Selected Publications

AI Debaters are More Persuasive when Arguing in Alignment with Their Own Beliefs
Conference paper

Poster @ NeurIPS 2025

This paper explores how LLMs perform in AI debates when their arguments align or conflict with their prior beliefs. By measuring models’ priors and comparing debate protocols, we find that models are more persuasive when defending positions consistent with their beliefs, offering insights into persuasion dynamics and bias in scalable oversight.

AI Debate Scalable Oversight Persuasiveness AI Safety Large Language Models Alignment

SignAttention: On the Interpretability of Transformer Models for Sign Language Translation
Conference paper

Spotlight @ NeurIPS 2024

This paper studies a Transformer-based model for translating Greek Sign Language to glosses and text. It finds that the model focuses on clusters of frames, with alignment weakening as glosses increase, and shifts from video frames to predicted tokens during translation. This work enhances understanding of SLT models and promotes more transparent translation systems.

Sign Language Translation Transformers Attention Mechanism Greek Sign Language Interpretability

VISTA: Vision Improvement via Split and Reconstruct Deep Neural Network for Fundus Image Quality Assessment
Journal paper

This paper introduces a deep learning model for assessing fundus image quality, essential for diagnosing eye conditions like cataracts and diabetic retinopathy. The model preserves high resolution and includes an autoencoder for reconstruction and classification. Results on the EyeQ dataset show 90.66% accuracy, 88.43% precision, 89.05% recall, and an F1-score of 88.68%.

Autoencoder Network Explainability Fundus Image Gradability Interpretability Quality Assessment Retinal Image

CB-RISE: Improving the RISE Interpretability Method Through Convergence Detection and Blurred Perturbations
Conference paper

This paper enhances the RISE algorithm for image interpretation by introducing C-RISE, which reduces computation time, and CB-RISE, which uses blurred masks instead of black patches. These improvements lead to better heatmap quality and a speedup of about three times.

Black Box Blurred Masks Computer Vision Convergence Detection Deep Learning Interpretability RISE

Bringing balance to hand shape classification: Mitigating data imbalance through generative models
Journal paper

This paper improves hand-shape classification in sign language by using GAN-generated images for pre-training. This boosts accuracy (up to 5% overall, 100% on rare classes), halves training time, and enables cross-dataset generalization. Pre-training with synthetic data outperforms other augmentation methods.

Handshape Recognition Unbalanced Data Limited Data Sign Language Generative Adversarial Networks

The Implementation of the RISE Algorithm for the Captum Framework
Conference paper

This paper introduces an efficient implementation of the RISE method in PyTorch using the Captum library. RISE generates explanations for deep learning model predictions by randomly masking input parts. Experiments highlight the significance of the number and size of masks for achieving accurate and versatile interpretations.

Black Box Computer Vision Deep Learning Interpretability RISE

LSA-T: The First Continuous Argentinian Sign Language Dataset for Sign Language Translation
Conference paper

LSA-T is the first continuous Argentinian Sign Language (LSA) dataset. It contains ~22 hs of video extracted from the CN Sordos YouTube channel with Spanish subtitles, joints for each signer, and the inferred signer if there is more than one person in a clip. Videos are in 30 FPS full HD (1920x1080). We've also developed https://app.seni.ar, a platform for exploring, validating, and augmenting LSA-T.

Computer Vision Dataset Deep Learning LSA-T Sign Language Translation

AI Debaters are More Persuasive when Arguing in Alignment with Their Own Beliefs Conference paper Poster @ NeurIPS 2025

SignAttention: On the Interpretability of Transformer Models for Sign Language Translation Conference paper Spotlight @ NeurIPS 2024

VISTA: Vision Improvement via Split and Reconstruct Deep Neural Network for Fundus Image Quality Assessment Journal paper

CB-RISE: Improving the RISE Interpretability Method Through Convergence Detection and Blurred Perturbations Conference paper

Bringing balance to hand shape classification: Mitigating data imbalance through generative models Journal paper

The Implementation of the RISE Algorithm for the Captum Framework Conference paper