my face


Ioannis Siglidis (he/they)

I'm a postdoctoral researcher in the Pioneer Center for AI in Copenhagen, working on multimodal data summarization, under the supervision of Serge Belongie. I defended my PhD in Computer Vision on "Image Mining via Synthesis" (2022-2025) at Ecole Des Ponts, advised by Mathieu Aubry, where I collaborated closely with Alyosha Efros and Shiry Ginosar. I'm a suma cum laude graduate of MVA and ECE-NTUA.

My main area of interest is how deep generative models can be used to summarize multimodal data in a way that is both informative and interpretable. Being a polymath my references are interdisciplinary, formed around three central beliefs: a) deep neural networks are the first technology to operationalize human perception outside reductive mathematical models; b) by operationalizing perceptual ontological questions, machine-learning turns modeling into ontology-in-action; c) being statistical sciences of perception, the "arts" and the "humanities" are ideal generators of machine learning problems.

I'm a postdoctoral researcher in the Pioneer Center for AI in Copenhagen, working with Serge Belongie on modeling Digital Narratives. I defended my PhD in Computer Vision on "Image Mining via Synthesis" (2022-2025) funded by a scholarship by the University of Ecole Des Ponts of the eastern Paris Region. There I was part of the IMAGINE lab, advised by Mathieu Aubry, where our work on The Learnable Typewriter, got the best paper award at ICDAR 2024. During the summer of 2023 I visited BAIR under the most loving hospitality of Alyosha Efros, where I made friends along the way. I'm an Antikythera 2024 Cognitive Infrastructure studio alumni, where I learned (from my friend Cezar) that my website has a Prof. Dr. Style, yet its base template comes from Armin Linke. I had the priviledge to work closely for almost 3 years (2020-2022) with the visionary Ilan Manouach doing futurist (AI-comics) projects like the Neural Yorker. My first useful project was GraKeL: A Graph Kernel Library in Python under the supervision of Giannis Nikolentzos in the DaSciM team at Ecole Polytechnique. I hold a degree with a suma cum laude from both the School of Electrical and Computer Engineering of the National Technical University of Athens and the MVA master in Paris of ENS-Paris-Saclay. I'm notorious for being a catophile and for saying /ˈeɪ/ (sounds like the letter "A") yet my good-old friend Chrysa started it.

CV
Contact
Scholar
GitHub
LinkedIn
Twitter
Personal Art

Publications Thesis Talks Art & Literary

(Notation: ⭐ current interest, * equal contribution, 👨🏻‍🎓 advising)

(2024) Diffusion Models as Data Mining Tools

diffmining logo Ioannis Siglidis, Aleksander Holynski, Alexei A. Efros, Mathieu Aubry, Shiry Ginosar
European Conference on Computer Vision (ECCV)

(2024) OpenStreetView-5M The Many Roads to Global Visual Geolocation

grakel logo Guillaume Astruc* Nicolas Dufour* Ioannis Siglidis*, Constantin Aronssohn, Nacim Bouia, Stephanie Fu,
Romain Loiseau, Van Nguyen Nguyen, Charles Raude, Elliot Vincent, Lintao XU, Hongyu Zhou, Loic Landrieu
Computer Vision and Pattern Recognition (CVPR)

(2024) The Learnable Typewriter: A generative approach to text analysis.

ltw logo Ioannis Siglidis, Nicolas Gonthier, Julien Gaubil, Tom Monnier, Mathieu Aubry
Best Paper Award; Internation Conference of Document Analysis and Recognition (ICDAR)

(2021) Graph kernels: A survey

grakel logo Giannis Nikolentzos, Giannis Siglidis, Michalis Vazirgiannis
Journal of Artificial Intelligence Research (JAIR)

(2020) GraKeL: A Graph Kernel Library in Python

grakel logo
Giannis Siglidis, Giannis Nikolentzos, Stratis Limnios, Christos Giatsidis, Konstantinos Skianis,
Michalis Vazirgiannis
Journal of Machine Learning Research (JMLR)

Advising 👨🏻‍🎓


(2026) HiddenObjects: Diffusion-Distilled Dense Spatial Priors for Object Placement

Diffusion Spatial Prior Extractor logo Marco Schouten, Ioannis Siglidis, Serge Belongie, Dim P. Papadopoulos,
Preprint (Arxiv, Under Review)

(2024) An Interpretable Deep Learning Approach for Morphological Script Type Analysis

learnable-handwriter logo Malamatenia Vlachou-Efstathiou, Ioannis Siglidis, Dominique Stutzmann, Mathieu Aubry,
Internation Workshop on Computational Paleography (IWCP)

(2022-2025) Image Mining via Synthesis (PhD Thesis)

phd logo

(2026) When Comics Chat: A Deep Prior for Text-Balloon Reading Order

Comma, KTH, 2026.

(2026) Slop, how did we get here. Lecture on Generative Models.

VIP class, UCPH, 2026.

(2025) Computational Impressionism; Hallucinating as a way of seeing.

DIKU-Bits, UCPH, 2025.

(2025) Computational Impressionism; Hallucinating as a way of seeing.

IMA, NYU Shanghai, 2025.

(2025) Computer Vision at the Mirror Stage: Questioning and Refining Visual Categorization

MPI, Tübingen, 2025.

(2025) For Lack of a Better Word: Mining Visual Variation Behind Labels

Pioneer Center for AI, Copenhagen, 2025.

(2024) Diffusion Models as Data Mining Tools

Google Deepmind, 2024.

(2024) Diffusion Models as Data Mining Tools

Workshop for Visual Concepts, ECCV, Milan, 2024.

(2024) For Lack of a Better Word: Mining Visual Variation Behind Labels

BAIR, Berkeley, 2024.

(2022) The Learnable Typewriter: Unsupervised Text Line Recognition

NYU (remote), 2022.

(2021) Applied Memetic: Comics Generated by Artificial Intelligence

IA fictions, PSL, Université (remote), 2021.

(2019) The stream-graph library for the analysis of temporal social graphs

ISTMAR, Venice, 2019.

(2018) GraKeL: A scikit-learn compatible extension for Graph Kernels

Centrale Supélec, Paris, 2018.

(2025) Antikythera, MIT Press.

antikythera logo
  • Generative Topolinguistics (p. 119)
    Expanding my previous idea of Latent Reading to a whole infrastructure.
  • The Chronoceptual Governor (p. 146)
    Reframing technology as the main mediator of an ecology of timescales.

(2022) Fastwalkers, Ilan Manouach.

fastwalkers logo
The first synthetic comic book co-created with emerging AI, a nonlinear meditation on deep learning that
celebrates the unexpected poetics of generative computing and explores its potential to form new reader sensibilities. It is the outcome of more than a 2-year research effort of trying to create a manga comic driven by the excitement and resources of an early stage of generative modelling, that at that time didn't meet the industrial needs of high-fidelity reconstruction and compositional generalization. A true futurist piece of work that both captures and invents a whole micro-style of imagery, from and as an implemented extra-human metaphor of Manga as big-data. As rare as a picture of a falling star, like certain websites of Web 1.0 it is an essential piece of avant-garde artistry.

(2020 - 2023) The Neural Yorker

grakel logo Synthetic Cartoons in the style of New Yorker on Twitter (~2K followers).
Press: Hyperallergic (2021), iMedd (2025)

(2022) Translation of Capital is Dead from McKenzie Wark, in greek.

grakel logo
A long interview I took from McKenzie Wark about the book in english and in greek.
A single conceptual podcast where I read excerpts in the voice of an AI hybrid of me and McKenzie Wark.

(2021) Latent Reading @ Chimeras: Inventory of Synthetic Cognition

Chimeras logo
Different from "Close Reading" that focuses on the close study of the artifacts of a certain cultural entity and "Distant Reading" that applies computational methods to them by treating them as big-data and extracts high level statistical conclusions and avoids a close qualitative analysis, Latent Reading speculates that to understand such an entity it may be meaningful to do instead a close-reading of the outputs of a generative model trained on its big-data.