Frontier Series 2026 — Örebro University and Lund University

Radoslaw Martin Cichy examines three contributions of empirical vision research to AI: visual representations developing from infancy to adulthood; feedforward and recurrent processing distinguished using layer-specific fMRI with EEG; and difficult object recognition recruiting cortex-wide recurrence. These findings test the limits of vision models inspired primarily by visual cortex.

Subcallosal cingulate stimulation for resistant depression needs objective biomarkers beyond stress-sensitive clinical ratings. Sankaraleengam Alagapan uses variational autoencoders to address high-dimensional longitudinal local-field potentials and unreliable labels. Supervised models track depressive states; a generative causal explainer identifies interpretable, low-dimensional features, advancing actionable biomarkers for stimulation management.

Brain function and psychiatric symptoms evolve dynamically. Manuel Brenner presents self-learning mathematical models that infer hidden states and patterns from brain activity and behaviour. Developed within LOEWE DYNAMIC, the interpretable models connect artificial intelligence, basic neuroscience and clinical psychiatry, seeking a path from temporal-data modelling to clinical application.