Dr.rer.nat. Daniel Nyga

daniel_nyga.jpg
Postdoctoral Researcher

Room: 1.77
Tel: –49 -421 218 64010
Fax: –49 -421 218 64047
Mail: nyga(at)cs[dot]uni-bremen[dot]de

About

Daniel Nyga is a postdoctoral researcher at the Institute for Artificial Intelligence (IAI), University of Bremen. Before he joined the IAI Bremen, he studied computer science at the Technical University of Munich, where he received a Bachelor's degree in 2008 and a Master's degree in computer science in 2010. In 2011, he started his PhD supervised by Prof. Michael Beetz at the Intelligent Autonomous Systems group at TUM, which he has finished at the Institute for Artificial Intelligence Bremen with his thesis on the Interpretation of Natural-language Robot Instructions: Probabilistic Knowledge Representation, Learning, and Reasoning (see below).

Dissertation

Abstract– A robot that can be simply told in natural language what to do – this has been one of the ultimate long-standing goals in both Artificial Intelligence and Robotics research. In near-future applications, robotic assistants and companions will have to understand and perform commands such as “set the table for dinner”, “make pancakes for breakfast”, or “cut the pizza into 8 pieces.” Although such instructions are only vaguely formulated, complex sequences of sophisticated and accurate manipulation activities need to be carried out in order to accomplish the respective tasks. The acquisition of knowledge about how to perform these activities from huge collections of natural-language instructions from the Internet has garnered a lot of attention within the last decade. However, natural language is typically massively unspecific, incomplete, ambiguous and vague and thus requires powerful means for interpretation. This work presents PRAC – Probabilistic Action Cores – an interpreter for natural-language instructions which is able to resolve vagueness and ambiguity in natural language and infer missing information pieces that are required to render an instruction executable by a robot. To this end, PRAC formulates the problem of instruction interpretation as a reasoning problem in first-order probabilistic knowledge bases. In particular, the system uses Markov logic networks as a carrier formalism for encoding uncertain knowledge. A novel framework for reasoning about unmodeled symbolic concepts is introduced, which incorporates ontological knowledge from taxonomies and exploits semantically similar relational structures in a domain of discourse. The resulting reasoning framework thus enables more compact representations of knowledge and exhibits strong generalization performance when being learnt from very sparse data. Furthermore, a novel approach for completing directives is presented, which applies semantic analogical reasoning to transfer knowledge collected from thousands of natural-language instruction sheets to new situations. In addition, a cohesive processing pipeline is described that transforms vague and incomplete task formulations into sequences of formally specified robot plans. The system is connected to a plan executive that is able to execute the computed plans in a simulator. Experiments conducted in a publicly accessible, browser-based web interface showcase that PRAC is capable of closing the loop from natural-language instructions to their execution by a robot.

Master's Thesis

Abstract– This thesis investigates boosting algorithms for classifier learning in the presence of imbalanced classes and uneven misclassification costs. In particular, we address the well-known AdaBoost procedure and its extensions for coping with class imbalance, which typically has a negative impact on the classification accuracy regarding the minority class. We give an extensive survey of existing boosting methods for classification and enhancements for tackling the class imbalance problem, including cost-sensitive variants. Regularized boosting methods, which are favourable when dealing with noise and overlapping class distributions, are considered too. We theoretically analyze several strategies for introducing costs and their applicability in the case of imbalance. For one variant (AdaC1) we show that it is instable under certain conditions. We identify drawbacks of an often-cited cost-sensitive boosting algorithm (AdaCost), both theoretically and empirically. We also expose that an algorithm for tackling imbalance without using explicit costs (RareBoost) is a special case of the RealBoost algorithm, a probabilistic variant of AdaBoost. We approve our findings by empirical evaluation on several real-world data sets and academic benchmarks.

Projects

Daniel Nyga's research interests revolve around topics on Artificial Intelligence and Data Science in general, as well as Machine Learning, Data Mining and Pattern Recognition techniques. In particular, he is interested in probabilistic graphical and relational knowledge representation, learning and inference methods, and in applications thereof in natural-language understanding, knowledge processing and robotics.

He was involved in the European FP7 research projects RoboHow and ACAT.

He is the lead developer in the projects pracmln and PRAC.

His GitHub profile can be found here.

Fields of Interest

  • Artificial Intelligence
  • Probability Theory
  • Probabilistic Knowledge Processing
  • Machine Learning
  • Statistical Relational Learning
  • Data Mining/Knowledge Discovery
  • Automated Learning/Understanding of WWW information
  • Natural-language understanding

Teaching

  • AI: Knowledge Acquisition and Representation (WS2017/18) (Lecturer)
  • Master Seminar: Data Mining and Data Analytics (SS2017)
  • AI: Knowledge Acquisition and Representation (WS2016/17) (Lecturer)
  • AI: Knowledge Acquisition and Representation (WS2015/16) (Lecturer)
  • Foundations of Artificial Intelligence (SS2015) (Tutorial/Co-Lecturer)
  • AI: Knowledge Acquisition and Representation (WS2014/15) (Lecturer)
  • Foundations of Artificial Intelligence (SS2014) (Tutorial)
  • AI: Knowledge Acquisition and Representation (WS2013/14) (Co-Lecturer)
  • Foundations of Artificial Intelligence (SS2013) (Tutorial)
  • Technical Cognitive Systems (Lecture & Tutorial, at TUM) (SS2012)
  • Techniques in Artificial Intelligence (Tutorial, at TUM) (WS2011/12)
  • Discrete Probability Theory (Tutorial, at TUM) (SS2011)

Supervised Theses

  • Lifelong Learning of First-order Probabilistic Models for Everyday Robot Manipulation (Master's Thesis, Marc Niehaus)
  • Scaling Probabilistic Completion of Robot Instructions through Semantic Information Retrieval (Master's Thesis, Sebastian Koralewski)
  • To see what no robot has seen before - Recognizing objects based on natural-language descriptions (Master's Thesis, Mareike Picklum)
  • Web-enabled Learning of Models for Word Sense Disambiguation (Bachelor Thesis, Stephan Epping)
  • Grounding Words to Objects: A Joint Model for Co-reference and Entity Resolution Using Markov Logic Networks for Robot Instruction Processing (Diploma Thesis, Florian Meyer)

Publications

Journal Articles and Book Chapters
[1] Daniel Nyga, Michael Beetz, "Cloud-based Probabilistic Knowledge Services for Instruction Interpretation", Chapter in Robotics Research, Springer, vol. 2, pp. 649-664, 2018. [bib]
[2] Michael Beetz, Hagen Langer, Daniel Nyga, "Planning Everyday Manipulation Tasks--Prediction-based Transformation of Structured Activity Descriptions", Chapter in Exploring Cybernetics, Springer, pp. 63-83, 2015. [bib]
[3] Michael Beetz, Ferenc Bálint-Benczédi, Nico Blodow, Christian Kerl, Zoltán-Csaba Márton, Daniel Nyga, Florian Seidel, Thiemo Wiedemeyer, Jan-Hendrik Worch, "RoboSherlock: Unstructured Information Processing Framework for Robotic Perception", In Handling Uncertainty and Networked Structure in Robot Control, Springer International Publishing, Cham, pp. 181-208, 2015. [bib] [pdf] [doi]
Conference Papers
[4] Mihai Pomarlan, Daniel Nyga, Mareike Picklum, Sebastian Koralewski, Michael Beetz, "Deeper Understanding of Vague Instructions through Simulated Execution (Extended Abstract)", In Proceedings of the 2017 International Conference on Autonomous Agents & Multiagent Systems, International Foundation for Autonomous Agents and Multiagent Systems, 2017. [bib]
[5] Daniel Nyga, Mareike Picklum, Sebastian Koralewski, Michael Beetz, "Instruction Completion through Instance-based Learning and Semantic Analogical Reasoning", In International Conference on Robotics and Automation (ICRA), Singapore, 2017. [bib]
[6] Daniel Nyga, Mareike Picklum, Michael Beetz, "What No Robot Has Seen Before -- Probabilistic Interpretation of Natural-language Object Descriptions", In International Conference on Robotics and Automation (ICRA), Singapore, 2017. [bib]
[7] Daniel Nyga, Michael Beetz, "Cloud-based Probabilistic Knowledge Services for Instruction Interpretation", In International Symposium of Robotics Research (ISRR), Sestri Levante (Genoa), Italy, 2015. [bib] [pdf]
[8] Daniel Nyga, Michael Beetz, "Reasoning about Unmodelled Concepts -- Incorporating Class Taxonomies in Probabilistic Relational Models", In Arxiv.org, 2015. Preprint [bib] [pdf]
[9] Gheorghe Lisca, Daniel Nyga, Ferenc Bálint-Benczédi, Hagen Langer, Michael Beetz, "Towards Robots Conducting Chemical Experiments", In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hamburg, Germany, 2015. [bib] [pdf]
[10] Michael Beetz, Ferenc Balint-Benczedi, Nico Blodow, Daniel Nyga, Thiemo Wiedemeyer, Zoltan-Csaba Marton, "RoboSherlock: Unstructured Information Processing for Robot Perception", In IEEE International Conference on Robotics and Automation (ICRA), Seattle, Washington, USA, 2015. Best Service Robotics Paper Award [bib] [pdf]
[11] Daniel Nyga, Ferenc Balint-Benczedi, Michael Beetz, "PR2 Looking at Things: Ensemble Learning for Unstructured Information Processing with Markov Logic Networks", In IEEE International Conference on Robotics and Automation (ICRA), Hong Kong, China, 2014. [bib] [pdf]
[12] Nicholas Hubert Kirk, Daniel Nyga, Michael Beetz, "Controlled Natural Languages for Language Generation in Artificial Cognition", In IEEE International Conference on Robotics and Automation (ICRA), Hong Kong, China, 2014. [bib] [pdf]
[13] Daniel Nyga, Michael Beetz, "Everything Robots Always Wanted to Know about Housework (But were afraid to ask)", In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vilamoura, Portugal, 2012. [bib] [pdf]
[14] Daniel Nyga, Moritz Tenorth, Michael Beetz, "How-Models of Human Reaching Movements in the Context of Everyday Manipulation Activities", In IEEE International Conference on Robotics and Automation (ICRA), Shanghai, China, 2011. [bib] [pdf]
[15] Moritz Tenorth, Daniel Nyga, Michael Beetz, "Understanding and Executing Instructions for Everyday Manipulation Tasks from the World Wide Web", In IEEE International Conference on Robotics and Automation (ICRA), Anchorage, AK, USA, pp. 1486-1491, 2010. [bib] [pdf]
Other Publications
[16] Daniel Nyga, "Interpretation of Natural-language Robot Instructions: Probabilistic Knowledge Representation, Learning, and Reasoning", PhD thesis, University of Bremen, 2017. [bib] [pdf]
[17] Moritz Tenorth, Daniel Nyga, Michael Beetz, "Understanding and Executing Instructions for Everyday Manipulation Tasks from the World Wide Web", Technical report, IAS group, Technische Universität München, Fakultät für Informatik, 2009. [bib]
Powered by bibtexbrowser