Sayan K Chaki

"

Final-year doctoral candidate specialising in computer vision and digital humanities, with expertise in unsupervised analysis of historical ornaments. Collaborated with University of Oxford's VGG, IMAGINE École des Ponts ParisTech, and TU Vienna on computational analysis of cultural artefacts.

Professional Experience

PhD Scholar in Image, Vision, and Signal Processing

Inria, Laboratoire Hubert Curien | Sep 2022– Present

  • Developing an equivariant model for unsupervised detection and segmentation of ornaments from historical manuscripts
  • Affine transformation invariance with lowdata solutions
  • Equivariance principles for model robustness in ANR project "Rey's Ornament Image Investigation"

Research (Master) Intern

Laboratoire Hubert Curien | Mar 2022– Jul 2022

  • Developed Spatial Transformer AutoEncoder for anomaly segmentation
  • Compiled benchmark vignettes dataset from ornaments in books

Software Developer Intern

IBM India & KLA Tencor | May 2021– Jul 2021

  • Developed anomaly segmentation algorithms for 3D fault detection
  • Created an integrated crossplatform ecosystem using React

Research & Publications

Rotation Equivariant Detector

Making Spatial Transformer Networks Rotation-Equivariant: Variational Autoencoding to Object Detection

Submitted to ICML 2025

Led a team of researchers in addressing limitations in 3D reconstructions and modelling of byzantine seals

Historical Ornaments

Historical Printed Ornaments Dataset and Tasks -

Published in ICDAR 2024

View Project Page

Anomaly localization and Pattern Comparison

One-to-many pattern comparison combining fully-connected autoencoder with spatial transformer for ornament investigation

View Paper

Computing the Early Page in Modern Europe

University of Oxford - March 2024

Invited talk on unsupervised decomposition of historical document elements, focused on early page structures of ancient European books

Intermediality and Computational Humanities Hackathon

University of Vienna - November 2024

Led a team of researchers in addressing limitations in 3D reconstructions and modelling of byzantine seals

Teaching Experience

Introduction to Machine Learning

Institut d'Optique Graduate School (24 hrs)

  • Supervised/unsupervised learning, neural networks (CNNs, GANs, VAEs)
  • Model evaluation with practical Python implementations

Introduction to Python

Institut d'Optique Graduate School (24 hrs)

  • Python basics, data structures, libraries (NumPy, pandas)
  • Applications in data analysis and machine learning

Summer Schools

Google DeepMind ProbAI Summer School

NTNU, Trondheim, Norway | June 2023

  • Advanced probabilistic AI techniques in generative modeling
  • Presented poster on probabilistic methods in document analysis

Generative Modelling Summer School (GeMSS)

TU Eindhoven, DTU, Inria | June 2024

  • Advanced methods in generative modeling and variational inference
  • Presented poster on applications in digital humanities