NitinVishalKulkarni

AnomalyDetection

MLP, LSTM, and Autoencoder models to detect anomalies on AWS CPU utilization.

DeepLearningVsGaussianProcessClassification

In this project, we employ both Gaussian Processes and Deep Learning to solve a real-world time series multi-class classification problem. We used the Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC) dataset. It's a huge dataset of about 37.37 GB containing the electromagnetic flux data measured from different celestial bodies in six different passbands. The task is to classify the type of astronomical bodies which include classes such as brown dwarfs, different types of supernovae, and galaxies.

Dynamic_Modeling_and_Forecasting_of_Epidemics_Incorporating_Age_and_Vaccination_Status

Efficient-Generalized-Reinforcement-Learning-through-Causal-Inference

One of the fundamental challenges in the successful deployment of reinforcement learning systems in the real world is their ability to generalize. Although the state-of-the-art deep reinforcement learning algorithms can solve complex tasks, they require a large number of interactions with the environment and don't perform well on new environments. These algorithms latch onto the specifics of the environments they are trained on instead of learning general concepts. We propose that generalization to out-of-distribution (OOD) tasks can be achieved through causal inference. We demonstrate that our solution achieves generalization to a number of OOD tasks while being trained on just a few variations of the environments.

FairnessInML

A model to replace COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) using five different post processing models: demographic parity, equal opportunity, predictive parity, maximum accuracy, and single threshold with SVM, neural networks, and Naive Bayes.

Fake_News_and_Propaganda_Detection

ImageNetClassification

Alex Net and ResNet (18/34/50 layers) models to classify a portion of the ImageNet dataset.

MovieGenrePrediction

Implementation of movie genre prediction using Apache Spark. The goal is to do a multi-class multi-label classification based on the synopsis of the movie. The model is able to achieve a F-score of 0.99978 on the test data containing 7000 movie plots.

MovieRecommendationSystem

KNN, MLP, and Collaborative Denoising Auto-Encoder models to build a movie recommendation system using the Movie-Lens dataset.

OfflineReinforcementLearning

The Advantage Weighted Regression algorithm based on the research paper by Xue Bin Peng, Aviral Kumar, Grace Zhang & Sergey Levine - "https://arxiv.org/pdf/1910.00177.pdf". The following code demonstrates its use for three environments 1. Open AI CartPole environment 2. A custom Wumpus World environment (Based on the Wumpus World environment from the book "Artificial Intelligence a Modern Approach" by Stuart Russell and Peter Norving 3. A modified multi-agent version of the Wumpus World environment.

NitinVishalKulkarni/OfflineReinforcementLearning

The Advantage Weighted Regression algorithm based on the research paper by Xue Bin Peng, Aviral Kumar, Grace Zhang & Sergey Levine - "https://arxiv.org/pdf/1910.00177.pdf". The following code demonstrates its use for three environments 1. Open AI CartPole environment 2. A custom Wumpus World environment (Based on the Wumpus World environment from the book "Artificial Intelligence a Modern Approach" by Stuart Russell and Peter Norving 3. A modified multi-agent version of the Wumpus World environment.

NitinVishalKulkarni/AnomalyDetection

MLP, LSTM, and Autoencoder models to detect anomalies on AWS CPU utilization.

NitinVishalKulkarni/DeepLearningVsGaussianProcessClassification

In this project, we employ both Gaussian Processes and Deep Learning to solve a real-world time series multi-class classification problem. We used the Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC) dataset. It's a huge dataset of about 37.37 GB containing the electromagnetic flux data measured from different celestial bodies in six different passbands. The task is to classify the type of astronomical bodies which include classes such as brown dwarfs, different types of supernovae, and galaxies.

NitinVishalKulkarni/Dynamic_Modeling_and_Forecasting_of_Epidemics_Incorporating_Age_and_Vaccination_Status

NitinVishalKulkarni/Efficient-Generalized-Reinforcement-Learning-through-Causal-Inference

One of the fundamental challenges in the successful deployment of reinforcement learning systems in the real world is their ability to generalize. Although the state-of-the-art deep reinforcement learning algorithms can solve complex tasks, they require a large number of interactions with the environment and don't perform well on new environments. These algorithms latch onto the specifics of the environments they are trained on instead of learning general concepts. We propose that generalization to out-of-distribution (OOD) tasks can be achieved through causal inference. We demonstrate that our solution achieves generalization to a number of OOD tasks while being trained on just a few variations of the environments.

NitinVishalKulkarni/FairnessInML

A model to replace COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) using five different post processing models: demographic parity, equal opportunity, predictive parity, maximum accuracy, and single threshold with SVM, neural networks, and Naive Bayes.

NitinVishalKulkarni/Fake_News_and_Propaganda_Detection

NitinVishalKulkarni/ImageNetClassification

Alex Net and ResNet (18/34/50 layers) models to classify a portion of the ImageNet dataset.

NitinVishalKulkarni/MovieGenrePrediction

Implementation of movie genre prediction using Apache Spark. The goal is to do a multi-class multi-label classification based on the synopsis of the movie. The model is able to achieve a F-score of 0.99978 on the test data containing 7000 movie plots.

NitinVishalKulkarni/MovieRecommendationSystem

KNN, MLP, and Collaborative Denoising Auto-Encoder models to build a movie recommendation system using the Movie-Lens dataset.

NitinVishalKulkarni/Optimization_of_Mitigation_Regulations_During_Epidemics_using_Offline_Reinforcement_Learning

NitinVishalKulkarni/Optimizing_Pharmaceutical_and_Non-Pharmaceutical_Interventions_During_Epidemics

NitinVishalKulkarni/StockPricePrediction

ARIMA, AR, VAR, MLP, and LSTM models to predict Walmart stock price.