Publications SISSA
URI permanente per questa collezione
Sfogliare
Immissioni recenti
1 - 5 di 14490
-
PubblicazioneAligned and oblique dynamics in recurrent neural networks( 2024)The relation between neural activity and behaviorally relevant variables is at the heart of neuroscience research. When strong, this relation is termed a neural representation. There is increasing evidence, however, for partial dissociations between activity in an area and relevant external variables. While many explanations have been proposed, a theoretical framework for the relationship between external and internal variables is lacking. Here, we utilize recurrent neural networks (RNNs) to explore the question of when and how neural dynamics and the network’s output are related from a geometrical point of view. We find that training RNNs can lead to two dynamical regimes: dynamics can either be aligned with the directions that generate output variables, or oblique to them. We show that the choice of readout weight magnitude before training can serve as a control knob between the regimes, similar to recent findings in feedforward networks. These regimes are functionally distinct. Oblique networks are more heterogeneous and suppress noise in their output directions. They are furthermore more robust to perturbations along the output directions. Crucially, the oblique regime is specific to recurrent (but not feedforward) networks, arising from dynamical stability considerations. Finally, we show that tendencies toward the aligned or the oblique regime can be dissociated in neural recordings. Altogether, our results open a new perspective for interpreting neural activity by relating network dynamics and their output.
-
PubblicazioneFunctional Profiling of Olfactory Sensory Neurons: Electrophysiological Characterization of Human Olfactory Epithelium and Maturation-Dependent Changes in Mouse Neuronal Excitability(SISSA, 2026-03-05)The olfactory system is crucial for the detection of chemical cues from the environment, allowing species survival. Olfactory sensory neurons (OSNs) within the olfactory epithelium (OE) are responsible for odorant detection and signal transmission to the brain. To investigate olfaction, rodents, along with other species such as amphibians and fishes, have been extensively used as laboratory models. In the first part of this thesis, we provided the first electrophysiological characterization of human OSNs. Current knowledge about human OE is primarily confined to its morphology and molecular profile. However, little is known about the functional properties of human OSNs and supporting cells. We obtained acute slices of human OE from nasal biopsies and demonstrated their viability for whole-cell patch-clamp recordings. We measured voltage-gated currents from both OSNs and supporting cells in voltage-clamp configuration. Current-clamp protocols allowed us to assess the excitability of OSNs, which exhibited diverse firing patterns. Moreover, we demonstrated that these acute slices are also feasible for studying olfactory transduction, as we obtained the first electrophysiological responses of human OSNs upon odorant stimulation. Stimulation with a phosphodiesterase inhibitor elicited neuronal inward currents and action potentials, providing evidence that cyclic adenosine monophosphate (cAMP) is involved in the transduction pathway of human olfaction. In the second part of the thesis, we investigated immature OSNs from the mouse OE. The OE has the capability to continuously regenerate throughout life. To better characterize epithelial regeneration, a deeper knowledge of immature OSNs is required. While gene remodelling and morphological rearrangements are well established, changes in electrophysiological properties across maturation, remain largely unexplored. Using an electrophysiological approach, we explored the intrinsic properties of immature OSNs. Through loose-patch recordings, we demonstrated that immature OSNs are already endowed with a spontaneous activity. Currentclamp experiments showed that these neurons are excitable, although displaying lower excitability and slower action potential kinetics compared to mature OSNs. Both electrophysiological and transcriptomic analyses revealed differences in voltage-gated currents along development. Focusing on voltage-gated Na+ and K+ channels, we found the emergence of tetrodotoxin-resistant Na+ currents and transient A-type K+ currents when neurons become mature, likely influencing changes in firing behaviour. Altogether, these findings provide a comprehensive electrophysiological characterization of human OSNs, contributing to a deeper understanding of olfactory mechanisms in humans, and expand the current knowledge of OSN functional maturation through the functional description of immature OSNs in a mouse model.
-
PubblicazioneA brief review of reduced order models using intrusive and non‐intrusive techniques( 2024)Reduced Order Models (ROMs) have gained a great attention by the scientific community in the last years thanks to their capabilities of significantly reducing the computational cost of the numerical simulations, which is a crucial objective in applications like real time control and shape optimization. This contribution aims to provide a brief overview about such a topic. We discuss both a classic intrusive framework based on a Galerkin projection technique and hybrid/non-intrusive approaches, including Physics Informed Neural Networks (PINN), purely Data-Driven Neural Networks (NN), Radial Basis Functions (RBF), Dynamic Mode Decomposition (DMD) and Gaussian Process Regression (GPR). We also briefly mention geometrical parametrization and dimensionality reduction methods like Active Subspaces (ASs). Then we test the performance of such approaches in terms of efficiency and accuracy against three academic test cases, the lid driven cavity, the flow past a cylinder and the geometrically parametrized Stanford Bunny. Moreover, we also briefly present some preliminary results related to a more complex case involving an industrial application.
-
PubblicazioneUniqueness of asymptotic solutions for linear systems of ODEs with isolated singularities of general type( 2026)Abstract. This article is a review of our paper [18] which, for a wide class of ODEs, provides sufficient conditions of existence and uniqueness of a fundamental system of solutions, with specified asymptotic behaviour, in wide sectors centered at an isolated singularity of the coefficients. These singularity can be general, not just of pole type.
-
PubblicazioneDesign and Implementation of a Prediction-Serving System for Runtime and Parallel Performance in Quantum ESPRESSO(SISSA, 2025-12-16)Ab initio simulations, such as those performed with Quantum ESPRESSO (QE), play a central role in materials science but are often limited by their high computational cost. Predicting the execution time of self-consistent field (SCF) iterations is particularly challenging, as performance depends on both the physical characteristics of the simulated system and the parallelization parameters of the underlying hardware. This thesis investigates the use of machine learning (ML) techniques to predict the time required per SCF iteration directly from QE inputs, pseudopotentials, and computational settings. A complete workflow was designed to process raw benchmarking data into structured datasets, evaluate multiple regression approaches, and integrate the trained models into a web-based prediction-serving system. Among the tested models, Random Forests achieved the highest overall predictive accuracy and interpretability, revealing that the number of Kohn–Sham states, total cores, and electrons are the most influential factors affecting runtime. Fully Connected Neural Networks showed comparable performance and offered smooth, consistent predictions across a wide range of execution times. Simpler models, such as Kernel Ridge Regression and Linear Regression, provided useful baselines for comparison but were less effective in capturing nonlinear dependencies. Beyond model evaluation, a practical web interface was developed to make runtime prediction accessible to users in real time. By uploading QE input files and specifying hardware configurations, users can obtain immediate predictions of computational cost, supporting more informed resource allocation and efficient planning of large-scale simulations. Overall, this work demonstrates how data-driven approaches can complement traditional performance modeling in high-performance computing. By combining interpretability, predictive accuracy, and real-world deployment, the developed system represents a step toward intelligent, ML-assisted simulation workflows.