Ananta Bhattarai

- Implemented image explanation methods for classification and segmentation models - Designed a safety validation framework for autonomous trains with GANs (Work got accepted at SAFECOMP 2023) - Proposed a novel approach to improve faithfulness and stability of explanation methods (Work got accepted at ICML 2024)

- Implemented a pipeline for data processing, model training, and deployment - Developed a U-Net segmentation model that segments lesions in an MRI Spine scan with a Dice Score of 0.83 - Modified BioBERT language model such that it can answer whether radiologist reports indicate abnormalities in the Spine with an accuracy of 93%

- Developed an ML model to detect real-time changes in returns co-movements in digital asset markets - Implemented and integrated the model into the existing pipeline for simulation - Improved risk management and enhanced cross-sectional investment strategies with the deployed model

- Teaching Assistant for Intelligent Mobile Systems and Machine Learning Courses - Research with Prof. Francesco Maurelli on autonomous driving algorithms for Duckietown

- Proposed a novel approach to learning programmable potentials using deep learning - Led the experiments by running hyperparameter optimization and verifying the robustness of the model (Work got accepted at AAAI Spring Symposium: MLPS 2021)

- Proposed an ML model to perform unsupervised clustering of robot trajectories - Implemented the proposed method, combining an LSTM auto-encoder and Gaussian mixture model

- Focus Area: Computer Vision, Machine Learning - Thesis: 3D GAN Inversion with Deep Learning