Recent studies have deciphered the nature of the neural code for facial identification in the primate inferotemporal cortex. The neural code is believed to consist in a linear projection of the facial image onto principal axes of variation, preceded by a disentanglement of the facial coordinates of shape (Cartesian positions of facial elements) and texture (image variation at fixed shape coordinates). We present an estimation of the efficiency, on information-theoretical grounds, of such a probabilistic representation of facial images (known as Active Appearance Model), comparing it with the simpler representation in terms of Eigenfaces. The seminar focus will be on the methods: we will review the relation between information theory and Bayesian statistics, the notion of description length, and the multivariate normal distribution.

Ponente: Miguel Ibáñez Berganza. Università di Roma, «La Sapienza».

Fecha y hora: miércoles, 29 de junio de 2022 a las 12:00.

Lugar: Seminario de Física Computacional, planta baja del edificio de Física (junto a las pantallas). Facultad de Ciencias.

Cell differentiation is one of the most fascinating areas of developmental biology. This was long thought to be an irreversible process, however it has been shown recently that it is possible to reprogramme fully differentiated cells into a state of induced pluripotency, via the introduction of a few transcription factors. This opens up exciting perspectives in the field of regenerative medicine, however, no universally accepted theory exists that explains the phenomena. The purpose of this work is to drive forward our understanding of the dynamics of gene-regulatory networks by introducing two analytical models, one based on neural networks, the other based on bipartite graphs. Using neural networks techniques, we model cell types as hierarchically organized dynamical attractors corresponding to cell cycles. Stages of the cell cycle are fully characterised by the configuration of gene expression levels, and reprogramming corresponds to triggering transitions between such configurations. Two mechanisms were found for reprogramming: cycle-state specific perturbations and a noise-induced switching. We then model the mechanism for the effective interactions arising between genes, by a bipartite graph model, that integrates the genome and transcriptome into a single regulatory network. With this perspective, we are able to deduce important features of the regulatory network that exists in every cell type. We use an efficient implementation of the dynamic cavity method, based on dynamic programming, to analyse the heterogeneous statistics of single node activation in fully asymmetric networks with fat-tailed degree distribution. In addition, we extend the dynamical cavity approach to calculate the pairwise correlations induced by different motifs in the network. Our results suggest that the entire statistics of observed correlations can be accurately described in terms of just two basic motifs. Finally, we investigate networks with bi-directional links. We observe that the stationary state associated with symmetric or anti-symmetric interactions is biased towards the active or inactive state respectively, even when interaction are independently drawn from an unbiased distribution. This phenomenon, which can be regarded as a form of spontaneous symmetry-breaking, is peculiar to systems formulated in terms of Boolean variables, as opposed to Ising spins.

Ponente: Alessia Annibale. King’s College London.

Fecha y hora: martes, 14 de junio de 2022 a las 11:00.

Lugar: Seminario de Física Computacional, planta baja del edificio de Física (junto a las pantallas). Facultad de Ciencias.

We propose an algebraic method to systematically approach the solution of an ordinary differential equation (ODE) with any boundary conditions. We define an extended ODE (eODE) composed of a linear generic differential operator that depends on free parameters plus an ε formal perturbation formed by the original ODE minus the same linear term. After an eODE’s formal ε expansion, we can solve order by order a hierarchy of linear ODEs. We get a sequence of functions that converge exponentially fast to the solution/s when ε = 1 and after determining the free set of parameters by minimising a distance-to-the solution function. Therefore, we get a formal expansion of the solution that we call Ghost Expansion that can be used as a multiscaling decomposition of the ODE’s solution. The method permits the detection of several solutions to Boundary Value Problems just by looking at the number of minima of the distance function. We present the method by its application to several cases where we discuss its properties, benefits and shortcomings and some practical algorithmic improvements on it.

Ponente: Pedro Garrido. Universidad de Granada.

Fecha y hora: miércoles, 1 de junio de 2022 a las 11:00.

Lugar: Seminario de Física Computacional, planta baja del edificio de Física (junto a las pantallas). Facultad de Ciencias.

The question of how coexistence of species is generated and maintained is as old as ecology itself. Many ecological “forces” such as competition, cooperation, demographic fluctuation, environmental fluctuation and filtering, migration, predation etc. are expected to act together in ecosystems. Disentangling the effect and intensity of each of theses forces in natural communities is the focus of present research. Thanks to a recent revolution in the availability of high quality data, natural microbial ecosystems offer an invaluable possibility to tackle this question. Here, we explore if it is possible to discriminate a dominant force at a give resolution of genetic similarity. By using both relative-abundances and metagenomic data, we reveal the presence of a new macroecological law relating correlation and phylogenetic similarity. In particular, the average correlation of species abundance fluctuation decays with phylogenetic distance from positive to null values following a stretch exponential function consistently in all empirically analyzed biomes both across communities (hosts) and in temporal data for each community. By scrutinizing different ecological models, we show that competition cannot reproduce the observed pattern. Instead, the elucidated macroecological law is explained quantitatively by the “correlated stochastic logistic model” (CSLM) pointing to environmental filtering as the dominant ecological force at this resolution level. We conclude by arguing that in order to understand interactions in microbial ecosystems one needs to abandon the concept of species and study the system from different scales, much as done in physics exploiting renormalization-group ideas.

Ponente: Matteo Sireci. Universidad de Granada.

Fecha y hora: viernes, 4 de febrero de 2022 a las 12:00.

Lugar: Seminario de Física Computacional, planta baja del edificio de Física (junto a las pantallas). Facultad de Ciencias. Online a través de Google Meet en el siguiente enlace: https://meet.google.com/vub-uoiw-piz

The dynamics of open quantum systems can be described by a Liouvillian, which in the Markovian approximation fulfills the Lindblad master equation. We present a family of integrable many-body Liouvillians based on Richardson-Gaudin models with a complex structure of the jump operators. Making use of this new region of integrability, we study the transition to chaos in terms of a two-parameter Liouvillian. The transition is characterized by the spectral statistics of the complex eigenvalues of the Liouvillian operators using the nearest neighbor spacing distribution and by the ratios between eigenvalue distances.

Ponente: Rafael A. Molina Fernández. Instituto de Estructura de la Materia, CSIC, Madrid.

Fecha y hora: miércoles, 24 de noviembre de 2021 a las 12:00.

Lugar: Presencial: hasta completar aforo en el seminario de Física Computacional, planta baja del edificio de Física (junto a las pantallas), Facultad de Ciencias. Online: a través de Google Meet en https://meet.google.com/kye-ynsm-vzd.

Hemispherectomy is a last-resource treatment for some neurological disorders. This radical intervention allows some patients to live normally, with better odds the earlier in life it happens. Somehow, the remaining hemisphere takes on the outstanding computational burden. Brain plasticity at smaller scales shows how functionality is adopted by adjacent tissue. In models of brain rewiring after stroke, circuits accepting new workload are close and similar to the damaged ones. Hemispherectomy demands more drastic changes, mixing far and functionally diverse regions. We lack mathematical models of this. We introduce a simple model of brain reorganization after hemispherectomy based on Self-Organized Maps (SOMs). We show how emerging representations in SOMs constrain brain reorganization after simulated hemispherectomy, resulting in some forbidden and some other favored rearrangement pathways, each with distinct symmetries and properties. We discuss what the enabled paths imply for the recovery of topographic maps and language functionality after hemispherectomy. We find how too much symmetry can be detrimental for the proper formation of representation systems. We also obtain results regarding the existence of window periods – a critical age after which hemispherectomy causes irreversible function loss. These findings illuminate various (hitherto unexplained) clinical facts about window periods for language recovery. Our model offers a powerful thinking tool and suggests simple guiding principles for large-scale brain plasticity – notably, that the geometry of emerging representations turns into topological constraints for large-scale brain rearrangement. This offers insights about why such an aggressive intervention results in highly functional brains nevertheless, and suggests specific treatments for simulated, pathological disorders observed in our SOM models.

Ponente: Luis Seoane. Centro Nacional de Biotecnología, CSIC.

Fecha y hora: jueves, 16 de septiembre de 2021 a las 12:00.

Lugar:

Presencial (hasta completar aforo): seminario de Física Computacional, en la planta baja del edificio de Física (junto a las pantallas) de la Facultad de Ciencias.

We compare various known and original strategies of overfitting mitigation in correlation matrices, in the context of brain functional connectivity. In particular, we infer a database of human brain activity from functional Magnetic Resonance Imaging (fMRI), beyond Maximum Likelihood inference and using the multivariate Gaussian as likelihood. We show that the relative algorithm performances are consistent across subjects, and across samples of a synthetic database of similar characteristics. We observe as well that the resulting cleaned correlation matrices, that are proposed as a refined model of functional connectivity, depend crucially on the cleaning algorithm. We discuss possible applications of these findings to network neuroscience.

Ponente: Dr. Miguel Ibáñez. ISTC-CNR (Italy).

Fecha y hora: lunes, 28 de junio de 2021 a las 12:00.

Lugar: Seminario de Física Computacional, planta baja del edificio de Física (hasta completar aforo) y vía Google Meet: https://meet.google.com/ijd-agcd-rye..

Los seminarios se encuentran financiados por el Ministerio de Ciencia, Innovación y Universidades.