Search for: All records

Abstract Forecasting all components in complex systems is an open and challenging task, possibly due to high dimensionality and undesirable predictors. We bridge this gap by proposing a data-driven and model-free framework, namely, feature-and-reconstructed manifold mapping (FRMM), which is a combination of feature embedding and delay embedding. For a high-dimensional dynamical system, FRMM finds its topologically equivalent manifolds with low dimensions from feature embedding and delay embedding and then sets the low-dimensional feature manifold as a generalized predictor to achieve predictions of all components. The substantial potential of FRMM is shown for both representative models and real-world data involving Indian monsoon, electroencephalogram (EEG) signals, foreign exchange market, and traffic speed in Los Angeles Country. FRMM overcomes the curse of dimensionality and finds a generalized predictor, and thus has potential for applications in many other real-world systems. 

The capacity to generalize to future unseen data stands as one of the utmost crucial attributes of deep neural networks. Sharpness-Aware Minimization (SAM) aims to enhance the generalizability by minimizing worst-case loss using one-step gradient ascent as an approximation. However, as training progresses, the non-linearity of the loss landscape increases, rendering one-step gradient ascent less effective. On the other hand, multi-step gradient ascent will incur higher training cost. In this paper, we introduce a normalized Hessian trace to accurately measure the curvature of loss landscape on both training and test sets. In particular, to counter excessive non-linearity of loss landscape, we propose Curvature Regularized SAM (CR-SAM), integrating the normalized Hessian trace as a SAM regularizer. Additionally, we present an efficient way to compute the trace via finite differences with parallelism. Our theoretical analysis based on PAC-Bayes bounds establishes the regularizer's efficacy in reducing generalization error. Empirical evaluation on CIFAR and ImageNet datasets shows that CR-SAM consistently enhances classification performance for ResNet and Vision Transformer (ViT) models across various datasets. Our code is available at https://github.com/TrustAIoT/CR-SAM. 

The transferability of adversarial examples is of central importance to transfer-based black-box adversarial attacks. Previous works for generating transferable adversarial examples focus on attacking given pretrained surrogate models while the connections between surrogate models and adversarial trasferability have been overlooked. In this paper, we propose Lipschitz Regularized Surrogate (LRS) for transfer-based black-box attacks, a novel approach that transforms surrogate models towards favorable adversarial transferability. Using such transformed surrogate models, any existing transfer-based black-box attack can run without any change, yet achieving much better performance. Specifically, we impose Lipschitz regularization on the loss landscape of surrogate models to enable a smoother and more controlled optimization process for generating more transferable adversarial examples. In addition, this paper also sheds light on the connection between the inner properties of surrogate models and adversarial transferability, where three factors are identified: smaller local Lipschitz constant, smoother loss landscape, and stronger adversarial robustness. We evaluate our proposed LRS approach by attacking state-of-the-art standard deep neural networks and defense models. The results demonstrate significant improvement on the attack success rates and transferability. Our code is available at https://github.com/TrustAIoT/LRS. 

North Atlantic Deep Water (NADW), the return flow component of the Atlantic Meridional Overturning Circulation (AMOC), is a major inter-hemispheric ocean water mass with strong climate effects but the evolution of its source components on million-year timescales is poorly known. Today, two major NADW components that flow southward over volcanic ridges to the east and west of Iceland are associated with distinct contourite drift systems that are forming off the coast of Greenland and on the eastern flank of the Reykjanes (mid-Atlantic) Ridge. Here we provide direct records of the early history of this drift sedimentation based on cores collected during International Ocean Discovery Programme (IODP) Expeditions 395C and 395. We find rapid acceleration of drift deposition linked to the eastern component of NADW, known as Iceland–Scotland Overflow Water at 3.6 million years ago (Ma). In contrast, the Denmark Strait Overflow Water feeding the western Eirik Drift has been persistent since the Late Miocene. These observations constrain the long-term evolution of the two NADW components, revealing their contrasting independent histories and allowing their links with climatic events such as Northern Hemisphere cooling at 3.6 Ma, to be assessed. 

Abstract In high-temperature ( T c ) cuprate superconductors, many exotic phenomena are rooted in the enigmatic pseudogap state, which has been interpreted as consisting of preformed Cooper pairs or competing orders or a combination thereof. Observation of pseudogap phenomenologically in electron-doped Sr 2 IrO 4 —the 5d electron counterpart of the cuprates, has spurred intense interest in the strontium iridates as a testbed for exploring the exotic physics of the cuprates. Here, we examine the pseudogap state of electron-doped Sr 2 IrO 4 by angle-resolved photoemission spectroscopy (ARPES) and parallel theoretical modeling. Our analysis demonstrates that the pseudogap state of Sr 2 IrO 4 appears without breaking the particle–hole symmetry or inducing spectral broadening which are telltale signatures of competing orders in the cuprates. We find quasiparticle dispersion and its temperature dependence in the pseudogap state of Sr 2 IrO 4 to point to an electronic order with a zero scattering wave vector and limited correlation length. Particle–hole symmetric preformed Cooper pairs are discussed as a viable mechanism for such an electronic order. The potential roles of incommensurate density waves are also discussed. 

Zeolites (ZSM-5 and Beta) with different SiO2/Al2O3 ratios were synthesized as solid acids for hydrolyzing cellulose in an inorganic ionic liquid system (lithium bromide trihydrate solution, LBTH) under mild conditions. The results indicated that the texture properties of zeolite had little effect on catalytic activity, while acidity of zeolite was crucial to the cellulose hydrolysis. In the LBTH system, H-form zeolites released H+ into the solution from their acid sites via ion-exchange with Li+, which hydrolyzed the cellulose already dissolved. This unique homogeneous hydrolysis mechanism was the primary reason for the excellent performance of the zeolites in catalyzing cellulose hydrolysis in the LBTH system. It was found cellulose could be completely hydrolyzed to glucose and oligoglucan by 2% (w/w on cellulose) zeolite at 140 °C within 3 h with a single-pass glucose yield 61%. The zeolites could be recovered with 50% initial catalytic activity after regeneration and reused with stable catalytic activity.