Mihir Mulye

I am a Machine Learning engineer and have completed Master of Autonomous Systems from the University of Applied Sciences Bonn. I have experience in deep learning frameworks such as TensorFlow and PyTorch as well as libraries such as scikit-learn, opencv and numpy. 

1. I worked as a Junior Machine Learning Engineer at PreciBake GmbH Munich. My primary tasks were to train object detectors primarily YOLO-X and couple them with object trackers namely Observation Centric Simple Online Realtime Tracker (OCSORT) to monitor food items in a given input video sequence. I also worked on integrating OCSORT into the already existing codebase of the company. 
2.  I worked as a Wissenschaftliche Hilfskraft at the Institute of Technology, Resource & Energy Efficient Engineering (TREE). I worked on creating a pipeline for semantic segmentation of the aerial images of a forest area recorded using UAV. For this purpose, I used software tools such as OpenCV, TensorFlow, QGIS & Agisoft Metashape. 

In my Master thesis titled Uncertainty in Explanation methods for Neural Networks, I focussed on combining explanation methods such as Local Interpretable Model-agnostic Explanation (LIME), Guided Backpropagation and Integrated Gradients with uncertainty estimation methods such as Deep Ensembles, Monte Carlo Dropout, Monte Carlo Dropconnect and Flipouts in order to determine the uncertainty in the explanation method output of a neural network. To test the robustness of the proposed concept, the task of regression and image classification were selected with the visual and non-visual modality of input data. This project has introduced me to the domain of explainable AI and interpretable Machine Learning.