I am currently a postdoc in the Technion - Israel Institute of Technology, in the Department of Industrial Engineering & Management.
Research Interests
Topics: Artificial intelligence, electronic commerce, decision
theory, preference elicitation and representation, mechanism design, auction
theory, multi-agent planning.
Thesis (completed September 2008): Structured Preference Representation and Multiattribute Auctions
Electronic Commerce: My overall interest is in the foundations of Electronic Commerce. Many research areas contribute to the technologic enablement of Electronic Commerce: Artificial Intelligence (AI), Algorithmic Game Theory, Mechanism Design, Information Retrieval, Cryptography, and more. In particular, many applications involve automation of common economic applications, such as auctions, coalition formation, preference elicitation, and recommendation. While these tasks usually involve hard computational problems, modern AI tools can solve many of these relatively efficiently by exploiting structural properties of real-life instances. On the other hand, economists suggest generic models and analyze efficiency for many of these situations. However, a central challenge in electronic commerce remains to apply sophisticated computational tools to enable these tasks, while at the same time maintaining desired economic properties. Example applications include: the design of economically and computationally efficient trading protocols (such as combinatorial and multiattribute auctions), interactive and collaborative recommendation systems, customization, and many more.
Preference Handling: Intelligent systems make decisions to the benefit of users. Therefore, they need to extract and reason about the user's preferences. Though preferences have an equal role in AI decision making as do beliefs and probabilities, the field of preference handling traditionally did not get the level of attention and successful models as probabilities.
There are obvious reasons for the relatively lower interest and weaker results: unlike probabilities, preferences are subjective, and the assessment of results often requires the involvement of human subjects. Furthermore, data is rarely available for evaluation and benchmarking of new technologies. The main challenge is to balance between the need to get accurate preference information, and to limit elicitation burden on the user. This requires preference modeling tools which are compact and geared towards particular tasks; users of simple recommendation website will not be as attentive for a complex elicitation scheme as, for example, users of complex control systems. Furthermore, given a particular representation scheme, preference elicitation queries must be carefully crafted to minimize the burden on the user. I focus mainly on the representation side, seeking compact and informative models which can be employed to reduce the amount of data needed in order to make optimal decisions.
|
