I teach undergraduate computing at Purdue University in Indianapolis. I have worked on multimodal attention architectures for detecting manipulated audio-video media.
AreasMultimodal representation learning; audio-visual deepfake detection; attention and transformer architectures; numerical methods and PDE-based image restoration.
arXiv:2309.06511 · cs.CV, cs.MM · September 2023
A multi-modal framework that processes audio and video concurrently. A cross-attention mechanism scores lip synchronization against the input audio while a fine-tuned VGG-16 extracts visual cues; a transformer encoder then applies facial self-attention. The approach exceeds prior multimodal detectors on F-1 and per-video AUC.
US 2025/0005925 A1 · Purdue Research Foundation · filed 24 Apr 2024, published 2 Jan 2025
Discloses a two-pipeline framework. A video-only pipeline uses a vision encoder with a feature extractor and a transformer encoder applying self-attention to facial-region artifacts. A separate audio-video pipeline uses an audio and lip encoder whose transformer applies cross-attention to detect discrepancies between lip motion and spoken words. The two modalities are combined for a joint manipulation inference.
Honors thesis · University of Southern Mississippi
An automatic solver for a nonlinear PDE image-processing model. The Perona-Malik equation performs simultaneous denoising and deblurring across forward and backward diffusion regimes, but large gradients in the backward regime cause staircasing. This work combines Perona-Malik with the Guidotti-Kim-Lambers bound on backward diffusion and selects parameters automatically by Nelder-Mead simplex optimization.
Instructor terms link to the course Brightspace page.
My teaching spans the systems and numerical sides of the undergraduate curriculum, from introductory C programming through operating systems, numerical methods, and applied deep learning.
My past research sat where those two halves meet: attention architectures that fuse audio and video, and the numerical machinery underneath image restoration.