Key-Value Store in Variable Client-Server Configurations
Multi-Threaded, Event Polling and Event Signaling based servers were developed in Python. Log Structured Merge Trees were then incorporated, resulting in better I/O performance. Servers were distributed over AWS instances for higher concurrency. ABD algorithm was used to ensure sequential consistency between processes.
Image Captioning using Flickr8k
Inception-ResNet-V2 was used to extract the most prominent features per image. These features were fed into a recurrent neural network along with the captions provided in Flickr8k. The network was able to produce simple but accurate descriptions of test images often. BLEU (Bilingual Evaluation Understudy) scores were just above 0.6.
Analysis of On-The-Job Training Participation
Google Time Series charts was used to visualize patterns of a two-decade long dataset from IPUMS. Prominent patterns suggesting racism and sexism allowed us to exploit specific attributes. Using Decision Trees to select relevant attributes and Random Forest to classify, important questions regarding the data were answered: racism and sexism encouraged higher minority and female participation in training programs.
Github Projects
- A Study of 17 Unique Wallpaper Patterns
- IDA* powered computer opponent in reversi with GUI
- Facial Recognition System
- 3D Multi-Colored Cube simulation under variable lighting using GLUT
- First come first serve vs round robin process scheduler
- Flash Flood Forecaster Performance Evaluation
- Multi-threaded proxy server
- OOP based POS System
- Academic website with forums, comments, emails backed by SQL database