Explore the vast range of titles available.

Big data can easily be contaminated by outliers or contain variables with heavy-tailed distributions, which makes many conventional methods inadequate. To address this challenge, we propose the adaptive Huber regression for robust estimation and inference. The key observation is that the robustification parameter should adapt to the sample size, dimension and moments for optimal tradeoff between bias and robustness. Our theoretical framework deals with heavy-tailed distributions with bounded th moment for any . We establish a sharp phase transition for robust estimation of regression parameters in both low and high dimensions: when , the estimator admits a sub-Gaussian-type deviation bound without sub-Gaussian assumptions on the data, while only a slower rate is available in the regime and the transition is smooth and optimal. In addition, we extend the methodology to allow both heavy-tailed predictors and observation noise. Simulation studies lend further support to the theory. In a genetic study of cancer cell lines that exhibit heavy-tailedness, the proposed methods are shown to be more robust and predictive. Supplementary materials for this article are available online.

Pt-based alloy catalysts may promise considerable mass activity (MA) for oxygen reduction but are generally unsustainable over long-term cycles, particularly in practical proton exchange membrane fuel cells (PEMFCs). Herein, we report a series of Pt-based intermetallic compounds (Pt₂Co, PtCo, and Pt₂Ti) enclosed by ultrathin Pt skin with an average particle size down to about 2.3 nm, which deliver outstanding cyclic MA and durability for oxygen reduction. By breaking size limitation during ordered atomic transformation in Pt alloy systems, the MA and durability of subsize Pt-based intermetallic compounds can be simultaneously optimized. The subsize scale was also found to enhance the stability of the membrane electrode through preventing the poisoning of catalysts by ionomers in humid fuel-cell conditions. We anticipate that subsize Pt-based intermetallic compounds set a good example for the rational design of high-performance oxygen reduction electrocatalysts for PEMFCs. Furthermore, the prevention of ionomer poisoning was identified as the critical parameter for assembling robust commercial membrane electrodes in PEMFCs.

Geometric properties of wave functions can explain the appearance of topological invariants in many condensed-matter and quantum systems1. For example, topological invariants describe the plateaux observed in the quantized Hall effect and the pumped charge in its dynamic analogue—the Thouless pump2–4. However, the presence of interparticle interactions can affect the topology of a material, invalidating the idealized formulation in terms of Bloch waves. Despite pioneering experiments in different platforms5–9, the study of topological matter under variations in interparticle interactions has proven challenging10. Here we experimentally realize a topological Thouless pump with fully tuneable Hubbard interactions in an optical lattice and observe regimes with robust pumping, as well as an interaction-induced breakdown. We confirm the pump’s robustness against interactions that are smaller than the protecting gap for both repulsive and attractive interactions. Furthermore, we identify that bound pairs of fermions are responsible for quantized transport at strongly attractive interactions. However, for strong repulsive interactions, topological pumping breaks down, but we show how to reinstate it by modifying the pump trajectory. Our results will prove useful for further investigations of interacting topological matter10, including edge effects11 and interaction-induced topological phases12–15.Thouless pumping is the quantization of charge transport through the adiabatic variation of a system’s parameters. The robustness and breakdown of pumping under variations in interparticle interactions have now been shown with ultracold atoms in an optical lattice.

Chinese pre-training is an essential direction in Chinese natural language processing, and vocabulary constitutes the foundation of pre-trained models. Existing methods for constructing vocabularies for Chinese pre-trained models typically treat each Chinese character as an indivisible token, overlooking the additional information embedded in the intrinsic structure of Chinese characters and its impact on model performance. To leverage this information, we propose a method of training with rule-based fine-grained vocabularies to directly learn the sequence of intrinsic structures of Chinese characters in Chinese pre-training. Specifically, we first construct a mapping rule-based fine-grained vocabulary based on the glyph and radical splitting mapping relationship of Chinese characters. Subsequently, we employ a whole char masking strategy to train Chinese pre-trained models based on this new vocabulary. Experimental results demonstrate that compared to the BERT baseline model, our model achieves promising performance on multiple downstream tasks, with fewer model parameters and stronger robustness. The proposed Chinese pre-training method exploits the intrinsic structural information of Chinese characters, providing a novel method for subsequent research in Chinese natural language processing.

The sampling problem lies at the heart of atomistic simulations and over the years many different enhanced sampling methods have been suggested toward its solution. These methods are often grouped into two broad families. On the one hand, are methods such as umbrella sampling and metadynamics that build a bias potential based on few order parameters or collective variables. On the other hand, are tempering methods such as replica exchange that combine different thermodynamic ensembles in one single expanded ensemble. We instead adopt a unifying perspective, focusing on the target probability distribution sampled by the different methods. This allows us to introduce a new class of collective-variables-based bias potentials that can be used to sample any of the expanded ensembles normally sampled via replica exchange. We also provide a practical implementation by properly adapting the iterative scheme of the recently developed on-the-fly probability enhanced sampling method [M. Invernizzi and M. Parrinello, J. Phys. Chem. Lett. 11, 2731 (2020)], which was originally introduced for metadynamicslike sampling. The resulting method is very general and can be used to achieve different types of enhanced sampling. It is also reliable and simple to use, since it presents only few and robust external parameters and has a straightforward reweighting scheme. Furthermore, it can be used with any number of parallel replicas. We show the versatility of our approach with applications to multicanonical and multithermal-multibaric simulations, thermodynamic integration, umbrella sampling, and combinations thereof.

Numerics converging on stripesThe Hubbard model (HM) describes the behavior of interacting particles on a lattice where the particles can hop from one lattice site to the next. Although it appears simple, solving the HM when the interactions are repulsive, the particles are fermions, and the temperature is low—all of which applies in the case of correlated electron systems—is computationally challenging. Two groups have tackled this important problem. Huang et al. studied a three-band version of the HM at finite temperature, whereas Zheng et al. used five complementary numerical methods that kept each other in check to discern the ground state of the HM. Both groups found evidence for stripes, or one-dimensional charge and/or spin density modulations.Science, this issue p. 1161, p. 1155Upon doping, Mott insulators often exhibit symmetry breaking where charge carriers and their spins organize into patterns known as stripes. For high–transition temperature cuprate superconductors, stripes are widely suspected to exist in a fluctuating form. We used numerically exact determinant quantum Monte Carlo calculations to demonstrate dynamical stripe correlations in the three-band Hubbard model, which represents the local electronic structure of the copper-oxygen plane. Our results, which are robust to varying parameters, cluster size, and boundary conditions, support the interpretation of experimental observations such as the hourglass magnetic dispersion and the Yamada plot of incommensurability versus doping in terms of the physics of fluctuating stripes. These findings provide a different perspective on the intertwined orders emerging from the cuprates’ normal state.

In this paper, we investigate the (2+1)-dimensional nonlinear Schrödinger equation (NLSE) which characterizes the transmission of optical pulses through optical fibers exhibiting refractive index variations corresponding to light intensity changes. Traditional numerical methods typically require a substantial amount of data to ensure the accuracy when solving high-dimensional NLSE, resulting in high experimental costs as well as a significant demand for storage space and computing power. With physical knowledge embedded into deep neural networks, physics-informed neural network (PINN) has been widely applied to solve various complex nonlinear problems and achieved significant results with small amount of data. Setting different groups of initial conditions and boundary conditions with hyperbolic and exponential functions, we construct the corresponding loss functions which will be further applied to train PINN. All data studied here is generated on Python. Based on the predicted results, we depict different types of optical pulses. According to our data experiments, lower prediction errors can be achieved with small volume of data, which fully demonstrates the effectiveness of the PINN. In the meantime, we also perform data-driven parameter discovery to the (2+1)-dimensional NLSE to study the coefficients of the group velocity dispersion and self-phase modulation terms. It can be seen that the PINN has high accuracy and robustness for parameter discovery to the (2+1)-dimensional NLSE. In brief, the use of PINN greatly enriches the diversity of solving methods, providing a reference for research of (2+1)-dimensional solitons in the field of optical fiber communications.

Artificial neural networks (ANNs) are at the core of most Deep Learning (DL) algorithms that successfully tackle complex problems like image recognition, autonomous driving, and natural language processing. However, unlike biological brains who tackle similar problems in a very efficient manner, DL algorithms require a large number of trainable parameters, making them energy-intensive and prone to overfitting. Here, we show that a new ANN architecture that incorporates the structured connectivity and restricted sampling properties of biological dendrites counteracts these limitations. We find that dendritic ANNs are more robust to overfitting and match or outperform traditional ANNs on several image classification tasks while using significantly fewer trainable parameters. These advantages are likely the result of a different learning strategy, whereby most of the nodes in dendritic ANNs respond to multiple classes, unlike classical ANNs that strive for class-specificity. Our findings suggest that the incorporation of dendritic properties can make learning in ANNs more precise, resilient, and parameter-efficient and shed new light on how biological features can impact the learning strategies of ANNs. Artificial neural networks, central to deep learning, are powerful but energy-consuming and prone to overfitting. The authors propose a network design inspired by biological dendrites, which offers better robustness and efficiency, using fewer trainable parameters, thus enhancing precision and resilience in artificial neural networks.

The stable functionality of networked systems is a hallmark of their natural ability to coordinate between their multiple interacting components. Yet, real-world networks often appear random and highly irregular, raising the question of what are the naturally emerging organizing principles of complex system stability. The answer is encoded within the system’s stability matrix—the Jacobian—but is hard to retrieve, due to the scale and diversity of the relevant systems, their broad parameter space and their nonlinear interaction dynamics. Here we introduce the dynamic Jacobian ensemble, which allows us to systematically investigate the fixed-point dynamics of a range of relevant network-based models. Within this ensemble, we find that complex systems exhibit discrete stability classes. These range from asymptotically unstable (where stability is unattainable) to sensitive (where stability abides within a bounded range of system parameters). Alongside these two classes, we uncover a third asymptotically stable class in which a sufficiently large and heterogeneous network acquires a guaranteed stability, independent of its microscopic parameters and robust against external perturbation. Hence, in this ensemble, two of the most ubiquitous characteristics of real-world networks—scale and heterogeneity—emerge as natural organizing principles to ensure fixed-point stability in the face of changing environmental conditions.Despite looking highly irregular, most real-world networks exhibit natural stability to external perturbations. A study of the properties of the stability matrix of networks now sheds light on the principles underlying this emerging stability.

Predicting individual differences in sustained attention is a key challenge in cognitive neuroscience. This study developed and validated an individualized prediction model using task-state functional magnetic resonance imaging (fMRI) from the psychomotor vigilance task (PVT). Task-state fMRI from 29 participants, each performing three PVT sessions, was analysed using leave-one-out cross-validation (LOOCV). Results demonstrated that the connectome-based predictive modeling (CPM) framework significantly predicted individual mean reaction times. Contributing connectivity features revealed distinct neural biomarkers: contributing positive features involved primarily frontal and default-mode network regions, whereas contributing negative features were characterized by temporal-subcortical connections. A comprehensive validation analysis with various parameters demonstrated the robustness of introduced framework in sustained attention prediction. These findings underscore the efficacy of task-state fMRI and CPM for individualized sustained attention assessment and deepen understanding of the neural mechanisms underlying attention.