Learning Matters! The physics of continual learning in biological and soft matter

Descripción del proyecto

Lo que una sola célula puede revelar sobre el aprendizaje

El aprendizaje suele asociarse al cerebro y a las neuronas. No obstante, la capacidad de responder a señales repetidas podría constituir una propiedad más fundamental de la materia. Comprender cómo surge este tipo de aprendizaje podría favorecer nuevas oportunidades en la ciencia de materiales y la medicina. Por ello, en el proyecto LearningMatters, financiado por el Consejo Europeo de Investigación, se estudiará «Physarum polycephalum», un organismo unicelular que forma redes ramificadas pese a carecer de cerebro. Las señales se propagan a través de su estructura tubular mediante flujos y cambios químicos. El organismo se adapta modificando las propiedades mecánicas de sus paredes. A partir de mecanismos similares, reproducidos en materiales blandos como los hidrogeles, el equipo del proyecto busca desarrollar una «materia capaz de aprender». En el futuro, esta línea de investigación podría dar lugar a nuevas tecnologías biomédicas, entre ellas implantes tisulares avanzados.

Objetivo

How many times have you wished you could simply pause your brain, only to find, to your frustration, that it is still busy, revisiting and processing inputs?
Response to continual inputs is normal and essential, it is how we learn. In fact, any matter continually responds to input. Yet, our understanding of learning struggles with continual input.
I here break with the current focus on the brain to uncover the physics of continual learning matter instead by investigating the emergence of learning bottom-up in life reduced in complexity to a network-shaped single cell – Physarum.
Lacking any neurons, flows flushing throughout Physarum’s tubular network propagate input packaged as chemical concentration and flow shear force. Tube wall visco-elasticity re-organises in response – continually – learning its future response. Such mechanical learning serves as the blueprint for implementing learning in soft matter, as Physarum’s responsiveness in wall visco-elasticity has its soft matter twin in synthetic hydrogels.
Implementing Physarum’s learning mechanics, cast in the language of theoretical physics, directly into soft matter, I uncover the physics of continual learning matter and, thus, develop biocompatible learning matter. My ambitious aim requires an interdisciplinary approach spanning biology, soft matter, and theoretical physics to
1) reveal how Physarum learns as its matter mechanically responds to input,
2) manufacture synthetic, soft matter flow networks that mimic Physarum’s learning mechanics,
3) derive a theory of continual learning from both biological and soft matter learning data to identify how learned behaviour is selected to last or vanish and
4) apply soft matter learning to initiate proof-of-concept vascularised tissue implants.
My conceptual insight may inspire artificial intelligence but unfolds its ground-breaking power by making learning available in biocompatible soft matter with massive potential for medical treatment with tissue implants.