Search for: All records

Abstract We propose a sparse deep ReLU network (SDRN) estimator of the regression function obtained from regularized empirical risk minimization with a Lipschitz loss function. Our framework can be applied to a variety of regression and classification problems. We establish novel nonasymptotic excess risk bounds for our SDRN estimator when the regression function belongs to a Sobolev space with mixed derivatives. We obtain a new, nearly optimal, risk rate in the sense that the SDRN estimator can achieve nearly the same optimal minimax convergence rate as one-dimensional nonparametric regression with the dimension involved in a logarithm term only when the feature dimension is fixed. The estimator has a slightly slower rate when the dimension grows with the sample size. We show that the depth of the SDRN estimator grows with the sample size in logarithmic order, and the total number of nodes and weights grows in polynomial order of the sample size to have the nearly optimal risk rate. The proposed SDRN can go deeper with fewer parameters to well estimate the regression and overcome the overfitting problem encountered by conventional feedforward neural networks. 

A two-dimensional electron system exposed to a strong magnetic field produces a plethora of strongly interacting fractional quantum Hall (FQH) states, the complex topological orders of which are revealed through exotic emergent particles, such as composite fermions, and fractionally charged Abelian and non-Abelian anyons. Much insight has been gained by the study of multicomponent FQH states, where spin and pseudospin indices of the electron contribute additional correlation. Traditional multicomponent FQH states develop in situations where the components share the same orbital states and the resulting interactions are pseudospin independent; this homo-orbital nature is also crucial to their theoretical understanding. Here, we study “hetero-orbital” two-component FQH states, in which the orbital index is part of the pseudospin, rendering the multicomponent interactions strongly SU(2) anisotropic in the pseudospin space. Such states, obtained in bilayer graphene at the isospin transition between $\mathit{N}=0$and $\mathit{N}=1$electron Landau levels, are markedly different from previous homo-orbital two-component FQH states. In particular, we observe strikingly different behaviors for the parallel-vortex and reverse-vortex attachment composite fermion states, and an anomalously strong two-component $2/5$state over a wide range of magnetic field before it abruptly disappears at a high field. Our findings, combined with detailed theoretical calculations, reveal the surprising robustness of the hetero-orbital FQH effects, significantly enriching our understanding of FQH physics in this novel regime. 

Water-processable MXene layer applied to microneedle toward on-demand drug delivery bioelectronics. 

The northwest-trending Altai Mountains of central Asia expose a complex network of thrust and strike-slip faults that are key features accommodating intracontinental crustal shortening related to the Cenozoic India-Asia collision. In this study, we investigated the Quaternary slip history of the Fuyun fault, a right-lateral strike-slip fault bounding the southwestern margin of the Altai Mountains, through geologic mapping, geomorphic surveying, and optically stimulated luminescence (OSL) geochronology. At the Kuoyibagaer site, the Fuyun fault displaces three generations of Pleistocene–Holocene fill-cut river terraces (i.e., T3, T2, and T1) containing landslide and debris-flow deposits. The right-lateral offsets are magnified by erosion of terrace risers, suggesting that river course migration has been faster than slip along the Fuyun fault. The highest Tp2 terrace was abandoned in the middle Pleistocene (150.4 ± 8.1 ka uppermost OSL age) and was displaced 145.5 +45.6/–12.1 m along the Fuyun fault, yielding a slip rate of 1.0 +0.4/–0.1 mm/yr since the middle Pleistocene. The lower Tp1 terrace was abandoned in the late Pleistocene and aggraded by landslides and debris flows in the latest Pleistocene–Holocene (36.7 ± 1.6 ka uppermost OSL age). Tp1 was displaced 67.5 +14.2/–6.1 m along the Fuyun fault, yielding a slip rate of 1.8 +0.5/–0.2 mm/yr since the late Pleistocene. Our preferred minimum slip rate of ~1 mm/yr suggests the Fuyun fault accommodates ~16% of the average geodetic velocity of ~6 mm/yr across the Altai Mountains. Integration of our new Fuyun slip rate with other published fault slip rates accounts for ~4.2 mm/yr of convergence across the Chinese Altai, or ~70% of the geodetic velocity field. 

Abstract Platelets play a pivotal role in hemostasis and wound healing and conditional shape change is an important component of platelet functionality. In normal circumstances, platelets travel through the circulatory system in an inactive rounded state, which enables platelets to easily move to vessel walls for attachment. When an injury occurs, platelets are prompted by molecules, such as thrombin, to shift into a stellate shape and increase exposure of fibrin‐binding receptors. When active, platelets promote hemostasis and clot retraction, which enhances clot stability and promotes healing. However, in conditions where platelets are depleted or hyporeactive, these functions are diminished and lead to inhibited hemostasis and healing. To treat platelet depletion, our group developed platelet‐like particles (PLPs) which consist of highly deformable microgels coupled to fibrin binding motif. However, first generation PLPs do not exhibit wound‐triggered shape change like native platelets. Thus, the objective of these studies was to develop a PLP formulation that changes shape when prompted by thrombin. To create thrombin‐sensitive PLPs (TS‐PLPs), we incorporated a thrombin‐cleavable peptide into the microgel body and then evaluated PLP properties before and after exposure to thrombin including morphology, size, and in vitro clot retraction. Once thrombin‐prompted shape change ability was confirmed, the TS‐PLPs were tested in vivo for hemostatic ability and subsequent wound healing outcomes in a murine liver trauma model. We found that TS‐PLPs exhibit a wound‐triggered shape change, induce significant clot retraction following exposure to thrombin and promote hemostasis and healing in vivo after trauma. 

Binary neural network (BNN) delivers increased compute intensity and reduces memory/data requirements for computation. Scalable BNN enables inference in a limited time due to different constraints. This paper explores the application of Scalable BNN in oblivious inference, a service provided by a server to mistrusting clients. Using this service, a client can obtain the inference result on his/her data by a trained model held by the server without disclosing the data or learning the model parameters. Two contributions of this paper are: 1) we devise lightweight cryptographic protocols explicitly designed to exploit the unique characteristics of BNNs. 2) we present an advanced dynamic exploration of the runtime-accuracy tradeoff of scalable BNNs in a single-shot training process. While previous works trained multiple BNNs with different computational complexities (which is cumbersome due to the slow convergence of BNNs), we train a single BNN that can perform inference under various computational budgets. Compared to CryptFlow2, the state-of-the-art technique in the oblivious inference of non-binary DNNs, our approach reaches 3 × faster inference while keeping the same accuracy. Compared to XONN, the state-of-the-art technique in the oblivious inference of binary networks, we achieve 2 × to 12 × faster inference while obtaining higher accuracy.