Data Science & Neuroscience Projects

Multi-addicted State Q-agent Decision Making

In nature, there may exist common characteristics in information processing and decision making that are exhibited by different type of systems. It has long been proposed that reinforcement learning algorithms and the neural mechanism of human decision making are highly alike. In this study, we investigated different Q-agents with different exploration strategies decision-making performance under a discrete multi-addicted state environment. The below figure demonstates teh environmental setup of this study. Manuscript for this study is here.

Intuitive Laplacien Eigenmap Visualization

Dimensionality reduction technique is a crucial technique in machine learning as they are an approach in finding the "principals" in high dimensional data, which tends to be very noisy and would affect the accuracy of simple classifier significantly. However, these techniques are usually very mathematically intense and complicated for the general public to understand without prior mathematical background. Therefore, we developed this visualization project using D3, JavaScript to deliver an intuitive understanding of one of the most popular dimensionality reduction algorithms (Laplacien Eigenmap) on the art work collections in MET, New York.

Robust Ensemble Learning

Recipes and ratings play a pivotal role in our everyday lives, influencing various aspects from culinary experiences to social interactions. Predicting users’ preferences is particularly crucial as it enables personalized experiences and enhances efficiency in content discovery. In our analysis, we focus on how to predict user preferences based on various numerical and textual features. Our approach involves employing advanced techniques such as TF-IDF transformation, PCA, and a homogenous ensemble learning method, specifically Random Forest, to construct a reliable multi-class classifier with more robust and reliable predictions even facing imbalanced datasets, ensuring dependable predictions in scenarios where data distribution is skewed.