SearcharxivSearch

arXiv subjects

Marco Baiesi

Publications and source records attributed to Marco Baiesi.

At least 19 recordsLinked to original sources

Emergence of Nonequilibrium Latent Cycles in Unsupervised Generative Modeling

We show that nonequilibrium dynamics can play a constructive role in unsupervised machine learning by inducing the spontaneous emergence of latent-state cycles. We introduce a model in which visible and hidden variables interact through two independently parametrized transition matrices, defining a Markov chain whose steady state is intrinsically out of equilibrium. Likelihood maximization drives this system toward nonequilibrium steady states with finite entropy production, reduced self-transition probabilities, and persistent probability currents in the latent space. These cycles are not imposed by the architecture but arise from training, and models that develop them reproduce the empirical distribution of data classes more faithfully, with a clear correlation between agreement with the data and entropy production. Compared with equilibrium approaches such as restricted Boltzmann machines, our model breaks the detailed balance between the forward and backward conditional transitions and relies on a log-likelihood gradient that depends explicitly on the last two steps of the Markov chain. Hence, this exploration of the interface between nonequilibrium statistical physics and modern machine learning suggests that introducing irreversibility into latent-variable models can improve the fidelity of the generated data distribution.

cond-mat.stat-mech

Fluctuations of driven probes reveal nonequilibrium transitions in complex fluids

Complex fluids subjected to localized microscopic energy inputs, typical of active microrheology setups, exhibit poorly understood nonequilibrium behaviors because of the intricate self-organization of their mesoscopic constituents. In this work we show how to identify changes in the microstructural conformation of the fluid by monitoring the variance of the probe position, based on a general method grounded in the breakdown of the equipartition theorem. To illustrate our method, we perform large-scale Brownian dynamics simulations of an effective model of micellar solution, and we link the different scaling regimes in the variance of the probe's position to the transitions from diffusive to jump dynamics, where the fluid intermittently relaxes the accumulated stress. This suggests stored elastic stress may be the physical mechanism behind the nonlinear friction curves recently measured in micellar solutions, pointing at a mechanism for the observed multi-step rheology. Our approach overcomes the limitations of continuum macroscopic descriptions and introduces an empirical method, applicable in experiments, to detect nonequilibrium transitions in the structure of complex fluids.

cond-mat.stat-mech

xEEGNet: Towards Explainable AI in EEG Dementia Classification

This work presents xEEGNet, a novel, compact, and explainable neural network for EEG data analysis. It is fully interpretable and reduces overfitting through major parameter reduction. As an applicative use case, we focused on classifying common dementia conditions, Alzheimer's and frontotemporal dementia, versus controls. xEEGNet is broadly applicable to other neurological conditions involving spectral alterations. We initially used ShallowNet, a simple and popular model from the EEGNet-family. Its structure was analyzed and gradually modified to move from a "black box" to a more transparent model, without compromising performance. The learned kernels and weights were examined from a clinical standpoint to assess medical relevance. Model variants, including ShallowNet and the final xEEGNet, were evaluated using robust Nested-Leave-N-Subjects-Out cross-validation for unbiased performance estimates. Variability across data splits was explained using embedded EEG representations, grouped by class and set, with pairwise separability to quantify group distinction. Overfitting was assessed through training-validation loss correlation and training speed. xEEGNet uses only 168 parameters, 200 times fewer than ShallowNet, yet retains interpretability, resists overfitting, achieves comparable median performance (-1.5%), and reduces variability across splits. This variability is explained by embedded EEG representations: higher accuracy correlates with greater separation between test set controls and Alzheimer's cases, without significant influence from training data. xEEGNet's ability to filter specific EEG bands, learn band-specific topographies, and use relevant spectral features demonstrates its interpretability. While large deep learning models are often prioritized for performance, this study shows smaller architectures like xEEGNet can be equally effective in EEG pathology classification.

cs.LG

Variational autoencoders understand knot topology

Supervised machine learning (ML) methods are emerging as valid alternatives to standard mathematical methods for identifying knots in long, collapsed polymers. Here, we introduce a hybrid supervised/unsupervised ML approach for knot classification based on a variational autoencoder enhanced with a knot type classifier (VAEC). The neat organization of knots in its latent representation suggests that the VAEC, only based on an arbitrary labeling of three-dimensional configurations, has grasped complex topological concepts such as chirality, unknotting number, braid index, and the grouping in families such as achiral, torus, and twist knots. The understanding of topological concepts is confirmed by the ability of the VAEC to distinguish the chirality of knots $9_{42}$ and $10_{71}$ not used for its training and with a notoriously undetected chirality to standard tools. The well-organized latent space is also key for generating configurations with the decoder that reliably preserves the topology of the input ones. Our findings demonstrate the ability of a hybrid supervised-generative ML algorithm to capture different topological features of entangled filaments and to exploit this knowledge to faithfully reconstruct or produce new knotted configurations without simulations.

cond-mat.stat-mech

Effective estimation of entropy production with lacking data

Observing stochastic trajectories with rare transitions between states, practically undetectable on time scales accessible to experiments, makes it impossible to directly quantify the entropy production and thus infer whether and how far systems are from equilibrium. To solve this issue for Markovian jump dynamics, we show a lower bound that outperforms any other estimation of entropy production (including Bayesian approaches) in regimes lacking data due to the strong irreversibility of state transitions. Moreover, in the limit of complete irreversibility, our new effective version of the thermodynamic uncertainty relation sets a lower bound to entropy production that depends only on nondissipative aspects of the dynamics. Such an approach is also valuable when dealing with jump dynamics with a deterministic limit, such as irreversible chemical reactions.

cond-mat.stat-mech

Constrained hidden Markov models reveal further Hsp90 protein states

Time series of conformational dynamics in proteins are usually evaluated with hidden Markov models (HMMs). This approach works well if the number of states and their connectivity is known. However, for the multi-domain protein Hsp90, a standard HMM analysis with optimization of the BIC (Bayesian information criterion) cannot explain long-lived states well. Therefore, here we employ constrained hidden Markov models, which neglect transitions between states by including assumptions. Gradually tuning a model with justified and focused changes allows us to improve its effectiveness and the score of the BIC. This became possible by analyzing time traces with several thousand observable transitions and, therefore, superb statistics. In this scheme, we also monitor the residences in the states reconstructed by the model, aiming to find exponentially distributed dwell times. We show how introducing new states can achieve these statistics but also point out limitations, e.g., for substantial similarity of two states connected to a common neighbor. One of the states displays the lowest free energy and could be the idle open "waiting state", in which Hsp90 waits for the binding of nucleotides, cochaperones, or clients.

cond-mat.stat-mech

Variance sum rule: proofs and solvable models

We derive, in more general conditions, a recently introduced variance sum rule (VSR) [I. Di Terlizzi et al., 2024 Science 383 971] involving variances of displacement and force impulse for overdamped Langevin systems in a nonequilibrium steady state (NESS). This formula allows visualising the effect of nonequilibrium as a deviation of the sum of variances from normal diffusion $2Dt$, with $D$ the diffusion constant and $t$ the time. From the VSR, we also derive formulas for the entropy production rate $σ$ that, differently from previous results, involve second-order time derivatives of position correlation functions. This novel feature gives a criterion for discriminating strong nonequilibrium regimes without measuring forces. We then apply and discuss our results to three analytically solved models: a stochastic switching trap, a Brownian vortex, and a Brownian gyrator. Finally, we compare the advantages and limitations of known and novel formulas for $σ$ in an overdamped NESS.

cond-mat.stat-mech

Earthquake-like dynamics in ultrathin magnetic film

We study the motion of a domain wall on an ultrathin magnetic film using the magneto-optical Kerr effect (MOKE). At tiny magnetic fields, the wall creeps only via thermal activation over the pinning centers present in the sample. Our results show that this creep dynamics is highly intermittent and correlated. A localized instability triggers a cascade, akin to aftershocks following a large earthquake, where the pinned wall undergoes large reorganizations in a compact active region for a few seconds. Surprisingly, the size and shape of these reorganizations display the same scale-free statistics of the depinning avalanches in agreement with the quenched Kardar-Parisi-Zhang universality class.

cond-mat.mtrl-sci

Interpretable machine learning of amino acid patterns in proteins: a statistical ensemble approach

Explainable and interpretable unsupervised machine learning helps understand the underlying structure of data. We introduce an ensemble analysis of machine learning models to consolidate their interpretation. Its application shows that restricted Boltzmann machines compress consistently into a few bits the information stored in a sequence of five amino acids at the start or end of $α$-helices or $β$-sheets. The weights learned by the machines reveal unexpected properties of the amino acids and the secondary structure of proteins: (i) His and Thr have a negligible contribution to the amphiphilic pattern of $α$-helices; (ii) there is a class of $α$-helices particularly rich in Ala at their end; (iii) Pro occupies most often slots otherwise occupied by polar or charged amino acids, and its presence at the start of helices is relevant; (iv) Glu and especially Asp on one side, and Val, Leu, Iso, and Phe on the other, display the strongest tendency to mark amphiphilic patterns, i.e., extreme values of an "effective hydrophobicity", though they are not the most powerful (non) hydrophobic amino acids.

q-bio.BM

Machine learning understands knotted polymers

Simulated configurations of flexible knotted rings confined inside a spherical cavity are fed into long-short term memory neural networks (LSTM NNs) designed to distinguish knot types. The results show that they perform well in knot recognition even if tested against flexible, strongly confined and therefore highly geometrically entangled rings. In agreement with the expectation that knots are delocalized in dense polymers, a suitable coarse-graining procedure on configurations boosts the performance of the LSTMs when knot identification is applied to rings much longer than those used for training. Notably, when the NNs fail, usually the wrong prediction still belongs to the same topological family of the correct one. The fact that the LSTMs are able to grasp some basic properties of the ring's topology is corroborated by a test on knot types not used for training. We also show that the choice of the NN architecture is important: simpler convolutional NNs do not perform so well. Finally, all results depend on the features used for input: surprisingly, coordinates or bond directions of the configurations provide the best accuracy to the NNs, even if they are not invariant under rotations (while the knot type is invariant). Other rotational invariant features we tested are based on distances, angles, and dihedral angles.

cond-mat.soft

Dynamic and Facilitated Binding of Topoisomerase Accelerates Topological Relaxation

How type 2 Topoisomerase (TopoII) proteins relax and simplify the topology of DNA molecules is one of the most intriguing open questions in biophysics. Most of the existing models neglect the dynamics of TopoII which is characteristics for proteins searching their targets via facilitated diffusion. Here, we show that dynamic binding of TopoII speeds up the topological relaxation of knotted substrates by enhancing the search of the knotted arc. Intriguingly, this in turn implies that the timescale of topological relaxation is virtually independent of the substrate length. We then discover that considering binding biases due to facilitated diffusion on looped substrates steers the sampling of the topological space closer to the boundaries between different topoisomers yielding an optimally fast topological relaxation. We discuss our findings in the context of topological simplification in vitro and in vivo.

cond-mat.soft

The rise and fall of branching: a slowing down mechanism in relaxing wormlike micellar networks

A mean-field kinetic model suggests that the relaxation dynamics of wormlike micellar networks is a long and complex process due to the problem of reducing the number of free end-caps (or dangling ends) while also reaching an equilibrium level of branching after an earlier overgrowth. The model is validated against mesoscopic molecular dynamics simulations and is based on kinetic equations accounting for scission and synthesis processes of blobs of surfactants. A long relaxation time scale is reached both with thermal quenches and small perturbations of the system. The scaling of this relaxation time is exponential with the free energy of an end cap and with the branching free energy. We argue that the subtle end-recombination dynamics might yield effects that are difficult to detect in rheology experiments, with possible underestimates of the typical time scales of viscoelastic fluids.

cond-mat.soft

Generalized equipartition for nonlinear multiplicative Langevin dynamics: application to laser-cooled atoms

The virial theorem, and the equipartition theorem in the case of quadratic degrees of freedom, are handy constraints on the statistics of equilibrium systems. Their violation is instrumental in determining how far from equilibrium a driven system might be. We extend the virial theorem to nonequilibrium conditions for Langevin dynamics with nonlinear friction and multiplicative noise. In particular, we generalize the equipartition theorem for confined laser-cooled atoms in the semi-classical regime. The resulting relation between the lowest moments of the atom position and velocity allows to measure in experiments how dissipative the cooling mechanism is. Moreover, its violation can reveal the departure from a strictly harmonic confinement or from the semi-classical regime.

cond-mat.stat-mech

Modelling the deceleration of COVID-19 spreading

By characterising the time evolution of COVID-19 in term of its "velocity" (log of the new cases per day) and its rate of variation, or "acceleration", we show that in many countries there has been a deceleration even before lockdowns were issued. This feature, possibly due to the increase of social awareness, can be rationalised by a susceptible-hidden-infected-recovered (SHIR) model introduced by Barnes, in which a hidden (isolated from the virus) compartment $H$ is gradually populated by susceptible people, thus reducing the effectiveness of the virus spreading. By introducing a partial hiding mechanism, for instance due to the impossibility for a fraction of the population to enter the hidden state, we obtain a model that, although still sufficiently simple, faithfully reproduces the different deceleration trends observed in several major countries.

q-bio.PE

On a possible nonequilibrium imprint in the cosmic background at low frequencies

The cosmic background radiation has been observed to deviate from the Planck law expected from a blackbody at $\sim$ 2.7 K at frequencies below $\sim 3$ GHz. We discuss the abundance of the low-energy photons from the perspective of nonequilibrium statistical mechanics. We propose a mechanism of stochastic frequency-diffusion, the counterpart to stochastic acceleration for charged particles in a turbulent plasma, to modify the standard Kompaneets equation. The resulting violation of the Einstein relation allows to take advantage of low-frequency localization and finally leads to photon cooling. The new equation predicts a frequency distribution in agreement with the absolute temperature measurements of the cosmic background radiation down to about 20 MHz, for which we offer here an updated compilation. In that sense, the so called 'space roar' we observe today is interpreted as a nonequilibrium echo of the early universe, and of nonequilibrium conditions in the primordial plasma more specifically.

astro-ph.CO

Explicit solution of the generalised Langevin equation

Generating an initial condition for a Langevin equation with memory is a non trivial issue. We introduce a generalisation of the Laplace transform as a useful tool for solving this problem, in which a limit procedure may send the extension of memory effects to arbitrary times in the past. This method allows us to compute average position, work, their variances and the entropy production rate of a particle dragged in a complex fluid by an harmonic potential, which could represent the effect of moving optical tweezers. For initial conditions in equilibrium we generalise the results by van Zon and Cohen, finding the variance of the work for generic protocols of the trap. In addition, we study a particle dragged for a long time captured in an optical trap with constant velocity in a steady state. Our formulas open the door to thermodynamic uncertainty relations in systems with memory.

cond-mat.stat-mech

Aging of living polymer networks: a model with patchy particles

Microrheology experiments show that viscoelastic media composed by wormlike micellar networks display complex relaxations lasting seconds even at the scale of micrometers. By mapping a model of patchy colloids with suitable mesoscopic elementary motifs to a system of worm-like micelles, we are able to simulate its relaxation dynamics, upon a thermal quench, spanning many decades, from microseconds up to tens of seconds. After mapping the model to real units and to experimental scission energies, we show that the relaxation process develops through a sequence of non-local and energetically challenging arrangements. These adjustments remove undesired structures formed as a temporary energetic solution for stabilizing the thermodynamically unstable free caps of the network. We claim that the observed scale-free nature of this stagnant process may complicate the correct quantification of experimentally relevant time scales as the Weissenberg number.

cond-mat.soft

A thermodynamic uncertainty relation for a system with memory

We introduce an example of thermodynamic uncertainty relation (TUR) for systems modeled by a one-dimensional generalised Langevin dynamics with memory, determining the motion of a micro-bead driven in a complex fluid. Contrary to TURs typically discussed in the previous years, our observables and the entropy production rate are one-time variables. The bound to the signal-to-noise ratio of such state-dependent observables only in some cases can be mapped to the entropy production rate. For example, this is true in Markovian systems. Hence, the presence of memory in the system complicates the thermodynamic interpretation of the uncertainty relation.

cond-mat.stat-mech