Discovering causal relations and equations from data

Physics Reports, Journal Year: 2023, Volume and Issue: 1044, P. 1 - 68

Published: Nov. 7, 2023

Physics is a field of science that has traditionally used the scientific method to answer questions about why natural phenomena occur and make testable models explain phenomena. Discovering equations, laws, principles are invariant, robust, causal been fundamental in physical sciences throughout centuries. Discoveries emerge from observing world and, when possible, performing interventions on system under study. With advent big data data-driven methods, fields equation discovery have developed accelerated progress computer science, physics, statistics, philosophy, many applied fields. This paper reviews concepts, relevant works broad physics outlines most important challenges promising future lines research. We also provide taxonomy for discovery, point out connections, showcase comprehensive case studies Earth climate sciences, fluid dynamics mechanics, neurosciences. review demonstrates discovering laws relations by revolutionised with efficient exploitation observational simulations, modern machine learning algorithms combination domain knowledge. Exciting times ahead opportunities improve our understanding complex systems.

Language: Английский

---

Perspectives on adaptive dynamical systems

Chaos An Interdisciplinary Journal of Nonlinear Science, Journal Year: 2023, Volume and Issue: 33(7)

Published: July 1, 2023

Adaptivity is a dynamical feature that omnipresent in nature, socio-economics, and technology. For example, adaptive couplings appear various real-world systems, such as the power grid, social, neural networks, they form backbone of closed-loop control strategies machine learning algorithms. In this article, we provide an interdisciplinary perspective on systems. We reflect notion terminology adaptivity different disciplines discuss which role plays for fields. highlight common open challenges give perspectives future research directions, looking to inspire approaches.

Language: Английский

---

Degeneracy in epilepsy: multiple routes to hyperexcitable brain circuits and their repair

Communications Biology, Journal Year: 2023, Volume and Issue: 6(1)

Published: May 3, 2023

Abstract Due to its complex and multifaceted nature, developing effective treatments for epilepsy is still a major challenge. To deal with this complexity we introduce the concept of degeneracy field research: ability disparate elements cause an analogous function or malfunction. Here, review examples epilepsy-related at multiple levels brain organisation, ranging from cellular network systems level. Based on these insights, outline new multiscale population modelling approaches disentangle web interactions underlying design personalised multitarget therapies.

Language: Английский

---

A new approach for estimating effective connectivity from activity in neural networks

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 6, 2024

Abstract Inferring and understanding the underlying connectivity structure of a system solely from observed activity its constituent components is challenge in many areas science. In neuroscience, techniques for estimating are paramount when attempting to understand network neural systems their recorded patterns. To date, no universally accepted method exists inference effective connectivity, which describes how node mechanistically affects other nodes. Here, focussing on purely excitatory networks small intermediate size continuous dynamics, we provide systematic comparison different approaches connectivity. Starting with Hopf neuron model conjunction known ground truth structural reconstruct system’s matrix using variety algorithms. We show that, sparse non-linear delays, combining lagged-cross-correlation (LCC) approach recently published derivative-based covariance analysis provides most reliable estimation matrix. also that linear networks, LCC has comparable performance based transfer entropy, at drastically lower computational cost. highlight works best decreases larger less networks. Applying dynamics without time find it does not outperform methods. Employing model, then use estimated as basis forward simulation order recreate under certain conditions, method, LCC, results higher trace-to-trace correlations than methods noise-driven systems. Finally, apply empirical biological data. subset nervous nematode C. Elegans . computationally simple performs better another published, more expensive reservoir computing-based method. Our comparatively can be used reliably estimate directed presence spatio-temporal delays noise. concrete suggestions scenario common research, where only neuronal set neurons known.

Language: Английский

---

Intrinsic neural diversity quenches the dynamic volatility of neural networks

Proceedings of the National Academy of Sciences, Journal Year: 2023, Volume and Issue: 120(28)

Published: July 3, 2023

Heterogeneity is the norm in biology. The brain no different: Neuronal cell types are myriad, reflected through their cellular morphology, type, excitability, connectivity motifs, and ion channel distributions. While this biophysical diversity enriches neural systems' dynamical repertoire, it remains challenging to reconcile with robustness persistence of function over time (resilience). To better understand relationship between excitability heterogeneity (variability within a population neurons) resilience, we analyzed both analytically numerically nonlinear sparse network balanced excitatory inhibitory connections evolving long scales. Homogeneous networks demonstrated increases strong firing rate correlations-signs instability-in response slowly varying modulatory fluctuation. Excitability tuned stability context-dependent way by restraining responses challenges limiting correlations, while enriching dynamics during states low drive. was found implement homeostatic control mechanism enhancing resilience changes size, connection probability, strength variability synaptic weights, quenching volatility (i.e., its susceptibility critical transitions) dynamics. Together, these results highlight fundamental role played cell-to-cell face change.

Language: Английский

---

Energy-efficient network activity from disparate circuit parameters

Proceedings of the National Academy of Sciences, Journal Year: 2022, Volume and Issue: 119(44)

Published: Oct. 24, 2022

Neural circuits can produce similar activity patterns from vastly different combinations of channel and synaptic conductances. These conductances are tuned for specific but might also reflect additional constraints, such as metabolic cost or robustness to perturbations. How do constraints influence the range permissible conductances? Here we investigate how affects parameters neural with in a model pyloric network crab Cancer borealis . We present machine learning method that identify models generate matching experimental data find consume largely amounts energy despite circuit activity. Furthermore, reduced still significant gives rise energy-efficient circuits. then examine space potential tuning strategies low cost. Finally, interaction between temperature robustness. show vary across temperatures changes does not necessarily incur an increased Our analyses efficiency constraining parameters, systems functional, efficient, robust widely disparate sets

Language: Английский

---

Pareto optimality, economy–effectiveness trade-offs and ion channel degeneracy: improving population modelling for single neurons

Open Biology, Journal Year: 2022, Volume and Issue: 12(7)

Published: July 1, 2022

Neurons encounter unavoidable evolutionary trade-offs between multiple tasks. They must consume as little energy possible while effectively fulfilling their functions. Cells displaying the best performance for such multi-task are said to be Pareto optimal, with ion channel configurations underpinning functionality. Ion degeneracy, however, implies that can lead functionally similar behaviour. Therefore, instead of a single model, neuroscientists often use populations models distinct combinations ionic conductances. This approach is called population (database or ensemble) modelling. It remains unclear, which parameters in vast functional more likely found brain. Here we argue optimality serve guiding principle addressing this issue by helping identify subpopulations conductance-based perform trade-off economy and In way, high-dimensional parameter space neuronal might reduced geometrically simple low-dimensional manifolds, potentially explaining experimentally observed correlations. Conversely, inference also help deduce functions from Patch-seq data. summary, promising framework improving modelling neurons circuits.

Language: Английский

---

Simulation-based inference with neural posterior estimation applied to X-ray spectral fitting

Astronomy and Astrophysics, Journal Year: 2024, Volume and Issue: 686, P. A133 - A133

Published: June 1, 2024

Context. Neural networks are being extensively used for modeling data, especially in the case where no likelihood can be formulated. Aims. Although of X-ray spectral fitting is known, we aim to investigate ability neural recover model parameters and their associated uncertainties compare performances with standard fitting, whether following a frequentist or Bayesian approach. Methods. We applied simulation-based inference posterior estimation (SBI-NPE) spectra. trained network simulated spectra generated from multiparameter source emission folded through an instrument response, so that it learns mapping between returns distribution. The sampled predefined prior To maximize efficiency training network, while limiting size sample speed up inference, introduce way reduce range priors, either classifier coarse quick one multiple observations. For sake demonstrating working principles, technique data recorded by NICER instrument, which medium-resolution spectrometer covering 0.2–12 keV band. consider here simple models five parameters. Results. SBI-NPE demonstrated work equally well as both Gaussian Poisson regimes, on real yielding fully consistent results terms best-fit distributions. time comparable smaller than needed when involving computation large Markov chain Monte Carlo chains derive On other hand, once properly trained, amortized generates distributions (less 1 second per spectrum 6-core laptop). show less sensitive local minima trapping fit statistic minimization techniques. With model, find dimension-reduced via principal component decomposition, leading faster significant degradation posteriors. Conclusions. complementary tool analysis. robust produces well-calibrated It holds great potential its integration pipelines developed processing sets. code demonstrate first principles introduced released Python package called SIXSA (Simulation-based Inference Spectral Analysis), available GitHub.

Language: Английский

---

sbi reloaded: a toolkit for simulation-based inference workflows

The Journal of Open Source Software, Journal Year: 2025, Volume and Issue: 10(108), P. 7754 - 7754

Published: April 8, 2025

Language: Английский

---

Uncertainty mapping and probabilistic tractography using Simulation-based Inference in diffusion MRI: A comparison with classical Bayes

Medical Image Analysis, Journal Year: 2025, Volume and Issue: 103, P. 103580 - 103580

Published: April 20, 2025

Language: Английский

---