Hyperparameter optimization for a Gaussian Process Regression on Mauna Loa Atmospheric C02 level. GPU acclerated implementation using GPyTorch.
GPU Accelerated Gaussian Process Regression using GPyTorch
Gaussian Processes
Bayesian Statistics
Hyperparameter Optimization
GPyTorch
Confidence Intervals
Gaussian Processes - Updating the posterior
Gaussian Processes from scratch. Updating the gaussian process posterior using conditioning and noise in the observations. Sampling distributions and plotting confidence intervals.
Gaussian Processes
Bayesian Statistics
Machine Learning
Interpretability
Confidence Intervals
Scalable and Weakly Supervised Bank Transaction Classification
arXiv paper written with my colleagues at Tillful showing how we tackled the challenge of categorizing bank transactions utilizing weak supervision, natural language processing, and deep neural network techniques.
Article
Natural Language Processing
FinTech
Deep Learning
Embeddings
Semi-Supervised learning
Informative Youtube Video Classifier without labels using transformers and flippers
Fine tuning a distillbert transformer for classification without labels using weak supervision and flippers. The model seperates informative (scientific, historical, +) and uninformative (music, video game, vlogs, +) Youtube videos just from their title.
Transformers
BERT
Machine Learning
Weak Supervision
NLP
Variational Auto Encoders for Weak-Supervision
Implementing a type of Variational Auto Encoder for Weak-Supervision. They use latent variables to generate both the data and the label that is used to decode the labeling functions.
Python
Machine Learning
Weak Supervision
Variational Auto Encoders
Generative Models
flippers library
Flippers is an open-source weak supervision library for creating high quality labels using your domain kownledge and weak supervision sources.
Python
Machine Learning
Weak Supervision
Probabilistic Models
Multi Armed Bandits for Recommendation Systems
GitHub repository showcasing different Multi Armed Bandits and Contextual MAB algorithms with implementations on real recommendation data.
Python
Recommendation Systems
Contextual
Feature selection
Uncovering the horseshoe effect
Research article written during my Internship at UCSD's Knight Lab, explaining how the horseshoe effect arises in data with an underlying gradient and how to learn from it to build a better metric.
Research Article
Data analysis
Unsupervised Learning
PCA / SVD
Feature selection for loan classification
Feature selection notebook for banking loan classification.
Finance
Feature encoding
Classification
Feature selection
World of Warcraft Item Quality Classifier
Classification of World of Warcraft gear and their in-game attributes.
Data Exploration & Cleaning
Supervised Learning
Classification
XGBoost
Bubble map using plotly
Short Kaggle kernel demonstrating how to make a bubble map using worldly, applied for quantifying job postings per country in a glassdoor dataset.
Data visualization
Data cleaning
Python
Seaborn
Plotly
Asteroids & regressions
Notebook benchmarking the different classifiers in the scikit-learn suite on a NASA dataset.
Python
Supervised Learning
Regression
sklearn
Keras
XGBoost
Using Deep Q Learning and OpenAI Gym
Implementation of Deep Q Learning (both with keras and using my own neural network implementation) on different OpenAI Gym games.
Reinforcement Learning
OpenAI Gym
Keras
Deep Q Learning