Two CS @ ILLINOIS faculty members received the College of Engineering Dean’s Award for Excellence in Research: Grigore Roşu has been recognized by the college and his colleagues as one of four associate professors to have conducted the best research during the past five academic years. Derek Hoiem has been recognized as one of five assistant professors to have conducted the best research during the last academic year.
Roşu is considered one of the foremost researchers in programming languages and formal methods, using mathematical techniques to verify software programs. His approach to programming languages involves two novel ideas: the K language definitional framework and matching logic. K is a rewrite-based framework for defining formal operational semantics of programming languages. A K semantics can be executed and tested as if it was an interpreter. This way, there can be confidence in the correctness of the semantics. Then matching logic uses precisely that semantics, unchanged, for program analysis and verification, without a need to give the language another axiomatic or denotational or dynamic semantics. Matching logic consists of a language-independent proof system to reason about programs in any language that has a rewrite-based operational semantics.
Roşu has also helped develop a new subfield of formal methods called runtime verification, a computing system analysis and execution approach based on extracting information from a running system and using it to detect and avoid errors in computer systems as they execute. He has started a company, Runtime Verifications, Inc., and currently has contracts with Toyota (on runtime verification) and NASA (on semantics-based verification.)
Much of Hoiem’s research in computer vision has focused on the subjects of object recognition and physical scene interpretation. In object recognition, our understanding of an object might be limited by simply naming its category—is the object a “dog” or a “cat” or a “toaster”? Hoiem has explored algorithms to make statements about objects by describing their attributes. This is a revision of object recognition, by which he can demonstrate methods that describe unfamiliar objects, and recognize objects entirely from textual description. He further proposed a system that uses the idea that unfamiliar objects are like familiar objects with similar parts, i.e., a cart is a means of transportation because it has wheels like a bus.
Hoiem is also making exciting strides in the area of physical scene interpretation. Hoiem has shown that approximate estimates of geometry, obtained from a single picture, can enhance other visual processing. Using a single image of a location to build a representation, one can view the location from other directions, insert 3D object models, and so on. This information significantly improved the performance of several modern statistical object recognition methods. He has also shown how to identify the layout of a cluttered room from a single image, and, from this, infer free space in which an agent may move. This layout information also gives significant performance improvements from a furniture recognizer, a way to relight objects inserted into a scene, and a way to reason about where all large items are in a room.
Roşu and Hoiem have made innovative and exceptional contributions in their areas of research, and we congratulate them on this well-deserved recognition from the College of Engineering.
Media inquiries may be directed to:
Associate Director of Communications
moone [at] illinois [dot] edu