BEGIN:VCALENDAR VERSION:2.0 PRODID:Icfo X-PUBLISHED-TTL:P1W BEGIN:VEVENT UID:69d4b3d5ef206 DTSTART:20251014T080000Z SEQUENCE:0 TRANSP:OPAQUE LOCATION:Elements Room and Online (Teams) SUMMARY:ICFO | GABRIEL FERNÁNDEZ CLASS:PUBLIC DESCRIPTION:Understanding how a complex system works from its components\, such as a virus invading a cell or particles aggregating in a liquid\, is a fundamental question in the study of nature that provides great biologic al benefits. To solve this question\, it is interesting to observe the pat h taken by the components of a system\, as this contains valuable informat ion that helps us to characterize them and understand how they interact wi th each other. Advances in the last decade in the field of machine learnin g offer a promising numerical tool\, as they allow the automatic extractio n of relevant features and relationships\, while also predicting the syste m's behavior.\nIn this thesis\, we focus on the analysis of particle traje ctories observed in complex systems\, addressing two fundamental aspects: the random and therefore difficult-to-characterize individual behavior\, a s occurs in the lungs\, where we inhale air and oxygen diffuses into the c apillaries of the alveoli\; and behavior due to multiple ways of interacti ng\, in some cases unknown\, such as that of a large flock of birds migrat ing together.\nIn particular\, we consider three problems:\n1) the accurat e estimation of parameters that characterize the anomalous diffusion obser ved in biological processes\,\n2) the identification of significant parame ters to describe stochastic processes\,\nand 3) the extraction of the func tional form of the multiple forces present in particle systems.\nTo tackle each of the problems\, we developed a specific machine learning model des igned to extract meaningful information from trajectories and rigorously e valuated it on a series of simulated systems with known dynamics.\nThe fir st method\, KISTEP\, predicts anomalous diffusion properties at each time step\, for trajectory segments\, and for a set of trajectories\, allowing for detailed analysis at each level\, based on individual trajectories. Wi th this method\, we participated in the AnDi Challenge 2\, a scientific co mpetition comparing computational methods dedicated to characterizing frac tional Brownian motion trajectories that resemble biological phenomena obs erved in experiments such as cell endocytosis or protein immobilization.\n The second method\, SPIVAE\, helps to identify the minimal representation of stochastic processes thanks to its unsupervised\, interpretable\, and g enerative features. Furthermore\, it is capable of generating new trajecto ries that reproduce the learned characteristics of the process. The analys is performed with SPIVAE revealed the expected parameters of BM\, fraction al BM\, and confined BM\, while it learned a nonlinear combination in the case of the scaled BM.\nThe third method\, FISGAE\, employs a graph neural network to infer in an unsupervised manner the functional form of the for ces acting between particles. FISGAE successfully learned the forces betwe en 21 interacting particles with non-reciprocal linear forces\, while in t he more complex scenario of a Lennard-Jones gas\, it learned well the forc e at short distances.\nIn conclusion\, this research provides methods to f acilitate the analysis of particle systems directly from their trajectorie s\, unlocking insights otherwise unavailable. The proposed methods have th e potential to benefit experimental and theoretical researchers\, and even artificial intelligence developers\, by enabling a more comprehensive und erstanding of complex systems. Furthermore\, the developed frameworks are ready for future improvements\, which could be achieved through the integr ation of more sophisticated architectures\, thereby paving the way for eve n more advanced applications and discoveries.\n \;\nTuesday October 14 \, 10:00 h. Elements room \nThesis Director: Prof. Dr. Maciej Lewenstein a nd Dr. Carlo Manzo\n \; DTSTAMP:20260407T073549Z END:VEVENT END:VCALENDAR