Benjamin Robson

Machine learning researcher at Aalto University, working on self-supervised and multimodal representation learning, supervised by Arno Solin.

Previously a software engineer at Apple, Instagram, WhatsApp, Microsoft Research, and a few startups.

Portrait of Benjamin Robson

Publications

2026

LeAVJEPA audio-to-video attention heatmap highlighting an oboe

LeAVJEPA: A Minimalist Architecture for Audio-Visual Self-Supervised Learning

A single shared transformer learns aligned audio and video representations using modality dropout and the LeJEPA objective, supporting classification and zero-shot audio-visual retrieval.

Project page arXiv

PCA of AV-JEPA video patch-token features on a violin frame

AV-JEPA: Extending LeJEPA to Audio-Visual Self-Supervised Learning

An early-fusion transformer with modality dropout as masking: no decoder, no EMA teacher, no contrastive negatives. Accepted at the Machine Learning for Audio workshop at ICML 2026.

arXiv