- Currently designing a robust auto-encoder and GAN for detecting adversarial images.
- Developed a close proximity approximation model which is also known as on manifold adversarial detectors.
- Enhanced the vanilla auto-encoder to a deep architecture with enforced learning from memory elements.
- Seq-GAN is a unique approach which models the data generator as a stochastic policy in reinforcement learning to solve the problem.
- The RL reward signal comes from the GAN discriminator judged on a complete sequence, and is passed back to the intermediate state-action steps using Monte Carlo search.
- Working with the accuracies and flaw removal strategies with re-implementation of Open-Face/Googles Face-Net, for improving the range of applications in the domain of Security.
- Resolved the false positive 2-D inputs by introducing more features in Stage 1 (face detection) as a.) Orientation Normalization b.) 3D surface representation.
- The search for applying machine learning to real world problems led me to work in the area of Astro-turfing. This is a filtering model with machine learning algorithm to filter-out the fake reviews and hence the fake reviewers from committing review frauds. I used amazon reviews data sets to built a stochastic model on top of it.