74 papers
This paper investigates the one-dimensional stochastic porous medium equation with additive white noise, combining functional renormalization group predictions and extensive numerical simulations to characterize its growth exponents, anomalous scaling, and multiscaling properties, while identifying its stationary measure with a random walk model related to a Bessel process.
This paper analytically and numerically investigates a lattice Lorentz gas on a cylinder, demonstrating how crowding and quasi-confinement induce a dimensional crossover from two to one dimensions in equilibrium and characterizing the resulting stationary velocity and diffusion under an external pulling force to first order in obstacle density.
This paper investigates a discrete model of a heterogeneous elastic line with random springs, demonstrating that when the spring constant distribution follows a power law with exponent , the system exhibits anomalous scaling driven by sample-to-sample fluctuations and abrupt shape jumps, a finding that challenges previous theoretical predictions and is validated by numerical simulations.
This paper presents a theoretical model of multi-head softmax attention that explains the sequential emergence of head specialization during training, demonstrates the noise-reducing benefits of softmax-1 activation, and introduces Bayes-softmax attention to achieve optimal prediction performance.
This paper argues that adversarial examples arise from an exponential misalignment between the high-dimensional perceptual manifolds of neural networks and human concepts, suggesting that achieving robustness requires aligning these dimensions to match human perception.
This paper challenges the conventional practice of stopping diffusion model training at the minimum test loss by identifying a "biased generalization" phase where models continue to lower loss while overfitting to training data, a phenomenon driven by the sequential nature of feature learning that poses risks for privacy-critical applications.
This paper theoretically and numerically investigates light propagation in hyperuniform disordered photonic crystal slabs, revealing that intrinsic radiative loss (non-Hermiticity) fundamentally alters disorder scattering from a power-law dependence to a form dominated by a finite constant plus a modified power-law term, a finding validated through simulations with realistic parameters.
This paper proposes a strategy to design and identify disorder-induced topological Anderson insulators protected by latent symmetries—hidden in the original system but revealed through isospectral reduction—thereby extending topological classification to cases where symmetries are not immediately visible.
This paper investigates quantum kinetically constrained models with symmetry to demonstrate how the interplay of constraints and chiral symmetry induces Hilbert space fragmentation and a parametric increase in zero modes, leading to the emergence of robust collective bound states that challenge ergodicity.
This paper demonstrates that the low-temperature ferromagnet-to-paramagnet transition in the two-dimensional random-bond Ising model is controlled by a zero-temperature fixed point that can be understood via a renormalization group mapping to a noninteracting quantum problem exhibiting an infinite randomness fixed point, where the tunneling exponent equals the spin stiffness exponent.
This paper proposes an analog thermodynamic coprocessor based on open quantum systems that performs parallel vector-matrix multiplications with stochastic matrices in a time independent of input size by encoding inputs as reservoir occupancies and reading outputs as stationary energy flows, while establishing a direct mapping between these quantum dynamics and electrical crossbar structures.
This paper proposes an annealing-based algorithm that solves partial differential equations by transforming discretized linear systems into generalized eigenvalue problems, enabling efficient computation of eigenvectors to arbitrary precision without increasing the number of variables.
This paper proposes a general construction principle for non-Hermitian topological operators in any dimension that maintain real eigenvalues and robust zero-energy boundary modes without exhibiting the non-Hermitian skin effect, thereby extending the bulk-boundary correspondence to a broad class of non-Hermitian insulators and semimetals.
This paper introduces a physics-informed neural network framework that enables the unsupervised learning of effective Hamiltonian parameters directly from experimental transport data in solid-state quantum simulators, utilizing an autoencoder architecture with a physics-decoder to ensure physically meaningful representations, robust generalization, and noise resilience.