ARC-AGI Challenge
(Knowledge Based AI)
This project aimed to develop an agent capable of solving specific problems that traditional machine learning struggles with due to small training datasets and unique challenges. It aimed to enhance Artificial General Intelligence (AGI) by applying knowledge-based AI techniques effectively in solving tasks like ARC, aiming for more human-like intelligence.
Skills: Domain specific language (DSL), Frame generation, Mean sense analysis
Arc website: ARC Prize
Death Analysis Website
I participated in the Georgia Tech Data Science Hackathon, collaborating with a team of four to develop a website analyzing deaths and growth rates in every country over 20 years.
Tools that I utilized were Streamlit, Jupyter notebook, and GitHub.
Website: Death Analysis of Every Country (deathanalyzer.fly.dev)
AI Projects
Searching (Top Left):
I utilized BFS, DFS, and A* algorithms to enable Pacman to efficiently search for his food.
Tracking (Top Right):
I incorporated probabilistic inference techniques like Naive Bayes Network and Particle Filter. Now, Pacman effectively hunts down ghosts using a noisy sensor.
Reinforcement Learning (Bottom Left):
I applied reinforcement learning techniques, including value iteration and Q-learning, to train Pacman to navigate the world while evading ghosts.
Machine Learning (Bottom Right):
I built and trained a Neural Network from scratch for classification purposes.
Robotics
I utilized Webots to simulate robot movement using a particle filter with Lidar and Stereo sensors. The robot navigates through an unknown environment using markers.
Aquabots (Computer Vision Team)
In a team of 5, utilized computer vision and neural networks to classify plankton types. Assisted professor in optimizing code with TensorFlow, resulting in a notable increase in model accuracy from 80% to 85% through hyperparameter fine-tuning.