Objects in the Library Scenario of the Robotic System FRIEND
Dennis MrongaChallenges in real-time application development - The I4Copter project
Universität Erlangen Peter UlbrichMaster´s Thesis on the Global Path Planning of Multiple Robotic Manipulators in Static and Time-Varying Environments
Achint AggarwalLearning and Prediction of Autobiographical Episodic Experiences using Sparse Distributed Memory
Sascha Jockel, Dipl.-Inf.Presentation of Master Thesis in Biologically Inspired Neuro-locomotion Control in the Sensorimotor loop of Modular Walking Machine and Previous Work in Mobile Robotics
Vishal PatelReal-time event detection in a time-varied background for UAVs, Self-supervised road detection for UGVs using stereovision and Inter-Aural Time Differentiation based recovery system for AUVs
Punet ChhabraDevelopment of an artificial neural network on a heterogeneous multicore architecture to predict a successful weight loss in obese individuals (Presentation of my diploma thesis)
Markus ScholzKontext Klassifizierung und Schlussfolgerung - unscharfe und ungewisse Informationen in ubiquitären Systemen
Martin Berchtold1. Klassifikation von Melaninstrukturen in Hautaufnahmen, 2. Lokale Positionsbestimmung (LPS)
M. JansenComparison of "Neural Network Techniques" for "Data Approximation" in Wireless Sensor Network
University of Bremen Ramanan MuthuramanEstimating and Using Structure Information for Mobile Robot Exploration
University of Freiburg Kai M. WurmBiologically inspired walking robot Underwater Robot Master Thesis
Universität zu Lübeck und Fachhochschule Lübeck Arora, Vishal PreetDistributed Autonomous Systems
Universität Kassel, Informatik (Robotik, Kommunikation und Kooperation, Sicherheit) Philipp BearStability analysis of time-delayed controller & Stereo robot-vision calibration
Hamburg University of Technology Babu, AjishIntegrating Planning and Control in Mobile Robots and Manipulators
Artificial Intelligence Laboratory, Stanford University Roland PhilippsenFeasibility Study for Onboard Image Processing Algorithms for Lander Navigation.
Electronic Engineering Surrey University Aftab KhanAssistance Systems for the Remote Control of Rovers in Harsh Environments
Technische Informatik, Uni Würzburg Prof. Klaus SchillingAufbau eines Demonstrators zur visuellen Positionsbestimmung mittels Scale Invariant Feature Transform (SIFT) auf Basis einer autonomen, mobilen und echtzeitfähigen Roboterplattform
Technische Univerität Dresden, Institut für Automatisierungstechnik Dipl.-Ing. Andreas VogtVortragsdetails
| Ort: | DFKI Bremen Robotics Innovation Center Robert-Hooke-Str. 5 Konferenzraum 117  Anfahrtsbeschreibung |
| Tobias Jung, University of Texas at Austin |
| Reinforcement Learning with Regularization Networks |
| (Abstract) Reinforcement learning addresses the most intriguing class of problems faced by living creatures and artificial agents alike: that of making (or learning how to make) optimal decisions in a complex world without knowing the exact rules by which the world will respond to the decisions made. Reinforcement learning is a universal methodology that is widely applicable: it is useful for any task that involves taking a sequence of actions and where the outcome of one action influences the utility of subsequent actions. Practical applications abound and range from business and operations research to optimal control and robotics. Reinforcement learning has its roots in classical dynamic programming. Central to this methodology is the concept of a value function, which measures the utility and desirability of states in the world (similar to an evaluation function in board games). The optimal value function is obtained by solving a functional equation (called Bellmans equation). Unfortunately, for all problems of practical interest, we can solve this equation only approximately, borrowing various techniques ranging from function approximation and statistical regression to pattern recognition and linear programming. In this talk I will discuss regularization networks as a modern approach to function approximation in reinforcement learning. Powerful nonparametric methods (such as regularization networks and the related Gaussian process regression) expand the solution directly in the data, thus allowing the parametrization to automatically adapt itself to the complexity of the function we are trying to estimate. Combining regularization networks and least-squares-based policy evaluation, we are able to develop fast and efficient reinforcement learning algorithms that can scale to high-dimensional state-spaces without requiring manual tuning or engineering of basis functions. As applications we consider challenging real-world tasks, such as RoboCup-Keepaway, where we can demonstrate that this solution achieves a superior performance in less time. |