Ranjitha Subramaniam

• Analyzed the differences between non-robust and robust CNNs’ weights and internal activations. • Investigated and improved the data augmentation strategies to achieve higher corruption robustness.

• Proposed an effective approach to improve the shape based decision making of Convolutional Neural Networks (CNNs) that are recently shown to exhibit texture-bias in image recognition tasks. • Utilized edge information of images along with an additional in-network component to encode stronger shape bias. • Studied the significance of shape bias on common image corruptions and established that there exists no clear correlation between shape bias of a CNN and its robustness against corruptions.

• Supported the development of a Modular Data Analysis Framework for optimizing the injector geometry selection technique. • The framework is developed in an Object Oriented Programming paradigm and it involved the implementation of different Machine Learning algorithms. • Developed a unit testing framework using Pytest to assess the correctness of the Data Analysis Framework. • Created conda packages for effortless distribution and installation of the framework.

• Design and development of software for functional safety requirements in Powertrain Electronic Control Units (ECUs). • Verification and validation of code through unit and functional tests. • System test of the ECU software in Hardware-in-the-loop (HIL) systems. • Analysis, Design, Implementation and Automation of system tests. • Industrialization of Series Release Software for Engine ECUs.

Machine Learning, Computer Vision, Multicore Programming, Automotive Software Engineering Lab, Software Platform for Automotive Systems, Design of Software for Embedded Systems, Hardware-Software Co-design, Dependable Systems

Embedded Systems, Digital Signal Processing, Microprocessors & Microcontrollers, Computer Architecture, Programming Languages (C, C++)