AI4GRAPES selected as an AI-BOOST Spark-phase winner for NL2MISSION
Athena Research Center announced on 5 November 2026 that AI4GRAPES, a team from its Institute… Continue Reading AI4GRAPES selected as an AI-BOOST Spark-phase winner for NL2MISSION
Research Highlights
Selected research outputs, publications and technical milestones from MediaLab projects.
TMLR 2026: prescriptive SVD-inspired attention via spectral energy retention
The paper “Prescriptive SVD-Inspired Attention via Spectral Energy Retention” has been published in Transactions on… Continue Reading TMLR 2026: prescriptive SVD-inspired attention via spectral energy retention
Towards Heritage World Models: from digital twins to predictive heritage systems
The article “Towards Heritage World Models” extends the discussion around ARGUS, PANOPTES, semantic digital twins… Continue Reading Towards Heritage World Models: from digital twins to predictive heritage systems
ICASSP 2026: representation-diverse self-supervision for bioacoustic learning
“Representation-Diverse Self-Supervision for Cross-Domain Bioacoustic Learning in Low-Resource Settings” was presented at ICASSP 2026. The… Continue Reading ICASSP 2026: representation-diverse self-supervision for bioacoustic learning
AISTATS 2026: process-tensor tomography of SGD and training memory
The paper “Process-Tensor Tomography of SGD: Measuring Non-Markovian Memory via Back-Flow of Distinguishability” was published… Continue Reading AISTATS 2026: process-tensor tomography of SGD and training memory
ICLR 2026: measuring the hidden geometry of learned representations
Representation holonomy examines how neural features change along input paths, extending analysis beyond pointwise similarity. Continue Reading ICLR 2026: measuring the hidden geometry of learned representations
ICPRAM 2026: two-stage angular alignment for positive-unlabeled learning
“Two-Stage Angular Alignment for Positive-Unlabeled Learning” proposes a geometric weak-supervision method for settings where only… Continue Reading ICPRAM 2026: two-stage angular alignment for positive-unlabeled learning
IEEE Access 2026: interpretable Vision Transformers via SVDA
“Interpretable Vision Transformers in Image Classification via SVDA” extends the group’s SVD-inspired attention work to… Continue Reading IEEE Access 2026: interpretable Vision Transformers via SVDA
TMLR: angular regularization for positive-unlabeled learning
The paper “Angular Regularization for Positive-Unlabeled Learning on the Hypersphere” received TMLR certification and was… Continue Reading TMLR: angular regularization for positive-unlabeled learning
NeurIPS 2025: Auto-Compressing Networks for efficient neural representations
Auto-Compressing Networks explore how architectural design can reduce redundancy while retaining useful learned representations. Continue Reading NeurIPS 2025: Auto-Compressing Networks for efficient neural representations