Nishanth Koganti
- +91 9347233393
- nishanth.koganti7@gmail.com
- buntyke.github.io
- Hyderabad, India
Researcher/Engineer with 5+ years of experience in Machine Learning (ML) and Robotics. Experienced building production-ready machine learning products leveraging Deep Learning libraries and Azure cloud services. Comprehensive publication record with a best application paper award at IEEE IROS Conference 2015 among 950 paper submissions. Demonstrated leadership skills by co-supervising 18 member team winning first position in Customer Interaction task in World Robotics Summit, Tokyo, Japan 2018.
Work Experience
Senior Data Scientist
- Developed spend analysis API which reduced human efforts by upto 80%. Designed deep learning models to perform extreme multi-label classification of 20,000 labels and engineering pipeline to process over 100 million transaction records. Leveraged Pytorch, Azure Machine Learning, Databricks and Azure Data Factory to engineer the product.
- Developed legal text analysis APIs for contract lifecycle management. Designed deep learning models for language modeling of legal text, clause classification in contracts and clause deviation analysis with respect to a clause library. Leveraged Pytorch, Azure Machine Learning and Azure Kubernetes Services to engineer the product.
- Submitted two patents on original research conducted for contract lifecycle management and spend classification currently review in US Patent Office.
- Conducted knowledge sharing sessions on reinforcement learning, meta learning and Azure machine learning service to team members.
Assistant Professor
- Research on Imitation Learning (IL) to train visual attention system applierd to Human-Robot interaction.
- Supervised 3 PhD students on robotics clothing assistance, Imitation learning to model shepherding behavior and bio-signal processing.
- Collaboration with Denso corporation to apply Maximum Mean Discrepancy for obstacle detection on roads.
- Supervised team for WRS 2018 to control mobile manipulation using Kuka IIWA LBR and custom designed end-effector.
Postdoctoral Researcher
- Novel research on policy gradient RL applied to visual attention tasks for applications such as 3D navigation and HRI.
- Development of simulation environment in Unity game engine as a test-bed for DRL using ML-agents library.
- Developed active vision system on Roomba robot for visual attention and integrated with Jetson TX2 using ROS.
Visiting Student Researcher
- Developed clothing assistance framework with integration of Baxter dual-arm robot and Kinect depth sensor.
- In-charge of research group on assistive robotics co-supervising 3 master students.
- Organized several summer seminars on ROS, Machine Learning using Scipy stack and Gaussian processes using GPy library.
- Introduction collaborative coding to lab with several published projects: Link
Projects
Azure Machine Learning Template
Open-source machine learning project using Azure machine learning as a learning resource to experiment with model auditing and providing ML as a service: Link
Imitation Learning in Pytorch
Open-source project with an implementation of Behavioral cloning algorithm in Pytorch and its integration with OpenAI Gym environments for experimentation: Link
Kinect Baxter Calibration
Open-source project on hand-eye calibration between a Kinect depth sensor and Baxter dual-arm robot using AR markers implemented using Robot Operating System: Link
Publications
Journals
- 2017 · Bayesian Nonparametric Learning of Cloth Models for State Estimation, IEEE Transactions Robotics
- 2019 · Data-efficient Learning of Robotic Clothing Assistance using BGPLVM, Advanced Robotics
- 2020 · Restock System for Retail Automation using Mobile Manipulation, Advanced Robotics
Conferences
- 2013 · Estimation of Human-Cloth Topological Relationship for Robotic Clothing Assistance, ACM AIR Conf.
- 2015 · Cloth Dynamics Modeling in Latent Spaces for Robotic Clothing Assistance, IEEE IROS Conf.
- 2018 · Virtual Reality as a User-friendly Interface for Learning from Demonstrations, ACM SIGCHI Conf.
- 2019 · Toward Imitating Visual Attention of Experts in Software Development Tasks, ACM EMIP Workshop