|
|
Jain, Dominik
Researcher
|
Table of ContentsResearch TopicsCurrent Research Projects Teaching Inf. IX(Current Term) Publications |
Before I joined Professor Beetz' Intelligent Autonomous Systems (IAS) group in 2007, I studied computer science and astronautics at TUM, and technology management at the Center for Digital Technology and Management (CDTM) in Munich. I am interested in all things that are of a statistical nature, specifically in finding representations for statistical knowledge that allow not only straightforward knowledge engineering but also support sound and efficient learning and inference mechanisms. Within this field of research, my primary focus is on robotic applications of statistical relational learning.
Outside the lab, you're likely to find me either listening to music or making music (MySpace) - unless I am reading a book by Thomas Bernhard.
Research Topics
- Knowledge Representation
- Probabilistic Reasoning
- Statistical Relational Learning
- Probabilistic Models
- Logical Models
- Parameter Learning
Current Research Projects
- ProbCog : ProbCog designs and implements a statistical relational learning system that supports efficient learning and inference in relational domains, focusing mainly on probabilistic logical models. Based on a common data model, we integrate various representation formalisms and have successfully extended the expressiveness and practical applicability of existing approaches. ProbCog thus provides a coherent probabilistic framework that enables cognitive technical systems to deal with a high degree of uncertainty and complexity in a variety of application contexts, including adaptive plan generation for under-specified tasks and heuristic generation.
Teaching General
Student projects on offer:
- Learning Probabilistic Models of Football Games [Bachelor/Master/Diploma thesis]
The probabilistic interpretation of actions and situations in football games is a highly interesting problem. Statistical relational learning methods may provide the tools that are necessary to answer questions such as: What actions are likely to be chosen by a certain player (or a generic player occupying a certain position) given the various features of a situation; and how will these actions be parameterized? Will the actions be successful? How can we determine from the sequence of actions and situations what is generally going on in a football game (build up, attack phase, counter-attack, etc.)? Can we build qualitative models that rank failed actions by considering the causes of failure appropriately? We have a large collection of semantically interpreted football games (about 20 games from last season's Bayern München games) that could serve as the basis for a number of exciting experiments. - Inference in Relational Mixed Networks [Bachelor/Master/Diploma thesis]
Mixed networks are probabilistic models that combine probabilistic components (i.e. a graphical model such as a Bayesian network) with explicit deterministic constraints. This is a very important class of models because in many real-world domains, we are required to consider probabilistic knowledge (such as preferences) and hard constraints limiting what is considered possible at the same time. Relational mixed networks are templates for the construction of mixed networks, and the networks that are constructed from them thus feature repeated substructes. The goal of the project is to design and implement (approximate) inference algorithms that exploit these repeated substructures. - Bayesian Logic Decision Networks [Bachelor/Master/Diploma thesis]
In order to make decisions in relational domains, an agent may need to consider the effect that a variable number of entities may have on the expected utility of its actions or the situations they lead to. Based on the probabilistic framework of Bayesian logic networks, the student is to develop a representation that explicitly considers utilities. The formalism is to be applied to simple example scenarios demonstrating its representational power. - Bayesian Logic Networks with Soft Logical Constraints [Bachelor/Master/Diploma thesis]
- Entity Resolution in Domestic Environments [Bachelor/Master/Diploma thesis]
A robot in a domestic environment (such as a human kitchen) is often faced with the problem of identifying the objects that it perceives. In particular, it faces the additional challenge of determining whether an object that it currently perceives is likely to be the same as another object it has perceived in the past. The goal of the project is to develop statistical relational models that are capable of solving this problem through probabilistic inference. - Learning about Human Everyday Activities from Body Motions and Object Interactions [Bachelor/Master/Diploma thesis]
- Learning to Choose a Way of Thinking [Bachelor/Master/Diploma thesis]
The performance of inference algorithms for statistical relational models can vary widely depending on the structure of the model and the evidence it is combined with for the task of inference. Different circumstances may lead to greatly reduced performance or even render an algorithm inapplicable that may otherwise be the best choice. The goal of the project is to analyse instantiations of statistical relational models and to propose a set of features with which an informed decision on the choice of an ideal inference procedure could be made using AI methods - and to propose a suitable parameterization of the method.
Past projects:
- Learning Action Models from Trajectory Data (Andreas Barthels, Interdisciplinary Project)
- Maximum a Posteriori Inference in Markov logic Networks for Room Planning (Ralf Wernicke, Bachelor Thesis)
- Applying Markov Logic as a Modelling Language for Weighted CSPs in the Context of Room Planning Optimization (Gregor Wylezich, Bachelor Thesis)
Past courses:
- SS2009: Artificial Intelligence (seminar), Knowledge Representation in Artificial Intelligence (seminar)
- WS2008: Techniques in Artificial Intelligence (tutorial)
- SS2008: Applications of Knowledge-Based Techniques: Knowledge Representation (tutorial)
- WS2007: Techniques in Artificial Intelligence (tutorial)
- SS2007: Discrete Probability Theory (tutorial)
