Jiri Sedlar

I'm a research scientist at the Czech Institute of Informatics, Robotics and Cybernetics (CIIRC), Czech Technical University in Prague, in the IMPACT group led by Josef Sivic.

I did my PhD at Charles University / Czech Academy of Sciences in Prague, where I was advised by Jan Flusser, followed by postdoc stays at Flinders University in Adelaide, Australia, and Mayo Clinic in Rochester, MN, USA.

Email  /  Google Scholar  /  arXiv  /  LinkedIn

News

ICLR 2024: Scientists from CIIRC CTU empower the computational design of biotherapeutics

Research

I'm interested in machine learning, protein engineering, robotics, computer vision, and image processing.

Introducing the task of protein flexibility engineering. Survey and comparison of methods for measuring/predicting protein flexibility. Introducing novel flexibility predictors based on sequence or sequence and structure, utilizing protein language models and Elastic Network Models. Proposing a training framework for fine-tuning of protein inverse folding models to make them steerable by requirements on the flexibility of the output sequence.

The introduced CoVAMPnet method allows interpretation and comparison of Markov state models of molecular dynamics, as demonstrated on a case study comparing the dynamics of three different systems of Aβ42 peptide.

Introducing a large dataset of protein-protein interfaces and a model for predicting the change in binding affinity of two protein structures upon mutation (ddG prediction).

We demonstrate that further development of machine learning methods may be hindered by the quality of existing train-test splits. To overcome the data leakage, we recommend constructing data splits based on 3D structural similarity of protein-protein interfaces and suggest corresponding algorithms.

A perspective and survey on challenges in protein engineering and progress enabled by machine learning.

A dataset for training and evaluating methods for 6D pose estimation of hand-held tools in task demonstrations captured by a standard RGB camera.

A visual recognition system that takes as input video frames together with a simple object model, and outputs a 3D motion of the person and the object, as well as contact forces and torques actuated by the human limbs.

A purely image-based method for finding ruler and compass constructions in the Euclidea geometric game.

A new pose verification approach combining appearance, geometry and semantics to improve large-scale indoor camera localization.

Introducing a new method to automatically reconstruct the 3D motion of a person interacting with an object from a single RGB video.

This page is using the design provided by Jon Barron.