Daniel Leeds – Research

Computer vision models of object representations in the brain: We explore how well different computer vision recognition models account for neural object encoding across human cortical visual pathway as measured using fMRI. These neural data, collected during the viewing of real-world objects, are analyzed with a searchlight procedure to identify models and cortical regions that group visual objects in a similar manner. Realtime fMRI search for visual properties encoded in the brain: I utilize realtime fMRI analysis in conjunction with approaches from optimization and computer vision to address the question of compositional features underlying object representation in the brain. Euclidean spaces representing complex visual properties are defined based on SIFT and "Fribble" representations. These spaces are searched using a variation of the simplex method, in conjunction with realtime processing of neuroimaging signals. Semantic representations of objects in the brain: We explore how well different semantic features account for neural object encoding across human cortical visual pathway as measured using fMRI, similar to the computer vision approach above. These neural data, collected during the viewing of real-world objects, are analyzed with a searchlight procedure to identify models and cortical regions that group objects in a similar manner.

Past projects
Efficient coding of natural sounds: I developed a multi-stage, probabilistic encoding method for sounds. Incorporating insights from neuroscience, this method captured non-linear acoustic structures using a small number of representational components.

Computer-assisted diagnosis of heart murmurs: I developed software to produce visualizations of patients' heart sounds to aid in diagnosis of murmurs. I use tailored methods to extract relevant acoustic features and self-organizing maps to project information from a multi-dimensional space into an intuitive two-dimensional grid.

Articulator motion in speech production: I have studied the roles of motion in the lips and tongue (among other articulators) for the production of speech in varying conditions — such as slow or fast.

Statistical structure of bird song: I used statistical analyses to characterize the variations in the properties of the zebra finch song across single recitations. This method can be extended to study the evolution of a song while it is learned by young finches.

Publications

Z Syed, D Leeds, D Curtis, F Nesta, R A Levine, and J Guttag, "A framework for the analysis of acoustical cardiac signals," IEEE Transactions on Biomedical Engineering, 54(4), April 2007. [link]

Presentations

DA Seibert, DD Leeds, JA Pyles and MJ Tarr, "Exploring computational models of visual object perception," Vision Sciences Society, May 2012. poster

DD Leeds, DA Seibert, JA Pyles and MJ Tarr, "Unraveling the visual and semantic components of object representation," Vision Sciences Society, May 2011. [poster, appendix]

DD Leeds and MJ Tarr, "Searching for the visual components of cortical object representation," Temporal Dynamics of Learning Center All Hands Meeting, January 2011. [video]

A Nestor, DD Leeds, JM Vettel and MJ Tarr, "Neurally-derived representations for face detection," Statistical Analysis of Neural Data, May 2010. [poster]

Z Syed, D Leeds, D Curtis, J Guttag, "Audio-visual tools for computer-assisted diagnosis of cardiac disorders," Computer Based Medical Systems 2006, June 2006. [link]

Reports

Independent Manifolds in the Zebra Finch Song: A Strategy for Robust Social Interaction (Intel Science Talent Search submission 2000/2001) [pdf]