Highly effective and passionate Data Scientist adept to collecting, analyzing and interpreting large data sets, developing new data regression models using predictive learning and delivering high impact data-driven insights to non-technical business audience. Possessing extensive analytical skills, strong attention to detail and a significant ability to work in team environment while leading by decision making. Looking for full-time opportunities.
-
Experience
Data Science Intern
Company Name - Autodesk
Dates Employed - May 2019 – Aug 2019
Employment Duration - 4 mos
Location-San Francisco Bay Area
- Outlined key sabbatical patterns of Autodesk employees to the Finance-Audit Team using Splunk’s active directory data
- Developed a predictive model for job retention and projected time series sabbatical trends
- Performed exploratory data analytics, data visualization and identified KPIs using SQL, Python and Tableau
- Deployed project on AWS, automated the Rest API data extraction and email notification
- Presented the data driven insights to the C-staff and recommended resource and software licensing reduction by 15%
-
Projects
Project -Few-Shot-Image-Generation-Using-VAE
Description -The aim of this project is to create novel and creative images using few-shot image generation. This project will provide assistance in creative process for different artists. Artists or designers who lack time or creative inspiration for multiple versions of an image could sketch a limited number of drawings. They can have the trained few-shot learning model, generate multiple similar versions of the sketches that they produced. Earlier, Generative Adversarial Networks (GAN) were used to generate realistic images. However, GAN’s require inordinate amount of data. Meta-learning can be used to bypass this hurdle of less data. Hence the goal of our project is to use Few-shot Image Generation by manipulating latent features of generative models.
Approach- In order to address the above mentioned goal, we took the following two routes:
Variational Auto-Encoder: VAE are a type of generative models. They are based off of auto-encoders. Auto encoders have two parts, encoders and decoders. VAE learn to generate new data by minimizing the reconstruction loss and latent loss. What goes in to the network is spit out making sure there is as little difference as possible. It is also made sure that the latent vector takes only specific set of values.
Reptile Algorithm to Meta train the model : Reptile seeks an initialization for the parameters of a neural network, such that the network can be fine-tuned using a small amount of data from a new task. Reptile simply performs stochastic gradient descent (SGD) on each task in a standard way.