iiinova — where your digital self keeps evolving

Hubert Baniecki

@hbaniecki.com

PhD student, University of Warsaw hbaniecki.com

Hubert Baniecki @hbaniecki.com · 25/09/2025

Explaining similarity in vision-language encoders with weighted Banzhaf interactions Check out the paper on arXiv: arxiv.org/abs/2508.05430 Code to be released soon 👆4/4

000

Hubert Baniecki @hbaniecki.com · 25/09/2025

Moreover, we derive three evaluation metrics to facilitate future work in this direction. 𝐅𝐈𝐱𝐋𝐈𝐏 achieves state-of-the-art faithfulness performance across the popular insertion/deletion and pointing game benchmarks. 👇3/4

100

Hubert Baniecki @hbaniecki.com · 25/09/2025

We show that explaining vision–language interactions is essential to faithfully interpret models like OpenAI CLIP & Google SigLIP-2. 𝐅𝐈𝐱𝐋𝐈𝐏 is grounded in cooperative game theory, where we analyze its intriguing properties compared to prior art like Shapley values. 👇2/4

130

Hubert Baniecki @hbaniecki.com · 25/09/2025

🎉 Our paper has been accepted at #NeurIPS2025! @neuripsconf.bsky.social We introduce faithful interaction explanations of CLIP models (FIxLIP), offering a unique perspective on interpreting image–text similarity predictions. 👇1/4

120

Reposted by Hubert Baniecki

Giuseppe Casalicchio @giuseppe88.bsky.social · 11/02/2025

Excited to share that our #ICLR paper, “Efficient & Accurate Explanation Estimation with Distribution Compression” made the top 5.1% of submissions at #ICLR and was selected as a Spotlight! Congrats to the first author @hbaniecki.com #xAI #interpretableML

092

Hubert Baniecki @hbaniecki.com · 30/01/2025

𝗖𝗧𝗘 is a simple yet powerful plug-in for any explainable AI (xAI) method that now relies on i.i.d. sampling. Check out the examples on GitHub! Code: github.com/hbaniecki/co... 5/5 @xai-research.bsky.social

github.com

GitHub - hbaniecki/compress-then-explain: Efficient and accurate explanation estimation with distribution compression (ICLR 2025)

Efficient and accurate explanation estimation with distribution compression (ICLR 2025) - hbaniecki/compress-then-explain

021

Hubert Baniecki @hbaniecki.com · 30/01/2025

𝗖𝗧𝗘 improves the accuracy and stability of explanation estimation with negligible computational overhead, often achieving an on-par error using 2–3× fewer samples, i.e. requiring 2–3× fewer model inferences (⌛ = 💰). 👇4/5

100

Hubert Baniecki @hbaniecki.com · 30/01/2025

𝗖𝗧𝗘 results in more accurate explanations of smaller variance as benchmarked with 4 popular methods (SHAP, SAGE, PDP, Expected Gradients) across 50 datasets and 2 model classes. 👇3/5

100

Hubert Baniecki @hbaniecki.com · 30/01/2025

𝗖𝗼𝗺𝗽𝗿𝗲𝘀𝘀 𝘁𝗵𝗲𝗻 𝗲𝘅𝗽𝗹𝗮𝗶𝗻! We discover a connection between explanation estimation and distribution compression that significantly improves the approximation of feature attributions, importance, and effects. Paper: arxiv.org/abs/2406.18334 👇2/5

arxiv.org

Efficient and Accurate Explanation Estimation with Distribution Compression

We discover a theoretical connection between explanation estimation and distribution compression that significantly improves the approximation of feature attributions, importance, and effects. While t...

100

Hubert Baniecki @hbaniecki.com · 30/01/2025

🚀 Our paper proposing a new paradigm for more efficient estimation of machine learning explanations is accepted at #ICLR2025! This is joint work with Giuseppe Casalicchio, Bernd Bischl & Przemyslaw Biecek, to be presented in Singapore 🇸🇬 👇1/5

151