Our paper "Towards a Real-World Aligned Benchmark for Unlearning in Recommender Systems" is accepted at the #RecSys2025 FAccTRec workshop. 🎉

We propose a realistic benchmark for machine unlearning that addresses real-world privacy challenges like GDPR's "right to be forgotten".

@bifold.berlin

Thanks to my supervisor
@mersault.bsky.social!

Paper: openreview.net/pdf?id=JJYHb...
Code: github.com/OlgaOvcharen...

Cross-modal errors span across multiple modaltities, like mismatched images or incorrect text fields, and are are an overlooked but critical issue in domains like e-Commerce and healthcare. Even with modern tools, large orgs struggle to catch cross-modal errors.

✨ Excited to present our workshop paper at DataWorld at #ICML2025 tomorrow 🇨🇦

We introduce the problem of detecting cross-modal errors in tabular data that originate from other modalities.

Visit our poster:
📅 Saturday, July 19, 10:05 AM - 11:20 AM
📍 West Meeting Room 208-209

Thanks to all co-authors @flobarkmann.bsky.social, Philip Toma, Imant Daunhawer, Julia Vogt, @mersault.bsky.social, and @valboeva.bsky.social.

📄 Full paper: openreview.net/pdf?id=jnPHZ...
💻 Code: github.com/BoevaLab/scS...

📊 Key findings:
- Specialized single-cell frameworks and scFMs excel at uni-modal batch correction
- Generic SSL methods like VICReg & SimCLR outperform on cell typing & multi-modal tasks
- Masking emerges as the most universally effective augmentation strategy

🎉 Happy to present our spotlight paper at #ICML2025 tomorrow 🇨🇦

We introduce scSSL-Bench, a benchmark evaluating 19 SSL methods across 9 single-cell multi-omics datasets🧬 and 3 tasks.

Visit our poster:
📅 Thu Jul 17, 11 AM - 1:30 PM PDT
📍 West Exhibition Hall B2-B3 #W-311

The DEEM Lab is at ICML this week for the first time, with two contributions!

(1/3)

Our paper "Towards Cross-Modal Error Detection with Tables and Images" was accepted for the DataWorld workshop at ICML'25! 🥳

Thanks to @mersault.bsky.social!

We have a PhD opening in Berlin on "Responsible Data Engineering", with a focus on data preparation pipelines that optimize ML models along responsibility objectives.

This is a fully-funded position at @bifold.berlin, co-supervised by Julia Stoyanovich from NYU.

Details: deem.berlin#jobs-17725

We have a PhD opening in Berlin on "Responsible Data Engineering", with a focus on data preparation for ML/AI systems.

This is a fully-funded position with salary level E13 at the newly founded DEEM Lab, as part of @bifold.berlin .

Details available at deem.berlin#jobs-2225

📢 Our extended benchmark on self-supervised learning for single-cell data, scSSL-Bench 🧬, is now accepted at ICML (spotlight)!

Thanks to all collaborators from @bifold.berlin and @ethzurich.bsky.social!

We have openings for student assistants in the DEEM Lab at @bifold.berlin. This is a great opportunity to work with PhD students, implement cool stuff, gather research experience and become a co-author of scientific publications :)

deem.berlin#jobs-193487

Our vision "Towards Regaining Control over Messy ML Pipelines" was accepted for the DAIS workshop at ICDE! 🥳

Initial experiments show LLMs are promising for extracting declarative query plans from messy ML code.

Joint work w/ @guangchen811.bsky.social @oovcharenko.bsky.social @mersault.bsky.social

We have a PhD opening in Berlin on "Responsible Data Engineering", with a focus on efficiently creating, maintaining and evaluating datasets and pipelines for ML use cases.

This is a fully-funded position at the newly founded DEEM Lab, as part of @bifold.berlin .

deem.berlin#jobs-2225

📢If you are interested in single-cell foundation models (scFMs), stop by our poster (West 109) at the AiDrugX Workshop at Neurips 2024. We will present CancerFoundation, a scFM tailored for studying cancer biology🧬.
Preprint: biorxiv.org/content/10.1...

📄 Read the full preprint: biorxiv.org/content/10.1...

Thanks to all co-authors Philip Toma, Imant Daunhawer, Julia Vogt, @flobarkmann.bsky.social, and @valboeva.bsky.social

Furthermore, we find that masking leads to the biggest performance improvement for creating augmented views of a cell.

Our bechmark shows that SimCLR and VICReg are the best methods for single-cell data across batch integration, query-to-reference, and missing modality prediction tasks.

📢 Our benchmark on self-supervised learning for single-cell data🧬 is accepted at the #NeurIPS2024 SSL workshop. We take a first step towards establishing best practices for SSL methods for single-cell data, and benchmark 8 SSL methods on 3 downstream tasks across 8 datasets.