Bharath Bhushan Damodaran Email & Phone Number

Rennes, Brittany, France

Neural image/video compression (coding): In this project, I am working on learning based compression techniques using variational auto-encoder and implicit neural representation based methods to improve the rate-distortion performance over the traditional and SoA nerual codecs. Publications:- BB Damodaran et.al, "RQAT-INR: Improved Implicit Neural Image Compression", Data Compression Conference (DCC) 2023.- M Balcilar, BB Damodaran, LATENT-SHIFT: Gradient of Entropy helps Neural Codecs, ICIP 2023.- M Shukor, BB Damodaran et.al, "Video coding using learned latent gan compression", Proceedings of the 30th ACM International Conference on Multimedia, 2022.- M Balcilar, BB Damodaran, P.Hellier, "Reducing the mismatch between marginal and learned distributions in neural video compression", IEEE International Conference on Visual Communications and Image Processing (VCIP), 2022.- M Balcilar, BB Damodaran, P.Hellier, "Reducing the amortization gap of entropy bottleneck in end-to-end image compression", Picture Coding Symposium (PCS) 2022.- M Shukor, X Yao, BB Damodaran, "Semantic Unfolding of StyleGAN Latent Space", ICIP 2022.Applied Machine learning for post-production : Sparse high-dimensional data embedding for video editing in post-production: In this project, I worked with a team (including post-production artists) developing and prototyping semi-automated efficient tools using machine learning for the movie post-production industry. Mainly, I was involved in developing a robust sparse high-dimensional data embedding method to capture the local and global geometrical structure of the motion tracks. Publications:BB Damodaran et.al, "FacialFilmroll: High-resolution multi-shot video editing", Proceedings of the 18th ACM SIGGRAPH European Conference on Visual Media Production (CVMP), 2021 (Received Best Paper Award)

Vannes Area, France

Post doctoral research scientist at IRISA, OBELIX in the project "optimal transport and machine learning". I developed techniques to regularize deep neural networks by injecting the geometrical structure of the data in the deep neural networks through optimal transport, to generalize deep neural networks in several real-world scenarios. The specific application domain includes: transfer learning for image classification, segmentation segmentation, semi-supervised learning, learning with inaccurate labels, adversarial regularization. The outcome of my research has been published in leading machine learning conferences and journals. For additional details please see my google scholar page.Keywords: Deep Learning, semantic segmentation, learning with noisy labels, object detection, optimal transport, semi-supervised learning, unsupervised domain adaptationPublished papers:- Large scale optimal transport and mapping applications (ICLR 2018)- Deep Joint distribution optimal transport for unsupervised domain adaption applications (ECCV 2018)- Entropic optimal transport loss for learning deep neural networks with noisy labels (Journal of Computer Vision and Image Understanding, 2019)- Adversarial training for learning with label noise (IEEE PAMI, 2021)The developed method has been implemented in Keras and PyTorch, and tested on several real-world complex datasets. The codes are deployed in the high-performance clustering environments.

Vannes Area, France

Prestige Marie Curie Research Fellow, co-financed under the Marie Curie Actions-COFUND of the FP7.I have developed several machine learning methods/frameworks for advancing the processing of high dimensional Earth observation datasets. - Developed (Implemented in R, and Matlab) a feature/variable selection method based kernel methods and LASSO for high dimensional datasets to solve the curse of dimensionality problem, and evaluated with several classification methods: SVM, PerTurbo, Extreme learning machine, Naive Bayes Classifier- Developed (Implemented in R, C++) a scalable machine learning framework using hierarchical image features, and Random Forest with active learning/semi-supervised learning for detection of woody features at European scale- Developed (Implemented in Python, R, Matlab) scale kernel method using random Fourier featuers for large scale classification, regression, and feature extraction (dimensionality reduction) problems. The codes are deployed in the high-performance clustering Environments.

Thiruvananthapuram Area, India

I have completed Ph.D in "Multiple Classifier System for Hyperspectral Image Classification". In my thesis, I developed to the framework to dynamically select the classifiers according to the input data sources from a pool of classifiers. I have proposed several combination schemes to effectively leverage the complementary benefit from the several classifiers to increase the classification performance, robustness, and reliability, and to solve curse of dimensionality. The developed methods are implemented in Matlab, and have been evaluated with real world applications including monitoring of mangroves. Keywords: Ensemble learning, classifier combination, input level classifier selection, Kernel methods, linear models, Markov random Field, KNN, NN, Beta distribution

Chennai Area, India

Junior project associate in the Institute of Remote Sensing working on the ISRO funded project in Space based information support for decentralized planning. The responsibility includes in image registration of the Cartosat images, map digitization and registration through control points.

Doctor Of Philosophy (Ph.D.), Remote Sensing, Machine Learning

Master’S Degree, Remote Sensing And Wireless Sensor Networks, 8.6 Cgpa

Completed M.Tech in Remote Sensing and wireless sensor networks from Amrita Vishwa Vidyapeetham, Coimbatore. Gained experience in Image.

Master’S Degree, Mathematics