Winners of Neukom Graduate Fellowships have been announced for the 2011-2012 academic year. Fellowships will provide a full year of funding, including stipend and benefits, to Ph.D. students engaged in faculty-advised research in the development of novel computational techniques as well as the application of computational methods to problems in the Sciences, Social Sciences, Humanities, and the Arts.
The 2011-2012 winners are:
One of the most fundamental questions that has been pursued by modern science is, “Are we alone?” To begin answering that question, astronomers have embarked on a mission to discover the first Earth-like exoplanet, or a planet outside our Solar System orbiting a star, other than our own, that is capable of having liquid water on its surface. Intuition suggests that searching around Sun-like stars would be best, as we have a direct comparison system: the Solar System.
However, much attention in the astronomical community has turned to searching around M dwarfs, or stars that are smaller and less massive than the Sun. M dwarf stars are defined to have masses between _0.1 and 0.6 times the mass of the Sun. It is the low mass of these stars that make the detection of an exoplanet comparatively easier. The kinds of observational evidence that astronomers search for is more pronounced for a given exoplanet orbiting an M dwarf than a Sun-like star. Once detected, the potential habitability of an exoplanet may be surmised based on several characteristics of the system: the proximity of the planet to the host star, the planet radius, and the planet composition.
The first characteristic determines if the planet has potential to support liquid water on its surface and is directly related to the effective (surface) temperature of the host star. If the planet receives too much radiation from the host star, liquid water evaporates; too little, it freezes. The final two characteristics are intimately tied to the host star radius. Current methods of detection only yield the ratio of the planet radius to the host star radius; neither value is uniquely determined by observations. As such, exoplanet observers need a way to estimate the radius of the host star. Of the various methods available, stellar models provide the most efficient means of doing so in terms time, cost, and accuracy.
Unfortunately, evidence gathered over the past decade suggests stellar models are not able to accurately predict the radius or effective temperature of low mass M dwarfs. Here, we propose to implement the effects of a large-scale magnetic field in a self-consistent manner within the framework provided by the Dartmouth Stellar Evolution Program (DSEP). Unlike previous studies, all effects will be accounted for in our approach, allowing for the full range of consequences to be studied; we will not simply alter one parameter arbitrarily. Fortunately, the framework provided by DSEP is currently more accurate than any other model code in the literature5, subject to discrepancies around 5-7%. The purpose of the work is to not only introduce the first of it’s kind, self-consistent magnetic models, but to also make easy the implementation of stellar models in other sub-fields of astronomy by providing access to a free, simple, and intuitive web interface.
There is no existing software package that provides a full user-enabling workflow for quantitative imaging with light in tissue, and here we propose to accomplish this. Light interaction with tissue is highly non-linear and distorted by tissue shape, absorption and scattering. Currently, we have a software tool, called Nirfast, which is a finite-element based package for modeling near-infrared light transport in tissue for medical applications. It is open source, free, and cross platform (www.nirfast.org), based largely on MATLAB with C mex files, and new linkages to QT and PYTHON produced visualization tools that utilize the VTK/ITK libraries. In the current version, there are approximately 60 unique downloads a month and 158 unique visits a month. Nirfast is kept under open version control via Google Code, making it easy to keep up to date with the latest developments and propose changes/improvements.
In this research, I plan to refine the clinical interface tools based upon PYTHON coding to make the tool useful to a novice user and test the utility in optical spectroscopy imaging of cancer tumors. The planned interface is not the same as is currently grant supported, but rather will be an easier use-of-interaction version which will be optimized for clinical utility. We currently have grant supported collaborations with Kitware Inc. (Clifton Park, NY) to produce and refine a tool that is our expert user interface and the plan in this proposal is to develop a more clinically-directed version of this allowing easy technical/clinician interface and use.
Large collections of music (>1M tracks) are available online to all. Current search techniques are either non musical, employing meta-data such as: song title, artist name, and tags, or they require an abstract music-theoretical representation of music, such as: melody, chord progression, timbre. We ask the question: what if we could search large collections of music by groove similarity? Groove describes the background, or foundational, rhythm of a track. As such, it is an essential component of most non-classical music: pop, rock, hip-hop, and dance music, for example. Perhaps more than melody, harmony, and beat (tempo), groove is the essence, or the gist, of a track. There are currently no systems that match tracks by their background rhythmic content, i.e. their groove.
To extract and represent groove, we must address one of the difficult challenges facing signal processing; namely, mixtures. Music recordings of all types consist of mixtures; the recorded sources are transformed via real or virtual acoustic processes and these are summed to make a stereo or mono track. We propose to investigate methods for separating and organizing signal-derived features into timbre channels— mid-level representations of multiple musical parts embedded in a mixed audio track. We propose to extract timbre channels using an extension of probabilistic latent component analysis algorithm (PLCA) that operates globally on the audio data-set, thereby producing features that capture the timbral variance across all tracks. A combined multi-channel distance metric, using Bhattacharyya distance, combined with the timbre-channel representation yields search by groove.
At the end of the project , I will have contributed to 1) a novel rhythm feature and distance measure 2) scalability to large (> 1M tracks) music collections 3) new knowledge about separated feature spaces versus sub-band spectrum features 4) contributions to the Bregman Python toolkit  for multimedia analysis and feature extraction 5) a publicly available marked-up data set of 1138 commercial dance music tracks for rhythm retrieval experiments, markup by actual NYC DJs.
Last Updated: 10/2/13