Caligiore, D. | Search bioRxiv

bioRxiv2026

Explainable machine learning identifies candidate shared neuroanatomical features in Alzheimer's and Parkinson's via importance inversion transfer

Despite significant neurobiological and pathological overlaps, Alzheimers and Parkinsons diseases--the primary threats to healthy aging--are still managed as distinct clinical entities. Standard machine learning exacerbates this diagnostic fragmentation by prioritizing divergent markers over shared traits, thereby obscuring the invariant foundations of neurodegeneration. This study introduces Importance Inversion Transfer, an explainable machine learning framework designed to identify neuroanatomical invariants across the neurodegenerative spectrum. Prioritizing structural stability over discriminative utility isolates a shared pathological core consisting of ten regional volumetric anchors, validated through an inductive protocol with high diagnostic fidelity (AUC = 0.894). The identified morphological continuum between healthy aging and neurodegeneration delineates shared structural substrates consistent with--though not demonstrative of--a potential common early-phase vulnerability. Aligned with the Neurodegenerative Elderly Syndrome hypothesis, this evidence establishes a possible paradigm for early, system-level diagnosis.

neuroscience↗

bioRxiv2022

Interactions between supervised and reinforcement learning processes in a neurorobotic model

Several influential works propose that the acquisition of motor behavior involves different learning mechanisms in the brain, in particular supervised and reinforcement learning, that are respectively associated with cerebellar-thalamocortical and basal ganglia-thalamocortical networks. Despite increasing evidence suggesting anatomical and functional interactions between these circuits, the learning processes operating within them are studied in isolation, neglecting their strong interdependence. This article proposes a bio-inspired neurorobotic model implementing a possible cooperation mechanism between supervised and reinforcement learning. The model, validated with empirical data from healthy participants and patients with cerebellar ataxia, shows how the integration of the two learning processes could lead to benefit both learning performance and movement accuracy.

bioengineering↗