Experts in machine learning leverage domain knowledge to navigate decisions in model selection, hyperparameter optimization, and resource allocation. This is particularly critical for fine-tuning language models (LMs), where repeated trials incur substantial computational overhead and environmental impact. However, no existing automated framework simultaneously tackles the entire model selection and hyperparameter optimization (HPO) task for resource-efficient LM fine-tuning. We introduce XAutoLM, a meta-learning-augmented AutoML framework that reuses past experiences to optimize discriminative and generative LM fine-tuning pipelines efficiently. XAutoLM learns from stored successes and failures by extracting task- and system-level meta-features to bias its sampling toward valuable configurations and away from costly dead ends. On four text classification and two question-answering benchmarks, XAutoLM surpasses zero-shot optimizer's peak F1 on five of six tasks, cuts mean evaluation time of pipelines by up to 4.5x, reduces search error ratios by up to sevenfold, and uncovers up to 50% more pipelines above the zero-shot Pareto front. In contrast, simpler memory-based baselines suffer negative transfer. We release XAutoLM and our experience store to catalyze resource-efficient, Green AI fine-tuning in the NLP community.

Cell cultures exhibit rich and complex behaviors driven by dynamic metabolic interactions among cells. In this work, we present a model that captures these interactions through a framework inspired by statistical mechanics. Using Monte Carlo simulations, we explore the equilibrium and dynamical properties of a population of cells arranged in a two-dimensional lattice, where each cell is characterized by fluxes of three reactions: glucose consumption ($g$), respiration ($r$), and waste production/absorption ($w$). The system minimizes an energy function influenced by competitive (J_g > 0) and cooperative (J_w < 0) couplings between cells. Our results reveal three distinct phases: a competitive phase dominated by glucose competition, a cooperative phase marked by ordered waste exchange, and a disordered phase with local-scale cooperation. By incorporating evolutionary dynamics, we demonstrate how initially non-interacting cells can develop effective metabolic interactions, leading to heterogeneous cultures sustained by cross-feeding. These findings are further supported by analytical solutions derived using mean-field approximations. The model provides insights into how environmental constraints and stochastic fluctuations shape community structures, offering a versatile approach to study several emergent phenomena in biological systems.

We present a general framework to study quantum disordered systems in the context of the Kikuchi's Cluster Variational Method (CVM). The method relies in the solution of message passing-like equations for single instances or in the iterative solution of complex population dynamic algorithms for an average case scenario. We first show how a standard application of the Kikuchi's Cluster Variational Method can be easily translated to message passing equations for specific instances of the disordered system. We then present an "ad-hoc" extension of these equations to a population dynamic algorithm representing an average case scenario. At the Bethe level, these equations are equivalent to the dynamic population equations that can be derived from a proper Cavity Ansatz. However, at the plaquette approximation, the interpretation is more subtle and we discuss it taking also into account previous results in classical disordered models. Moreover, we develop a formalism to properly deal with the average case scenario using a Replica-Symmetric ansatz within this CVM for quantum disordered systems. Finally, we present and discuss numerical solutions of the different approximations for the Quantum Transverse Ising model and the Quantum Random Field Ising model in two dimensional lattices.

The mechanisms controlling the growth rate and composition of epitaxial CdTe and CdZnTe films were studied. The films were grown by isothermal closed space configuration technique. A GaAs 100 substrate was exposed sequentially to the elemental sources, Zn, Te, and Cd, in isothermal conditions. While growth of ZnTe followed an atomic layer epitaxy, ALE, regime self regulated at one monolayer per cycle; the CdTe films revealed different growth rates in dependence of the growth parameters,exposure and purge times. Combination of short purge times and larger Cd exposure times led to not self regulated growth regime for CdTe. This is ascribed to large Cd coverages that were dependent on Cd exposure times, following a Brunauer-Emmett and Teller-type adsorption. However, for longer purge times and or short Cd exposure times, an ALE self regulated regime was achieved with 2 ML per cycle. In this sense, the self-regulation of the growth is limited by desorption, instead of absorption, as in the traditional growth technique. Cd atoms substitution by Zn atoms and subsequent evaporation of surface Cd atoms during Zn exposure has been proved. The influence of these facts on the growth and composition of the alloy is discussed.

Nonequilibrium systems with strong parameter fluctuations are challenging to describe with standard Statistical Mechanics techniques. Superstatistics -- that can be seen as statistics of an underlying family of statistical distributions -- has emerged as a tool able to describe these complex systems. This has been successfully applied, for example, to hydrodynamic turbulence, internal convection, and even DNA architecture, but not to macroscopic biological systems. Here we document the occurrence of Superstatistics in the animal world, and reveal the emergence of Log-normal Superstatistics in a living system, when ants in a confined space are exposed to a threat. We use a data-driven superstatistical model to explain both normal and "panic" dynamics, identifying non-Gaussian velocity distributions, time scale separation, and the Log-normal statistics of a stochastic diffusion coefficient. We also reveal distinct behavioral regimes in the ants' collective panic response and how it relates with the individual ant memory and cluster formation. Furthermore, our findings indicate that optical signals or simple antennation are not significant mechanisms for panic transmission. These discoveries provide a foundation for the understanding of the biological origin of Log-normal type diffusion in confined environments.

Cellular automata (CA) have long attracted attention as dynamical systems with local updating rules and yet can exhibit, for certain rules, complex, long space and time correlated patterns. This contrast with other rules which results in trivial patterns being homogeneous or periodic. In this article we approach CA from two related angles: we analyze the information transfer in the time evolution of CA driven sequences and; we revisit the sensibility of the initial configuration on sequence evolution. In order to do so, we borrow a recently reported information distance based on Kolmogorov algorithmic complexity. The normalized information distance has been used previously to find a hierarchical clustering of CA rules. What is different in our approach, is the temporal analysis of the sequence evolutions by correlating different calculated distances with entropy density. Entropy rate, is a length invariant measure of the amount of irreducible randomness in a dynamical process. In order to perform our analysis, we incorporate to the practical calculation of the entropy rate and the distance measure, the use of Lempel-Ziv complexity. Lempel-Ziv complexity carries a number of practical advantages while avoiding the uncomputable nature of Kolmogorov randomness. The reduction of entropy density during time evolution can be related to energy dissipation through Landauer principle. Related to the last fact, is the computational capabilities of CA as information processing rules, were the performed analysis could be used to select CA rules amiable for simulating different physical process. The tools developed in this article for the analysis of the CA are easily extendible to the study of other one dimensional dynamical systems.

While there has been a keen interest in studying computation at the edge of chaos for dynamical systems undergoing a phase transition, this has come under question for cellular automata. We show that for continuously deformed cellular automata there is an enhancement of computation capabilities as the system moves towards cellular automata with chaotic spatiotemporal behavior. The computation capabilities are followed by looking into the Shannon entropy rate and the excess entropy, which allows identifying the balance between unpredictability and complexity. Enhanced computation power shows as an increase of excess entropy while the system entropy density has a sudden jump to values near one. The analysis is extended to a system of non-linear locally coupled oscillators that have been reported to exhibit spatiotemporal diagrams similar to cellular automata.

Understanding the relationship between the structure and function of the human brain is one of the most important open questions in Neurosciences. In particular, Resting State Networks (RSN) and more specifically the Default Mode Network (DMN) of the brain, which are defined from the analysis of functional data lack a definitive justification consistent with the anatomical structure of the brain. In this work, we show that a possible connection may naturally rest on the idea that information flows in the brain through a neural message-passing dynamics between macroscopic structures, like those defined by the human connectome (HC). In our model, each brain region in the HC is assumed to have a binary behavior (active or not), the strength of interactions among them is encoded in the anatomical connectivity matrix defined by the HC, and the dynamics of the system is defined by a neural message-passing algorithm, Belief Propagation (BP), working near the critical point of the human connectome. We show that in the absence of direct external stimuli the BP algorithm converges to a spatial map of activations that is similar to the DMN. Moreover, we computed, using Susceptibility Propagation (SP), the matrix of correlations between the different regions and show that the modules defined by a clustering of this matrix resemble several Resting States Networks determined experimentally. Both results suggest that the functional DMN and RSNs can be seen as simple consequences of the anatomical structure of the brain and a neural message-passing dynamics between macroscopic regions. We then show preliminary results indicating our predictions on how functional DMN maps change when the anatomical brain network suffers structural anomalies, like in Alzheimers Disease and in lesions of the Corpus Callosum.

Progress in immunotherapy revolutionized the treatment landscape for advanced lung cancer, raising survival expectations beyond those that were historically anticipated with this disease. In the present study, we describe the methods for the adjustment of mixture parametric models of two populations for survival analysis in the presence of long survivors. A methodology is proposed in several five steps: first, it is proposed to use the multimodality test to decide the number of subpopulations to be considered in the model, second to adjust simple parametric survival models and mixture distribution models, to estimate the parameters and to select the best model fitted the data, finally, to test the hypotheses to compare the effectiveness of immunotherapies in the context of randomized clinical trials. The methodology is illustrated with data from a clinical trial that evaluates the effectiveness of the therapeutic vaccine CIMAvaxEGF vs the best supportive care for the treatment of advanced lung cancer. The mixture survival model allows estimating the presence of a subpopulation of long survivors that is 44% for vaccinated patients. The differences between the treated and control group were significant in both subpopulations (population of short-term survival: p = 0.001, the population of long-term survival: p = 0.0002). For cancer therapies, where a proportion of patients achieves long-term control of the disease, the heterogeneity of the population must be taken into account. Mixture parametric models may be more suitable to detect the effectiveness of immunotherapies compared to standard models.

Experimental observation of a new mechanism of sandpile formation is reported. As a steady stream of dry sand is poured onto a horizontal surface, a pile forms which has a thin river of sand on one side flowing from the apex of the pile to the edge of its base. The river rotates about the pile, depositing a new layer of sand with each revolution, thereby growing the pile. For small piles the river is steady and the pile formed is smooth. For larger piles, the river becomes intermittent and the surface of the pile becomes undulating. The frequency of revolution of the river is measured as the pile grows and the results are explained with a simple scaling argument. The essential features of the system that produce the phenomena are discussed.

The ancestral sequence reconstruction problem is the inference, back in time, of the properties of common sequence ancestors from measured properties of contemporary populations. Standard algorithms for this problem assume independent (factorized) evolution of the characters of the sequences, which is generally wrong (e.g. proteins and genome sequences). In this work, we have studied this problem for sequences described by global co-evolutionary models, which reproduce the global pattern of cooperative interactions between the elements that compose it. For this, we first modeled the temporal evolution of correlated real valued characters by a multivariate Ornstein-Uhlenbeck process on a finite tree. This represents sequences as Gaussian vectors evolving in a quadratic potential, who describe selection forces acting on the evolving entities. Under a Bayesian framework, we developed a reconstruction algorithm for these sequences and obtained an analytical expression to quantify the quality of our estimation. We extend this formalism to discrete valued sequences by applying our method to a Potts model. We showed that for both continuous and discrete configurations, there is a wide range of parameters where, to properly reconstruct the ancestral sequences, intra-species correlations must be taken into account. We also demonstrated that, for sequences with discrete elements, our reconstruction algorithm outperforms traditional schemes based on independent site approximations.

Herewith we discuss a network model of the ferroptosis avascular and vascular tumor growth based on our previous proposed framework. Chiefly, ferroptosis should be viewed as a first order phase transition characterized by a supercritical Andronov Hopf bifurcation, with the emergence of limit cycle. The increase of the population of the oxidized PUFA fragments, take as the control parameter, involves an inverse Feigenbaum, (a cascade of saddle foci Shilnikov's bifurcations) scenario, which results in the stabilization of the dynamics and in a decrease of complexity.

It has been recently reported that irregular objects sink irregularly when released in a granular medium: a subtle lack of symmetry in the density or shape of a macroscopic object may produce a large tilting and deviation from the vertical path when released from the free surface of a granular bed. This can be inconvenient -- even catastrophic -- in scenarios ranging from buildings to space rovers. Here, we take advantage of the high sensitivity of granular intruders to shape asymmetry: we introduce a granular intruder equipped with an inflatable bladder that protrudes from the intruder's surface as an autonomous response to an unwanted tilting. So, the intruder's symmetry is only slightly manipulated, resulting in the rectification of the undesired tilting. Our smart intruder is even able to rectify its settling path when perturbed by an external element, like a vertical wall. The general concept introduced here can be potentially expanded to real-life scenarios, such as ``smart foundations'' to mitigate the inclination of constructions on a partially fluidized soil.

During overflow metabolism, cells excrete glycolytic byproducts when growing under aerobic conditions in a seemingly wasteful fashion. While potentially advantageous for microbes with finite oxidative capacity, its role in higher organisms is harder to assess. Recent single-cell experiments suggest overflow metabolism arises due to imbalances in inter-cellular exchange networks. We quantitatively characterize this scenario by integrating spatial metabolic modeling with tools from statistical physics and experimental single-cell flux data. Our results provide a theoretical demonstration of how diffusion-limited exchanges shape the space of accessible multi-cellular metabolic states. Specifically, a phase transition from a balanced network of exchanges to an unbalanced, overflow regime occurs as mean glucose and oxygen uptake rates vary. Heterogeneous single-cell metabolic phenotypes occur near this transition. Time-resolved tumor-stroma co-culture data support the idea that overflow metabolism stems from failure of inter-cellular metabolic coordination. In summary, environmental control is an emergent multi-cellular property, rather than a cell-autonomous effect.

We use a coarse-grained model of superconducting vortices driven through a random pinning potential to study the nonlinear current-voltage (IV) characteristics of flux flow in type II superconductors with pinning. In experiments, the IV relation measures flux flow down a flux density gradient. The work presented here treats this key feature explicitly. As the vortex repulsion weakens, the vortex pile maintains a globally steeper slope, corresponding to a larger critical current, for the same pinning potential. In addition, the magnitude of the peak in the differential resistance falls as the resistance peak shifts to higher currents. The model also exhibits so-called "IV fingerprints", and crossover to Ohmic (linear) behavior at high currents. Thus, many of the varieties of plastic behavior observed experimentally for soft flux line systems in the ``peak regime'' are reproduced in numerical simulations of the zero temperature model. The nonlinear transport behaviors are related to the self-organized, large scale morphologies of the vortex river flow down the slope of the vortex pile. These morphologies include isolated filamentary channels, braided rivers, and flooded rivers. We propose that these self-organized morphologies of flux flow down a flux gradient govern the various plastic flow behaviors, including nonlinear IV characteristics, observed in type II superconductors.

Cadmium sulfide is a valuable material for solar cells, photovoltaic, and radiation detectors. It is thus important to evaluate the material damage mechanisms and damage threshold in response to irradiation. Here, we simulate the ultrafast XUV/X-ray irradiation of CdS with the combined model, XTANT-3. It accounts for nonequilibrium electronic and atomic dynamics, nonadiabatic coupling between the two systems, nonthermal melting and bond breaking due to electronic excitation. We find that the two phases of CdS, zinc blende and wurtzite, demonstrate very close damage threshold dose of ~0.4-0.5 eV/atom. The damage is mainly thermal, whereas with increase of the dose, nonthermal effects begin to dominate leading to nonthermal melting. The transient disordered state is a high-density liquid, which may be semiconducting or metallic depending on the dose. Later recrystallization may recover the material back to the crystalline phase, or at high doses create an amorphous phase with variable bandgap. The revealed effects may potentially allow for controllable tuning of the band gap via laser irradiation of CdS.

To understand the complex interplay of topography and surface chemistry in wetting, fundamental studies investigating both parameters are needed. Due to the sensitivity of wetting to miniscule changes in one of the parameters it is imperative to precisely control the experimental approach. A profound understanding of their influence on wetting facilitates a tailored design of surfaces with unique functionality. We present a multi-step study: The influence of surface chemistry is analyzed by determining the adsorption of volatile carbonous species (A) and by sputter deposition of metallic copper and copper oxides on flat copper substrates (B). A precise surface topography is created by laser processing. Isotropic topography is created by ps laser processing (C), and hierarchical anisotropic line patterns are produced by direct laser interference patterning (DLIP) with different pulse durations (D). Our results reveal that the long-term wetting response of polished copper surfaces stabilizes with time despite ongoing accumulation of hydrocarbons and is dominated by this adsorption layer over the oxide state of the substrate (Cu, CuO, Cu2O). The surfaces' wetting response can be precisely tuned by tailoring the topography via laser processing. The sub-pattern morphology of primary line-like patterns showed great impact on the static contact angle, wetting anisotropy, and water adhesion. An increased roughness inside the pattern valleys combined with a minor roughness on the peaks favors air-inclusions, isotropic hydrophobicity, and low water adhesion. Increasing the aspect ratio showed to enhance air-inclusions and hydrophobicity despite increased peak roughness while time dependent wetting transitions were observed.

This study pioneers the application of the market microstructure framework to an informal financial market. By scraping data from websites and social media about the Cuban informal currency market, we model the dynamics of bid/ask intentions using a Limit Order Book (LOB). This approach enables us to study key characteristics such as liquidity, stability and volume profiles. We continue exploiting the Avellaneda-Stoikov model to explore the impact of introducing a Market Maker (MM) into this informal setting, assessing its influence on the market structure and the bid/ask dynamics. We show that the Market Maker improves the quality of the market. Beyond their academic significance, we believe that our findings are relevant for policymakers seeking to intervene informal markets with limited resources.

CdS has broad applications in solar cells and radiation detectors. We study its response to irradiation with swift heavy ions and determine its damage thresholds. We apply a model combining the Monte Carlo code TREKIS-3 to simulate the kinetics of the electronic system and the molecular dynamics code LAMMPS to track the atomic reaction to the energy transfer. It is found that ion tracks in CdS differ between its zincblende and wurtzite phases in morphology and damage thresholds. The anisotropic wurtzite phase displays directional damage formation with transient hexagonal track shapes along the (001) plane. In contrast, zincblende CdS exhibits a more cylindrical damage distribution. High pressures near the ion path drive mass transport away from the melted region, forming cavities within the track core in the wurtzite phase. Although significant recrystallization is observed during post-irradiation relaxation, it does not fully restore the original phase: final tracks consist of the amorphous cores due to the low densities in this region and defect-containing crystalline halos. These findings suggest that ion irradiation could be used to create mixed-phase CdS-based materials.