- Data analytics for applications in brain-computer interfaces.- Development of algorithms and software for electron-tomography data restoration.

- Expert on simulations of direct-electron Silicon detectors for TEM (transmission electron microscopes). Developed a comprehensive electron and X-ray Monte Carlo simulation software that allows accurate and efficient predictions of imaging-performance metrics of direct-detection cameras under a wide range of voltages and other parameters. This has allowed Gatan to drive strategic decisions on camera development that resulted in considerable cost reductions, by avoiding the need to experimentally test a large matrix of possible alternatives.– Created algorithms to reduce or eliminate artifacts common in in-situ imaging.– Developed novel methodologies to increase electron detectivity relative to more traditional approaches.– Proposed a novel type of electron detector aiming at reducing scatter and thereby yielding improved resolution.– Analyzed experimental imaging data, both to evaluate performance of prototype detectors and to validate the simulation software.

My focus was the development of algorithms for applications in 3-D electron-microscopy and tomography image processing:– Designed and implemented a constrained-deconvolution-based approach for restoration of 3D electron-tomography (ET) images, which suffer from high noise levels and missing-wedge artifacts. (Collaboration with LBL / UC Berkeley, which provided experimental data.) Applied to trace filaments in actin ET maps. (Quadratic programming problem.)– Also, developed methods for fast and accurate calculation of Mutual Information, involving non-equispaced Fourier transforms and a novel kNN-based (k-nearest-neighbor) approach. Applied these methods to analyze molecular-dynamics trajectories, and to correlate speech with brain signals, with the goal of decoding the latter and reconstruct speech. (Collaboration with another lab at ODU which provided data.)

- Developed sophisticated software to generate quasicrystals from arbitrary lattices in any dimension, including phasons, which are essential for the modeling of particles according to our unification theory. Tree data structures used throughout the code yield a substantial reduction in complexity and storage, which is crucial for high-dimensional problems like E8.- Created a new type of quasicrystal with icosahedral symmetry made up of regular tetrahedra. A significant percentage of the tetrahedra can flip between two alternative configurations, providing a novel phason mechanism with potentially useful applications.- Contributed original ideas toward the development of a unification theory of physics, primarily on models of space and particles, based on key properties of mathematical quasicrystals, in particular those derived from the E8 lattice.- Carried out computer simulations and experiments of prototypes of low-cost power generators. (Electromagnetism equations via COMSOL.)

- Developed advanced correlation filters to optimize the performance of detection of fragments in 3D electron-microscopy images. This “generalized matched filtering” technique yields reliable fittings even in the presence of high levels of noise.

Research and development of technologies and software for satellite tracking and data processing:- Developed an approach to simultaneously move all angles in multi-axis antenna mounts to avoid keyholes in an optimal way, especially for tracking fast-moving targets. Applied it to SeaSpace’s AXYOM antenna geometry.- Designed a novel antenna mount geometry that combines a standard AZ/EL pedestal and a “sliding feed assembly.” This drastically reduces the complexity and weight of the AXYOM design, while keeping it keyhole-free, and is robust enough to eliminate the need for a radome.- Incorporated into TeraScan the attitude modulation (“yaw steering”) performed by the MetOp satellite, reducing geolocation errors from 50 km or more down to about 2 km — the normal value for polar-orbiting satellites.- Incorporated into the TeraScan's Batch Processing System several packages for the post-processing of satellite data (Modis, AIRS, AMSR-E, AAPP/IAPP), which generate a variety of products and atmospheric profiles.- Improved various parts and algorithms of the TeraScan processing software, including: sea-surface temperature calculations (for both polar-orbiting and geostationary satellites), calibration coefficients, ESA Sharp HRPT format, cloud and fog product scripts, and Advanced Data Collection System (A-DCS).

Developed advanced methods and software to approach challenging problems in drug design, structure determination, fitting and docking:- Introduced techniques to predict deformability of molecules and to model receptor flexibility in structure-based drug discovery. Improved success ratio 2–3 times relative to the use of rigid receptors. The internal-coordinate version achieves a speedup and storage reduction of about 15 times relative to the Cartesian version, and preserves the correct covalent geometry. A web applet that computes deformability predictions has been developed.- Formulated a fast protocol for helical molecules, 3–4 orders of magnitude faster than the standard. Involved energy minimization, PEA (Pairwise Energy Analysis), SCATD (Side-Chain Assignment via Tree Decomposition), and ICM (Internal Coordinate Mechanics). Applied to the transmembrane portions of gap-junction channels.- Invented an efficient approach, “Damped-Dynamics Flexible Fitting” (DDFF), to model conformational trajectories of protein molecules to fit them into EM maps or target structures. Works in internal coordinates and achieves the minimum possible complexity O(N^2), where N is the number of atoms.- Supervised Ph. D. student on harnessing FRM for protein-protein docking. This resulted in a novel approach with significantly higher efficiency than existing tools in its category.

Developed analytical and computational methods and software for biomedical applications, particularly in the area of 2D and 3D image processing of electron-microscopy data:- Developed a novel technology, “Fast Rotational Matching” (FRM), based on spherical harmonics and FFT (Fast Fourier Transform), to accelerate the pattern recognition process in 3D matching problems. FRM provided the solution to a 3-decade-standing problem, and achieved an increase in performance of between 1 and 3 orders of magnitude over previous methods. Software package for atomic-structure fitting into 3D EM images is 60 times faster than existing approaches. FRM has also been incorporated in the world-leader molecular-replacement package AMoRe, used to solve atomic structures by X-ray crystallography.- Created a fast and robust method, “Fast Bessel Matching” (FBM), for the alignment of noisy 2D EM projection images. FBM combines a novel parametrization (3 angles) of motions in the plane with Bessel functions and FFT, achieving significant matching speedups with a consequent productivity increase in obtaining the 3D images.- Codeveloped “Topology-representing neural networks” (TRN), which generate reduced models of the target and probe objects, making it easy to then find their best match.