Jeffrey A. Bilmes: Research
It may be best to visit my publications page sorted by research category to get a feeling for the research that I do, but some of the recent projects I have been involved in are briefly summarized below.
- The Graphical Models Toolkit (Binaries, Draft Documentation, and a Tutorial). As of March, 2011, a full new source release and new documentation are coming soon. This is funded by the NSF, and by gifts from Intel and Microsoft.
- Discrete optimization and in particular submodular functions applied to machine learning. Submodular functions are a class of discrete functions that on the one hand are powerful, flexible, and applicable to real problems but on the other hand are not too unrestricted so that hope for efficient algorithms needs to be given up. Submodular functions capture the notion of diminishing returns (an item is worth or costs less when it is considered in a larger context). We have been developing new and applying existing ideas in submodular functions to a variety of applications in machine learning. See the following page for recent publications on submodularity in machine learning. Some of this work is now sponsored by a gift from Google.
- Graphical Models and BioInformatics, joint with Bill Noble. The application of dynamic graphical models to protein identification and to genome detection. There are a bunch of ideas that we are developing using dynamic graphical models, see the relevant publications. This is funded by the NIH.
- Novel methods of speech recognition, including articulatory speech recognition and other aspects of speech are are currently under-exploited in existing speech recognition technology. This also involves dynamic graphical models, and this is joint work with Karen Livescu. This work is funded by the NSF.
- The Vocal Joystick is all about using your voice in non-verbal ways to control computing devices and electro-mechanical devices. The underlying application is for individuals with motor impairments and thus who otherwise would be unable to use a mouse or trackpad. This was funded by the NSF.
- Companions, A European project with the goal of developing intelligent "presence" devices to interact with and help individuals.
- Human Like Speech Processing - building new speech recognition systems by combining statistical modeling and human-like characteristics. This was funded by ONR.
- Sequential human activity recognition, a series of ongoing projects with Dieter Fox. This was funded by DARPA, and now is being funded by Intel.
- Dynamics of Human Social Interaction - designing dynamic models of the evolution of human social networks in-situ, with Tanzeem Choudhury. This was funded by the NSF.
- Intransitive Classifiers and Social Choice - producing methodologies that combine aspects of pattern classification and and social choice theory in a way that exploits intransitivity, with Marina Meila. This was funded by the NSF.