Hi, I'm
Kabilan Senapathy

Single Nucleotide Polymorphism Classification2018

Used ensemble methods to classify single nucleotide polymorphism (SNP) genotype data (synthetic) of 4,000 cases and 4,000 controls. The data had 29,623 features, of which only 15 were causal. Used PCA to reduce dimensionality and achieved a Top-1 accuracy of 66.2%.

Adaptive Step Size - Hinge Loss and Least Squares Loss2018

Developed a single-layer neural network in Python and implemented an adaptive learning rate for gradient descent. Applied it to both hinge loss and least squares loss and studied the effect of the learning rate on convergence.

Kaggle Dataset - Image Classification2019

Developed multiple Convolutional Neural Network classifiers in Keras for a wide range of Kaggle image classification datasets.

Mini-ImageNet Classification2019

Developed a 5-layer CNN for Mini-ImageNet classification with a minimum accuracy target of 80%. Used 50% dropout on every layer, an unusually high value that performed well, and studied the effect of high dropout during training.

Feature Space Transformation for Multi-Modality Data2019

Built a transformer that maps data from multiple unknown distributions into a known feature space. A single common model can then serve data from different modalities, reducing development time and engineering effort.

Visualization for Data Discovery in Open Data Portals (Discovery)2019

Helps users visually understand the links between arbitrary datasets and what information those links provide (e.g., linking datasets A, D and C might provide less information than linking datasets G, D and F).