ls ./machine-learning
Studying the limitations and potential applications of ML.
Existing methodologies of SimCLR, beta-VAE, and VICReg were replicated and applied to new datasets to test generalization. This study explores disentanglement within the context of representation learning in new domains.
Independent study under Dr. Richard Lange at RIT's Computer Science Department.
An end-to-end continual learning system using experience replay that learns from an online stream of network traffic. The project enables continual learning from multiple data sources over time via data homogenization, with uncertainty sampling to reduce labeling burden.
Conducted under Dr. Shanchieh Jay Yang at the RIT ESL Global Cybersecurity Institute.
Investigated divisive normalization, a canonical cortical gain-control computation, within CNNs designed as biologically plausible models of the auditory system. The work explores whether incorporating this normalization mechanism brings CNN-based hearing models closer to the response properties of biological auditory processing.
Conducted at the MIT McGovern Institute as part of the MSRP-Bio 2023 program.
Model-Agnostic Meta Learning implemented for EEG emotion recognition, capable of adapting quickly to new subjects with minimal data. Uses few-shot learning with normalization and alignment techniques to handle EEG variability across subjects.
In collaboration with the RIT Neurotechnology Exploration Team.
Active-learning surrogate models that optimize high power microwave device parameters, reducing the computational cost and time of simulation. An interactive query system obtains human labels for predictions with high uncertainty.
Conducted under Dr. Ashar Ali at the Air Force Research Laboratory Scholars Program.