Other Projects

1. Unsupervised Generative ML Models

• Conducted comprehensive evaluation of 4 unsupervised generative models in JAX
  • Energy Based Models (EBM)
  • Denoising Score Matching (DSM)
  • Denoising Diffusion Probabilistic Model (DDPM)
  • Generative Adversarial Networks (GAN)
• Multiple 2D datasets were used to evaluate generalized model performances
  • Checkerboard
  • Gaussian Mixtures
  • Pinwheel
  • Spiral
• GAN showed the best generations with the best mean KDE (-1.34) across all datasets. However, I could not realize the potential of GANs to act as the best generators. To that end, adding batch-normalization layers to the genera- tor and discriminator MLPs would have helped to capture more variance but my system compatibility with jax prevented me from such.
• The report can be found here.

2. NBA-Draft-Prediction

• Benchmarked several ML models like RBF Regression, SVMs, Neural Networks, KNN to predict NBA players’ performance and draft pick in next season, based on demographic or statistical attributes. KNN reached a best of 0.8 R2 score.
• The report can be found here.

3. Parallel Low-Latency Automated Trading System

• Parallelized trade orders for NASDAQ stocks of 30 companies over a market day using PThreads in C++
• Reduced latency from 200ms in a linear system to 30ms, increased profits 40-fold by higher trading frequency
• The report can be found here.

4. Course Web-Registration System by Socket Programming

• Built backend of a university’s course registration system in C++ across multiple department servers, a main server, and a client platform with dedicated TCP and UDP ports capable of handling multiple queries, with 0 error rate.

5. Load-Carrying Mule Bot for Supermarket Customers

• Headed team of 40 students in designing a trolley to follow customers via Bluetooth & GPS, and carry their supplies
• Planned installation of WiFi modules enabling P2P communication for optimal routing and sharing charging ports
• The report can be found here.