arXiv subjects

Marcelo Guzman

Publications and source records attributed to Marcelo Guzman.

10 recordsLinked to original sources

arXiv2025

Remembrance of Tasks Past in Tunable Physical Networks

Sequential learning in physical networks is hindered by catastrophic forgetting, where training a new task erases solutions to earlier ones. We show that we can significantly enhance memory of previous tasks by introducing a hard threshold in the learning rule, allowing only edges with sufficiently large training signals to be altered. Thresholding confines tuning to the spatial vicinity of inputs and outputs for each task, effectively partitioning the network into weakly overlapping functional regions. Using simulations of tunable resistor networks, we demonstrate that this strategy enables robust memory of multiple sequential tasks while reducing the number of edges and the overall tuning cost. Our results hint at constrained training as a simple, local, and scalable mechanism to overcome catastrophic forgetting in tunable matter.

cond-mat.dis-nn↗

arXiv2025

Unsupervised and probabilistic learning with Contrastive Local Learning Networks: The Restricted Kirchhoff Machine

Autonomous physical learning systems modify their internal parameters and solve computational tasks without relying on external computation. Compared to traditional computers, they enjoy distributed and energy-efficient learning due to their physical dynamics. In this paper, we introduce a self-learning resistor network, the Restricted Kirchhoff Machine, capable of solving unsupervised learning tasks akin to the Restricted Boltzmann Machine algorithm. The circuit relies on existing technology based on Contrastive Local Learning Networks, in which two identical networks compare different physical states to implement a contrastive local learning rule. We simulate the training of the machine on the binarized MNIST dataset, providing a proof of concept of its learning capabilities. Finally, we compare the scaling behavior of the time, power, and energy consumed per operation as more nodes are included in the machine to their Restricted Boltzmann Machine counterpart operated on CPU and GPU platforms.

cond-mat.dis-nn↗

arXiv2025

Understanding and Embracing Imperfection in Physical Learning Networks

Performing machine learning with analog signals offers advantages in speed and energy efficiency, but sensitivity to component and measurement imperfections often foils training without a system-specific companion digital model. Here we take a different perspective, accepting and characterizing these inherent imperfections and ultimately overcoming them without digital models. We train an analog network of self-adjusting resistors -- a contrastive local learning network -- for multiple tasks, and observe limit cycles and scaling behaviors that limit precision, erase memory of previous tasks, and are absent in `perfect' systems. We develop an analytical model capturing these phenomena as a consequence of an uncontrolled learning bias continuously modifying the underlying representation of learned tasks, reminiscent of representational drift in the brain. Finally, we introduce and demonstrate a system-agnostic training method that greatly suppresses these effects. Our work points to a new, scalable analog approach that eschews precise modeling and instead thrives in the mess of real systems.

cond-mat.dis-nn↗

arXiv2024

Physical networks become what they learn

Physical networks can develop diverse responses, or functions, by design, evolution or learning. We focus on electrical networks of nodes connected by resistive edges. Such networks can learn by adapting edge conductances to lower a cost function that penalizes deviations from a desired response. The network must also satisfy Kirchhoff's law, balancing currents at nodes, or, equivalently, minimizing total power dissipation by adjusting node voltages. The adaptation is thus a double optimization process, in which a cost function is minimized with respect to conductances, while dissipated power is minimized with respect to node voltages. Here we study how this physical adaptation couples the cost landscape, the landscape of the cost function in the high-dimensional space of edge conductances, to the physical landscape, the dissipated power in the high-dimensional space of node voltages. We show how adaptation links the physical and cost Hessian matrices, suggesting that the physical response of networks to perturbations holds significant information about the functions to which they are adapted.

cond-mat.dis-nn↗

arXiv2023

Model-free characterization of topological edge and corner states in mechanical networks

Topological materials can host edge and corner states that are protected from disorder and material imperfections. In particular, the topological edge states of mechanical structures present unmatched opportunities for achieving robust responses in wave guiding, sensing, computation, and filtering. However, determining whether a mechanical structure is topologically nontrivial and features topologically-protected modes has hitherto relied on theoretical models. This strong requirement has limited the experimental and practical significance of topological mechanics to laboratory demonstrations. Here, we introduce and validate an experimental method to detect the topologically protected zero modes of mechanical structures without resorting to any modeling step. Our practical method is based on a simple electrostatic analogy: topological zero modes are akin to electric charges. To detect them, we identify elementary mechanical molecules and measure their chiral polarization, a recently introduced marker of topology in chiral phases. Topological zero modes are then identified as singularities of the polarization field. Our method readily applies to any mechanical structure and effectively detects the edge and corner states of regular and higher-order topological insulators. Our findings extend the reach of chiral topological phases beyond designer materials, and allow their direct experimental investigation.

cond-mat.mtrl-sci↗

arXiv2021

Non-orientable order and non-Abelian response in frustrated metamaterials

From atomic crystals to bird flocks, most forms of order are captured by the concept of spontaneous symmetry breaking. This paradigm was challenged by the discovery of topological order, in materials where the number of accessible states is not solely determined by the number of broken symmetries, but also by space topology. Until now however, the concept of topological order has been linked to quantum entanglement and has therefore remained out of reach in classical systems. Here, we show that classical systems whose global geometry frustrates the emergence of homogeneous order realise an unanticipated form of topological order defined by non-orientable order-parameter bundles: non-orientable order. We validate experimentally and theoretically this concept by designing frustrated mechanical metamaterials that spontaneously break a discrete symmetry under homogeneous load. While conventional order leads to a discrete ground-state degeneracy, we show that non-orientable order implies an extensive ground-state degeneracy -- in the form of topologically protected zero-nodes and zero-lines. Our metamaterials escape the traditional classification of order by symmetry breaking. Considering more general stress distributions, we leverage non-orientable order to engineer robust mechanical memory and achieve non-Abelian mechanical responses that carry an imprint of the braiding of local loads. We envision this principle to open the way to designer materials that can robustly process information across multiple areas of physics, from mechanics to photonics and magnetism.

cond-mat.soft↗

arXiv2019

Non-ideal rheology of semidilute bacterial suspensions

The rheology of semidilute bacterial suspensions is studied with the tools of kinetic theory, considering binary interactions, going beyond the ideal gas approximation. Two models for the interactions are considered, which encompass both the steric and short range interactions. In these, swimmers can either align polarly regardless of the state previous to the collision or they can align axially, being possible the end up antiparallel if the relative angle between directors is large. In both cases, it is found that an ordered phase develops when increasing the density, where the shear stress oscillates with large amplitudes, when a constant shear rate is imposed. This oscillation disappears for large shear rates in a continuous or discontinuous transition, depending if the aligning is polar or axial, respectively. For pusher swimmers these non-linear effects can produce an increase on the shear stress, contrary to the prediction of viscosity reduction made for the dilute regime with the ideal gas approximation.

cond-mat.soft↗

arXiv2018

Hyperuniform states generated by a critical friction field

Hyperuniform states are an efficient way to fill up space for disordered systems. In these states the particle distribution is disordered at the short scale but becomes increasingly uniform when looked at large scales. Hyperuniformity appears in several systems, in static or quasistatic regimes as well as close to transitions to absorbing states. Here, we show that a vibrated granular layer, at the critical point of the liquid-to-solid transition, displays dynamic hyperuniformity. Prior to the transition, patches of the solid phase form, with length scales and mean lifetimes that diverge critically at the transition point. When reducing the wavenumber, density fluctuations encounter increasingly more patches that block their propagation, resulting in a static structure factor that tends to zero for small wavenumbers at the critical point, which is a signature of hyperuniformity. A simple model demonstrates that this coupling of a density field to a highly fluctuating scalar friction field gives rise to dynamic hyperuniform states. Finally, we show that the structure factor detects better the emergence of hyperuniformity, compared to the particle number variance.

cond-mat.stat-mech↗

arXiv2018

Effective two-dimensional model for granular matter with phase separation

Granular systems confined in vertically vibrated shallow horizontal boxes (quasi two-dimensional geometry) present a liquid to solid phase transition when the frequency of the periodic forcing is increased. An effective model, where grains move and collide in two-dimensions is presented, which reproduces the aforementioned phase transition. The key element is that besides the two-dimensional degrees of freedom, each grain has an additional variable $\epsilon$ that accounts for the kinetic energy stored in the vertical motion in the real quasi two-dimensional motion. This energy grows monotonically during free flight, mimicking the energy gain by collisions with the vibrating walls and, at collisions, this energy is instantaneously transferred to the horizontal degrees of freedom. As a result, the average values of $\epsilon$ and the kinetic temperature are decreasing functions of the local density, giving rise to an effective pressure that can present van der Waals loops. A kinetic theory approach predicts the conditions that must satisfy the energy grow function to obtain the phase separation, which are verified with molecular dynamics simulations. Notably, the effective equation of state and the critical points computed considering the velocity--time-of-flight correlations differ only slightly from those obtained by simple kinetic theory calculations that neglect those correlations.

cond-mat.stat-mech↗

arXiv2018

Critical Phenomena in Quasi-Two-Dimensional Vibrated Granular Systems

The critical phenomena associated to the liquid to solid transition of quasi-two-dimensional vibrated granular systems is studied using molecular dynamics simulations of the inelastic hard sphere model. The critical properties are associated to the fourfold bond-orientational order parameter $\chi_4$, which measures the level of square crystallization of the system. Previous experimental results have shown that the transition of $\chi_4$, when varying the vibration amplitude, can be either discontinuous or continuous, for two different values of the height of the box. Exploring the amplitude-height phase space, a transition line is found, which can be either discontinuous or continuous, merging at a tricritical point and the continuous branch ends in an upper critical point. In the continuous transition branch, the critical properties are studied. The exponent associated to the amplitude of the order parameter is $\beta=1/2$, for various system sizes, in complete agreement with the experimental results. However, the fluctuations of $\chi_4$ do not show any critical behavior, probably due to crossover effects by the close presence of the tricritical point. Finally, in quasi-one-dimensional systems, the transition is only discontinuous, limited by one critical point, indicating that two is the lower dimension for having a tricritical point.

cond-mat.soft↗