arXiv subjects

Peng Lan

Publications and source records attributed to Peng Lan.

7 recordsLinked to original sources

arXiv2026

Probabilistic indirect models for undrained shear strength: addressing significant data missing and variability with advanced imputation and machine learning techniques

Accurate prediction of undrained shear strength (su) is crucial for geotechnical design, but is often hampered by substantial uncertainty in traditional empirical methods. This study uses the CLAY/10/7490 global database to develop probabilistic indirect models to predict su based on Atterberg limits and piezocone cone penetration (CPTU) measurements. Firstly, the dataset has a high missing data rate and variability. We test three imputation methods - multivariate normal (MN), multiple imputation by chained equations (MICE), and miss forest (MF) - to fill the missing values. To validate their effectiveness, a Probabilistic Extreme Gradient Boosting (PXGB) model is developed, and the imputation methods are evaluated by comparing the PXGB's performance when trained on the imputed datasets against that on the original incomplete data. Secondly, the indirect model is built by integrating a multi-head attention (MHA) mechanism into an artificial neural network (ANN) to enhance information extraction from limited data, which leads to the MHA-based probabilistic neural networks (MHA-PNN) model. The models' performance, alongside a conventional MN-based prediction model, was evaluated using root mean square error (RMSE), coefficient of determination (R2), mean absolute percentage error (MAPE), conditional interval width (wCI), and coverage rate (CR). Results demonstrate that the proposed MN-enhanced MHA-PNN model substantially outperforms other models in both prediction accuracy and uncertainty quantification. These findings highlight the potential of this integrated strategy for building robust probabilistic indirect models in geotechnical applications, particularly when confronted with sparse and incomplete datasets.

cs.LG↗

arXiv2026

Private Information Retrieval from Joint Systematic MDS-Coded with Non-Colluding Servers: Bounds and Constructions

Consider a distributed storage system consisting of $N$ non-colluding servers that collectively store a database of $M$ files encoded using an $[N,K]$ maximum distance separable(MDS) code. A user wishes to retrieve one file privately by accessing the servers without revealing the identity of the requested file. A scheme designed for this purpose is called a joint MDS-coded private information retrieval(PIR) scheme, which was first introduced by Sun and Tian in 2019 to break the capacity $\frac{1-K/N}{1-(K/N)^M}$ of the separate MDS-coded PIR schemes established by Banawan and Ulukus. However, the capacity of joint MDS-coded PIR remains largely unexplored. In this paper, we study the capacity of joint MDS-coded PIR with systematic MDS array storage codes under prescribed storage patterns. Specifically, we first derive upper bounds on the capacity of joint MDS-coded PIR for $K=Mt$ and $K=Mt+1$, respectively. We then construct three joint MDS-coded PIR schemes for the cases $N\le K+t, K=Mt$, $N>K+t, K=Mt$ and $N\le K+t, K=Mt+1$. The proposed schemes require small file sizes and achieve higher retrieval rates: the first and third schemes exceed the capacity of separate MDS-coded PIR schemes, while the second scheme does so when the storage rate $\frac{K}{N}>r_M$ for some $0<r_M<\frac{M}{M+1}$. In particular, for $K=Mt$ and $N\leq K+t$, the proposed scheme achieves the derived upper bound, thereby establishing that the optimal joint MDS-coded PIR capacity under the considered storage pattern is $1-(1-\frac{1}{M})\frac{K}{N}$. Compared with capacity-achieving separate MDS-coded PIR schemes at the same storage-code rate, the proposed schemes may achieve a substantial relative retrieval-rate improvement: the maximum improvement can exceed $15\%$ when $M\geq 4$, exceed $20\%$ when $M\geq 9$, and asymptotically approach $1-2/e\approx 26.42\%$ as M increases.

cs.IT↗

arXiv2024

Influence of control polygon on the generalization of the conversion between ANCF and B-spline surfaces

The aim of this study is to establish a general transformation matrix between B-spline surfaces and ANCF surface elements. This study is a further study of the conversion between the ANCF and B-spline surfaces. In this paper, a general transformation matrix between the Bezier surfaces and ANCF surface element is established. This general transformation matrix essentially describes the linear relationship between ANCF and Bezier surfaces. Moreover, the general transformation matrix can help to improve the efficiency of the process to transfer the distorted configuration in the CAA back to the CAD, an urgent requirement in engineering practice. In addition, a special Bezier surface control polygon is given in this study. The Bezier surface described with this control polygon can be converted to an ANCF surface element with fewer d.o.f.. And the converted ANCF surface element with 36 d.o.f. was once addressed by Dufva and Shabana. So the special control polygon can be regarded as the geometric condition in conversion to an ANCF surface element with 36 d.o.f. Based on the fact that a B-spline surface can be seen as a set of Bezier surfaces connected together, the method to establish a general transformation matrix between the ANCF and lower-order B-spline surfaces is given. Specially, the general transformation is not in a recursive form, but in a simplified form.

cs.CG↗

arXiv2024

General Conversion between ANCF and B-spline Surfaces

In this paper, general conversion equations are derived between Absolute Nodal Coordinates Formulation (ANCF) finite surface elements and B-spline surfaces, an extension of our previous work on the conversion between ANCF cable elements and B-spline curves. The derivation of the conversion equations is the discovery of the geometric invariance of the ANCF displacement field before and after the conversion. Our study starts from proposing the conversion equation between ANCF finite surface elements and Bezier surfaces which are the special cases of B-spline surfaces, followed by establishing a general conversion equation between ANCF finite surface elements and Bezier surfaces. This general conversion equation has functionalities (1) to realize the one-step direct conversion between ANCF and Bezier surfaces (2) to convert ANCF finite surface elements directly to Bezier surfaces provided the ANCF nodal coordinates are not independent. The direct conversion from a conditional ANCF finite surface to Bezier surfaces enhances the efficiency and ability to control and store data in computers during the conversion process. The conversion between ANCF finite surface elements and B-spline surfaces is derived from a conversion of B-spline surfaces to a more general conversion of B-spline surfaces. B-spline basis functions are utilized in the non-recursive form, from which a more efficient conversion equation is obtained compared with an intuitive conversion semantics where one converts firstly B-spline surfaces to composite Bezier surfaces by inserting knot and converts to ANCF finite surface elements afterward. The obtained conversion equations between ANCF and B-spline surfaces realize the one-step direct conversion.

cs.CG↗

arXiv2024

Feedback Intensity Equalization Algorithm for Multi-Spots Holographic Tweezer

High degree of adjustability enables the holographic tweezer array a versatile platform for creating an arbitrary geometrical atomic array. In holographic tweezer array experiments, an optical tweezer generated by a spatial light modulator (SLM) usually is used as a static tweezer array. However, the alternating current (AC) stark effect generally induces the intensity difference of traps in terms of different light shifts. So, intensity equalization is an essential prerequisite for preparing a many-body system with individually controlled atoms. Here, we report an intensity equalization algorithm. In particular, we observe the non-uniformity of the tweezer array is below 1.1% when the array size is larger than 1000. Our analysis shows that by optimizing the hardware performance of the optical system, this uniformity could be further improved. Our work offers the opportunities for large-scale quantum computation and simulation with reconfigurable atom arrays.

physics.optics↗

arXiv2023

Elastically-Constrained Meta-Learner for Federated Learning

Federated learning is an approach to collaboratively training machine learning models for multiple parties that prohibit data sharing. One of the challenges in federated learning is non-IID data between clients, as a single model can not fit the data distribution for all clients. Meta-learning, such as Per-FedAvg, is introduced to cope with the challenge. Meta-learning learns shared initial parameters for all clients. Each client employs gradient descent to adapt the initialization to local data distributions quickly to realize model personalization. However, due to non-convex loss function and randomness of sampling update, meta-learning approaches have unstable goals in local adaptation for the same client. This fluctuation in different adaptation directions hinders the convergence in meta-learning. To overcome this challenge, we use the historical local adapted model to restrict the direction of the inner loop and propose an elastic-constrained method. As a result, the current round inner loop keeps historical goals and adapts to better solutions. Experiments show our method boosts meta-learning convergence and improves personalization without additional calculation and communication. Our method achieved SOTA on all metrics in three public datasets.

cs.LG↗

arXiv2022

Walle: An End-to-End, General-Purpose, and Large-Scale Production System for Device-Cloud Collaborative Machine Learning

To break the bottlenecks of mainstream cloud-based machine learning (ML) paradigm, we adopt device-cloud collaborative ML and build the first end-to-end and general-purpose system, called Walle, as the foundation. Walle consists of a deployment platform, distributing ML tasks to billion-scale devices in time; a data pipeline, efficiently preparing task input; and a compute container, providing a cross-platform and high-performance execution environment, while facilitating daily task iteration. Specifically, the compute container is based on Mobile Neural Network (MNN), a tensor compute engine along with the data processing and model execution libraries, which are exposed through a refined Python thread-level virtual machine (VM) to support diverse ML tasks and concurrent task execution. The core of MNN is the novel mechanisms of operator decomposition and semi-auto search, sharply reducing the workload in manually optimizing hundreds of operators for tens of hardware backends and further quickly identifying the best backend with runtime optimization for a computation graph. The data pipeline introduces an on-device stream processing framework to enable processing user behavior data at source. The deployment platform releases ML tasks with an efficient push-then-pull method and supports multi-granularity deployment policies. We evaluate Walle in practical e-commerce application scenarios to demonstrate its effectiveness, efficiency, and scalability. Extensive micro-benchmarks also highlight the superior performance of MNN and the Python thread-level VM. Walle has been in large-scale production use in Alibaba, while MNN has been open source with a broad impact in the community.

cs.LG↗