Ehsan Kourkchi, PhD

Experience

• 

  Springboard

  Jan 2020 - Jan 2021

  Springboard Machine Learning Career Track

  Project: Galaxy Image Classification: The objective of this project is to automatically determine the inclination of spiral galaxies with the human level accuracy providing their images (ideally in both colorful and black-and-white formats).The inclinations of a disk galaxy can be roughly derived from the axial ratio of the projected ellipse that defines its boundary. However, this approximation only provides good enough inclination estimates in ~30% of cases for various reasons, such as the existence of prominent bulges that dominate the axial ratio measurements. In this project, I built multiple models by employing the Convolutional Neural Network (CNN) of different architectures to determine the inclination of spiral galaxies from their visible images. I utilized ~20,000 galaxy images, taken from the SDSS image archive, to train and evaluate these models. I determined the inclination of the galaxies in my sample manually with the collaboration of citizen scientists. Exploring different CNN structures, I found that models with the convolutional filters of size 3x3 are simple but yet powerful to tackle the problem, given the limited computational resources. I report that averaging across multiple training scenarios and model architectures improves the overall accuracy of predictions. The trained networks can measure inclinations with the root-mean-square uncertainty of ~3 degrees, comparable to the average human performance, i.e. ~2.6 degrees. All models that I studied exhibit better accuracy than the ellipticity-based formalism.Codes and a complete report is available here: https://github.com/ekourkchi/inclinet_project Show less

• 

  University of Hawaii at Manoa

  May 2020 - Jun 2021

  Developing and deploying machine learning applications to forecast meteorological time series.In this project, I built a Bayesian machine learning model based on the Gaussian Process Regression (GPR) methodology to forecast the meteorological parameters such as terrestrial evapotranspiration (ET) and rainfall. This model takes an auto-regressive approach where each value in the time series is predicted based on a few previous data points. Our model benefits from the Particle Swarm Optimization (PSO) technique to optimize the hyper-parameters of the adopted GPR kernel.The main objective of this project is to collect and process data from the selected HOBO and CAMPBELL stations that are managed by this project. The stored data can be accessed through the URL queries. The server starts updating the local database at 1:00 AM (HST) and generates a set of forecasts for the future days. I developed and API that provides access to the latest forecasts to facilitate the agricultural decisions. Codes and more details are available here: https://github.com/ekourkchi/cropNetThe URL of the deployed project: https://cropnet.eng.hawaii.edu/ Show less

  • Postdoctoral Researcher

    Aug 2020 - Jun 2021

  • Cosmicflows-4 Program Research Assistant

    May 2020 - Jan 2021
• 

  Utah Valley University

  Jan 2021 - Jan 2023

  Postdoctoral Researcher

  Automating the pipeline to process images of elliptical galaxies observed by the Hubble Space Telescope (HST) to measure their distances through the Surface Brightness Fluctuation (SBF) methodologyUntil now, the workflow to process galaxy images to extract the SBF signal was manual and time consuming. In this project, I am working on automating solutions to minimize human supervision and improve the quality of the extracted signal. The constructed pipeline could be utilized to process the galaxy images of the LSST big sky surveys to accelerate the task of SBF distance measurements.I also continue working on the Cosmicflows project in collaboration with team to generate the next generation of the galaxy distance catalog. My work focuses on the quality assessments of distances, combining distances of multiple catalogs, visualizations and making astronomical measurements that mostly rely on data rather the models assumptions. This project involved statistical analysis (such as the MCMC simulation, mock data generation, parameter fitting, etc.). I also involke ML methodologies, if possible, to make strong data driven predictions in cases where data is missing, corrupted or biased. I adopt data-science best practices and latest available technologies to deal with big data sets and large number of parameters and features when constructing robust empirical formalisms. Show less

• 

  SensiML Corp

  Dec 2021 - now

  Chief Data Scientist

  In my role as a data scientist and machine learning engineer, the core of my daily tasks centers on adopting, developing, testing, and maintaining a toolkit designed for crafting machine learning models suitable for deployment on devices with constrained memory and processing capabilities. Some of my key responsibilities include:Developing machine learning models to analyze real-time streaming sensor data for various industrial applications, including predictive maintenance, speech recognition, sound classification, activity detection, and gesture classification.Providing consultation on the complete lifecycle of a machine learning project, including guidance on data collection, model testing in both controlled laboratory and real-world environments, and strategies for continuously improving model performance over time.Integrating state-of-the-art machine learning algorithms into the SensiNL system to offer seamless autoML services to all users, regardless of their background.Actively engaging with end-users and customers to design customized ML solutions tailored to their specific needs.Monitoring, improving, and debugging the SensiML system backend.Testing and validation of ML algorithms for deployment on edge devices.Optimizing the performance of ML algorithms deployed on small, resource-constrained devices. Show less