My research focuses on theoretical aspects of machine learning. I am particularly interested in how we can model and exploit special types of interaction with machines to make them learn using less data. This is not only a very interesting and challenging topic from a theoretical point of view, but it also has many potential applications.
Learning user preferences in web search, personalizing advertising in online markets, assisting clinicians in analyzing medical records, developing interactive robots - these and other problems require the use of machine learning, i.e., the design of algorithms that allow computers to learn based on data.
Unfortunately, standard machine learning models and techniques are not suited to exploit the potential benefit that all these problem scenarios have in common, namely the interaction with a potentially co-operative partner. Current machine learning models assume that machines learn from random real-world data. My research team studies models in which machines learn from particularly well-chosen data, as though they were interacting with a co-operative "teacher".
The crucial advantage that the resulting interactive machine learning algorithms have is that they require less data than current algorithms. Intuitively, our research will make intelligent machines exploit the quality of well-chosen data rather than requiring a large quantity of potentially expensive data.
The models and algorithmic techniques that will ultimately arise from this research may provide efficient solutions to complex problems in artificial intelligence - at a lower cost and with less data than is currently possible.
Keywords Relevant to My Research
Theory of Machine Learning ( Algorithmic Learning Theory , theory of co-operative learning, theory of incremental learning, theory of query learning, sample efficiency)
Artificial Intelligence (Search and planning)
Recursion Theory (Inductive inference)
Formal Language Theory (Pattern languages)
Algorithms (Design and analysis of efficient algorithms)