Byzantine Seals
Computer vision and multimodal methods for interpreting and restoring Byzantine seals.
- Computer Vision
- Multimodal Learning
- Image Restoration
ML Research · Paris
I’m Saranga, a machine-learning researcher interested in computer vision, multimodal learning, generative modeling, and representation learning. My work so far has involved projects across healthcare, scientific machine learning, cultural heritage, text, and audio.
From ancient seal impressions and blood-smear microscopy to the hidden structure of the cosmos and the acoustics of musical instruments.
Computer vision and multimodal methods for interpreting and restoring Byzantine seals.
Smartphone-embedded deep learning for microscopic and sub-microscopic malaria detection.
Machine-learning methods for strong gravitational lensing and dark-matter substructure analysis.
Neural-network methods for recognizing and classifying musical instruments from audio.
* Equal contribution where indicated in the author list.
Anis Bourou*, Saranga Kingkor Mahanta*, Thomas Boyer, Auguste Genovesio
Under review at MICCAI 2026@article{bourou2025diffex,
title={DiffEx: Explaining a Classifier with Diffusion Models to Identify Microscopic Cellular Variations},
author={Bourou, Anis and Mahanta, Saranga Kingkor and Boyer, Thomas and Mezger, Valérie and Genovesio, Auguste},
journal={arXiv preprint arXiv:2502.09663},
year={2025},
eprint={2502.09663},
archivePrefix={arXiv}
}
Nicolas Bourriez*, Saranga Kingkor Mahanta*, Ivan Svatko, Eleanor Lacassagne, et al.
medRxiv (preprint)@article{bourriez2025malaria,
title={Detection of Malaria Infection from parasite-free blood smears},
author={Bourriez, Nicolas and Mahanta, Saranga Kingkor and Svatko, Ivan and Lacassagne, Eleanor and Atchade, Aurore and Leonardi, Francesco and Cohen, Ethan and Massougbodji, Achille and Argy, Nicolas and Cottrell, Gilles and Genovesio, Auguste},
journal={medRxiv},
year={2025},
doi={10.64898/2025.12.29.25343125}
}
Saranga Kingkor Mahanta, Nihar Jyoti Basisth, Eisha Halder, Abdullah Faiz Ur Rahman Khilji, Partha Pakray
Evolving Systems (2023)@article{mahanta2023exploiting,
title={Exploiting Cepstral Coefficients and CNN for Efficient Musical Instrument Classification},
author={Mahanta, Saranga Kingkor and Basisth, Nihar Jyoti and Halder, Eisha and Khilji, Abdullah Faiz Ur Rahman and Pakray, Partha},
journal={Evolving Systems},
year={2023},
doi={10.1007/s12530-023-09540-x}
}
Swagat Shubham Bhuyan*, Saranga Kingkor Mahanta*, Partha Pakray, Benoit Favre
Natural Language Processing Journal (2023)@article{bhuyan2023textual,
title={Textual Entailment as an Evaluation Metric for Abstractive Text Summarization},
author={Bhuyan, Swagat Shubham and Mahanta, Saranga Kingkor and Pakray, Partha and Favre, Benoit},
journal={Natural Language Processing Journal},
volume={4},
pages={100028},
year={2023},
doi={10.1016/j.nlp.2023.100028}
}
Saranga Kingkor Mahanta, Shubham Jain, Hoang Van Truon, Darsh Kaushik, Koushik Guha
First International Conference on Advances in Computing and Future Communication Technologies · IEEE Xplore (2021)@inproceedings{mahanta2021covid,
title={COVID-19 Diagnosis from Cough Acoustics using ConvNets and Data Augmentation},
author={Mahanta, Saranga Kingkor and Jain, Shubham and Truong, Hoang Van and Kaushik, Darsh and Guha, Koushik},
booktitle={2021 First International Conference on Advances in Computing and Future Communication Technologies (ICACFCT)},
year={2021},
doi={10.1109/ICACFCT53978.2021.9837350}
}
Saranga Kingkor Mahanta, Abdullah Khilji, Partha Pakray
Computación y Sistemas (2021)@article{mahanta2021deep,
title={Deep Neural Network for Musical Instrument Recognition using MFCCs},
author={Mahanta, Saranga Kingkor and Khilji, Abdullah Faiz Ur Rahman and Pakray, Partha},
journal={Computación y Sistemas},
volume={25},
number={2},
pages={351--360},
year={2021},
doi={10.13053/CyS-25-2-3946}
}
Feb. 2026 – Present
Sorbonne Université
Applying computer vision and multimodal methods to interpret and restore Byzantine seals, in collaboration with historians.
Paris, FranceJan. 2025 – Jan. 2026
Institut de Biologie de l’École Normale Supérieure (IBENS)
Developed smartphone-embedded deep-learning models and an app for microscopic and sub-microscopic malaria detection.
Paris, FranceApr. 2024 – Oct. 2024
Institut de Biologie de l’École Normale Supérieure (IBENS)
Proposed diffusion-model counterfactual explanations for interpreting classifier decisions on biological datasets.
Paris, FranceMay – Sep. 2023 · Jun. – Sep. 2022
ML4SCI
Worked on regression, classification, anomaly detection, and dark-matter substructure analysis for DeepLense.
RemoteApr. 2023 – Jul. 2023
DXOMARK
Stress-tested image-quality assessment models and built an automated evaluation pipeline.
Boulogne-Billancourt, FranceJul. 2021 – Sep. 2021
NIT Silchar · Aix-Marseille University
Developed textual-entailment methods for evaluating semantic consistency in abstractive summaries.
Silchar, India · Marseille, FranceFeb. 2026 – Present
Sep. 2022 – Oct. 2024
Aug. 2018 – Jun. 2022
Diffusion-model counterfactual explanations for identifying microscopic cellular variations.
Read more ↗Textual entailment as a measure of semantic consistency in abstractive summaries.
Read more ↗Implemented neural style transfer from scratch in PyTorch and compared it with a pretrained TensorFlow Hub model.
Read more ↗Trained vanilla and variational autoencoders on the Free Spoken Digit Dataset and studied their performance.
Read more ↗Built a CNN to classify five facial expressions and run real-time recognition through a webcam with OpenCV.
Read more ↗Fine-tuned VGG19 to build a real-time face-mask detector using OpenCV and Keras/TensorFlow.
Read more ↗Used K-means clustering to identify dominant image colours and retrieve images using colour names.
Read more ↗Applied latent Dirichlet allocation to dialogue topics and generated day-wise WordClouds from WhatsApp chats.
Read more ↗Built a system that extracts title, author, and year from webpages to generate Chicago-style reference strings.
Read more ↗Combined multinomial Naive Bayes with additional checks for gibberish and selected abusive language.
Read more ↗Combined multiple regression models through weighted averaging to predict restaurant delivery times.
Read more ↗Built a SARIMA forecasting instance for bribery-case counts for a Smart India Hackathon project.
Read more ↗Full scholarship from the Embassy of France in India to pursue a Master’s degree in France.
Track 1: COVID-19 diagnosis from cough sound recordings.
Restaurant delivery-time prediction problem by Blue Yonder.
Built a spam classifier, SARIMA forecasting instance, and chatbot for the main website.
Beyond research · Personal
I often travel and seldom write...
To realize our dreams, we must first learn to normalize them.
— SKM