• Developed the synthetic data generator for time-series machine learning codebase.

• Learned traditional methods of generating synthetic data through statistical techniques such as ARIMA.

• Ran experiments on the TimeGAN (Tensorflow 2.0) and Copulas methods in Vertex AI (GCP).

• Applied statistical learning schemes on deep learning models to capture time-series trend/seasonality.

• Created a user interface to easily generate synthetic data based on real data or a desired distribution.

• Presented results of synthetic time-series data generation algorithms in a presentation fair.

Technologies Used: Python, Tensorflow (2.0), Keras, Numpy, GCP (Vertex AI), Conda, Jupyter, Scikit-Learn, Git

• Developed a scheme and pipeline responsible for randomly sampling 2D images of meshes rendered in OpenGL from 3D data environments such as Replica.

• Created a pipeline which supplied sampled images to ProgressiveGAN or StyleGAN algorithms.

• Ran various experiments on remote GPUs in a bash environment.

• Developed Python scripts and notebooks that used analysis tools such as PCA to create GIFs and figure which visualized the latent space of generative models of 3D space.

Technologies Used: C++, OpenGL, Python, Jupyter, PyTorch, Scikit-Learn, Git, bash, conda, numpy, matplotlib, Linux, bash

• Learned probabilistic and statistical techniques such as Markov Processes, principal component analysis, and time lagged component analysis and their implementations in Python 3 (Scikit-Learn, Numpy).

• Learned Latent NeuralODEs, autoencoder, and variational autoencoders (PyTorch).

• Learned and implemented (Python 3) the numerical method known as the Nudged Elastic Band (NEB) method which helped find minimum energy paths.

• Developed multiple NeuralODE based models with PyTorch (Python 3, conda environment) for the problem of finding minimum energy paths.

• Ran various experiments on remote GPUs in a bash environment showing that NeuralODE based models which were able to perform better than the traditional NEB method.

Technologies: Scikit-Learn, Numpy, Python, Jupyter, Git, Linux, bash, PyTorch, conda

• Learned about the Deep Equilibrium family of deep learning models and their implementations (PyTorch).

• Developed a feedforward Deep Equilibrium Model in PyTorch (Python 3).

• Compared the performance of the developed Deep Equilibrium model with Securiti.ai's own model and a state-of-the-art Transformer by performing experiments in Amazon's SageMaker services.

Technologies Used: Python, PyTorch, conda, Git, Jupyter, Linux, bash, AWS, SageMaker