Xerox Undergraduate Research Fellows Program
http://www.rochester.edu/college/kearnscenter/xeroxfellowsmain.html Program description
http://www.rochester.edu/college/kearnscenter/XeroxApply.html Fellowship application
Florian's call for ideas
So, here's a thought. If you have ideas on how to involve smart CS undergraduates in little programming projects in the lab, preferably in order to develop tools of general interest or to improve on existing tools (think web-experiments, corpus tools, visualization, etc.), we could see whether we can tap into this fund. In order to /find/ the smart undergraduates, we could have an open-lab day for CS undergraduates (I could advertise that to CS).
Ideas
Computational/psycholinguistic research
Marriage of real time language processing research and CS research via eye tracking. James Allen's group already does some of this, but my intuition is that there's plenty of room to explore more ideas. Eye trackers are used in industry for usability studies and design of visual interfaces. How can this usage be combined with the language processing work we do? --AustinFrank
Collection of reaction time data over a network. Combine client and server side javascript to get timing data from a user's machine and aggregate it on the server. Could do self-paced reading, lexical decision, or collect production data in the form of keypress latencies during typing. We get a new way to gather psycholinguistic data. I can imagine someone in CS taking the latency from each keypress during a search query and using it as an input into a machine learning algorithm to improve search results or infer user data. (At one point this is the idea I was sure Google would hire me for). Alternatively, infer user's processing load from this real time data and use it to offer help. Combine realtime behavioral data with AJAX and Comet to improve user experience-- slowdown in input leads to a "You seem to be having trouble. Would you like help with...?" sort of prompt. --AustinFrank
Comment: We (Carlos and Maria Polinsky) are already using a self-paced reading experiment over the web (See here: http://polinskylab.dnsalias.net/). The software was developed by Alex Drummond and can be downloaded here: http://code.google.com/p/webspr/
This software also requires a high polling rate keyboard, same as Linger, (see post: http://hlplab.wordpress.com/2009/02/02/diagnosing-the-linger-usb-keyboard-sampling-error/) but it is better than Linger in that a good keyboard and WebSPR works on any platform. A good keyboard and Linger still suffers from the "keyboard bug" when running on Windows. More details about how we've been testing some platform/keyboard/Linger vs WebSPR combinations can be found here: https://wiki.hmdc.harvard.edu/LPLab/Main/UsingLingerAndWebSPR --CarlosGomezGallo
Tool development
Integration of annotation, analysis, and visualization tools. Seamless integration of Praat, R, and R graphics (maybe also http://www.ggobi.org/ GGobi or http://rosuda.org/mondrian/ Mondrian). Ideally, import a data set, get a visualization, click on a point in the figure and have that sound file and associated annotations open in Praat. Automates outlier inspection, for example. Project would include UI design and usability testing for CS research angle. --AustinFrank
Toolkit for experimental interface creation. We use increasingly complex environments in our visual world eye tracking studies. Some common design elements are emerging (buttons for change of state and motion, visually segregated regions of the screen, etc.). These common elements should be easy to reuse without reinventing the wheel. They should be modular so that they can be combined into new designs. While this platform would obviously be useful for visual world studies, it could also serve as a test ground for usability testing on prototypes of UI designs, with eyetracking capabilities baked in. --AustinFrank
Develop a cross-platform MCMC sampling platform that doesn't suck. WinBUGS sucks. JAGS can be a pain to get working on different platforms. Implement a number of MCMC sampling algorithms on top of JVM or .NET platform. Add parsers for, for example, BUGS models specifications, lme4 model specifications, and some other specification that isn't horrible to use. Dump results in formats that are useful for analysis (can be read by R, matlab, etc). Take advantage of multiple processors to run multiple chains. Develop a server-client architecture to farm out several runs over many nodes on a cluster. Implement advanced algorithms like simulated annealing, Rao-Blackwellization, hybrid (Hamiltonian) MCMC, reversible jump, etc. Develop a GUI and/or a web app for specifying graphical models. --AustinFrank
Resource creation