Next Tuesday, I will take part in the eXplainable AI Day 2024. The event will consist of ten presentations by people from the AI industry and academia, including my talk focused on explainability in computer vision. The conference is online and free: bit.ly/XAIDay-2024-3

Are the explanations given by interpretable image recognition models reliable? If you are at #AAAI2024 in Vancouver this year, feel free to come to see our poster on Saturday's session from 6:15 to 8:15 PM. Otherwise, check out our paper arxiv.org/abs/2308.08162.

We will present our newest work on retrosynthesis modeling at Molecule.one at the ML4Molecules workshop tomorrow. The workshop is held online and free, so see you there!

Exciting news! Molecule.one is thrilled to announce our participation at #NeurIPS2023 in New Orleans, a premier conference for machine learning and AI. We will be presenting: Molecule-Edit Templates for Efficient and Accurate Retrosynthesis Prediction at 2 workshops

I am thrilled to present our work from Uniwersytet Jagielloński at the upcoming #WACV2023 ! Our model, ProtoSeg, adapts the Prototype Networks architecture to learn interpretable image segmentation. Check out the paper here: openaccess.thecvf.com/content/WACV20…

If you don't remember retweeting this, it means that you have leaked your Twitter Access Token in a public GitHub repository. Not the best practice, right?
For details, read our latest article: link.medium.com/L1VvZmhb5mb
#InfoSec #CyberSecurity #GitHub

Our work on MEGAN retrosynthesis prediction model has just been published at @JCIM_ACS . The paper includes more extensive evaluation and comparison to recent methods. Molecule.one

MEGAN: Modeling Chemical Reactions as Sequences of Graph Edits. Our poster about recent research at Molecule.one for the 3rd
RSC BMCS / RSC_CICAG AI in Chemistry meeting #AIChem20posters

We would like to invite everyone to the warsaw.ai meeting which takes place today at 6 PM CEST. Mikołaj Błaż from Molecule.one will be speaking about how we can leverage AI-powered synthesis planning against COVID-19. The event is remote and for free.

Training a Language Model on proteins using ~5600 GPUs 😮. It achieves 'slight' improvement - so maybe lots of data and absurd amount of computational power do not automatically guarantee outstanding results?

Happy to be coauthoring the first research published by Molecule.one.

Our new graph-based, template-free approach to retrosynthesis is now online, showing strong performance on the USPTO-50k single-step benchmark task. Work led by Vignesh Ram Somnath in collaboration with Charlotte Bunne Andreas Krause Regina Barzilay | arxiv.org/abs/2006.07038 #compchem MIT Jameel Clinic for AI & Health

MEGATHREAD TIME: In 40 tweets I will describe 40 powerful concepts for understanding the world. Some are complex so forgive me for oversimplifying, but the main purpose is to incite curiosity. Okay, here we go: