Selected work

Representative Papers

From data to design: Random forest regression model for predicting mechanical properties of alloy steel

Nov 04, 2025

This study addresses the challenge of accurately predicting mechanical properties—namely elongation, tensile strength, and yield strength—of alloy steels. We propose an ensemble learning prediction framework based on random forest regression. The model leverages chemical composition (Fe, Cr, Ni, Mn, Si, Cu, C, etc.) and cold-rolling reduction ratio as input features, integrated with systematic feature engineering, five-fold cross-validation, residual analysis, and learning curve diagnostics for robust modeling and optimization. Compared to conventional empirical formulas and single-model approaches, the proposed framework significantly enhances nonlinear relationship modeling capability and prediction robustness. On the test set, it achieves R² scores of 0.92–0.96 and reduces root-mean-square error (RMSE) by over 35%. These results demonstrate its practical utility in alloy design and process optimization, underscoring strong potential for industrial deployment.

1 citationsRead paper

Cellular Automata based Resource Efficient Maximally Equidistributed Pseudo-Random Number Generators

Mar 20, 2026

This work proposes a lightweight, composite cellular automaton (CA)-based pseudorandom number generator (PRNG) that overcomes the longstanding trade-off between maximal period length and high-dimensional equidistribution in linear CA-based designs. By integrating near-word-length linear maximum-length CAs through a temporal interleaving strategy, the proposed generator achieves—for the first time—simultaneous maximality in both period and equidistribution while maintaining computational efficiency. Theoretical analysis and empirical evaluations demonstrate that the generator passes nearly all standard statistical randomness tests, exhibiting performance and speed comparable to the Mersenne Twister while significantly enhancing the statistical quality and practical applicability of CA-based PRNGs.

0 citationsRead paper

Algebraic Characterization of Reversible First Degree Cellular Automata over $\mathbb{Z}_d$

Mar 05, 2026

This study addresses the problem of determining global reversibility for a class of one-dimensional, three-neighborhood, d-state first-degree cellular automata (FDCA) under zero boundary conditions, for arbitrary lattice sizes. By analyzing the eight parameters defining the local rule, the work proposes that verifying just three algebraic conditions is sufficient to guarantee reversibility for any lattice size, thereby enabling constant-time reversibility checking and rule synthesis. Built upon algebraic condition analysis, finite cellular automaton modeling, and exploration of the parameterized rule space, this approach establishes the first general framework for generating and verifying reversible FDCA rules applicable to any number of states d. The framework significantly enhances both the design efficiency and theoretical tractability of reversible cellular automata.

0 citationsRead paper

FedNET: Federated Learning for Proactive Traffic Management and Network Capacity Planning

Nov 10, 2025

To address the dual challenges of early identification of high-risk links and simultaneous privacy preservation and global modeling in large-scale communication networks, this paper proposes a federated learning-based distributed multi-step traffic forecasting framework. The method enables privacy-preserving node-level time-series modeling without sharing raw node traffic data, and infers link-level load evolution via source-destination routing mapping and traffic superposition, thereby supporting proactive congestion预警 and capacity planning. This work innovatively introduces federated learning to multi-step network traffic forecasting—marking the first decentralized approach to link risk assessment. Experiments demonstrate strong performance: short-term forecasts achieve R² > 0.92, while long-term forecasts attain R² = 0.45–0.55—comparable to centralized training. Moreover, on real-world topologies, the framework accurately identifies high-risk links up to three days in advance.

0 citationsRead paper

In-Memory Sorting-Searching with Cayley Tree

Jun 24, 2025

To address high CPU utilization and severe memory-access bottlenecks in data-intensive computing, this paper proposes a novel in-memory computing (IMC) architecture based on the Cayley tree. By reconfiguring conventional memory cells into logic-enabled processing nodes, the architecture natively supports in-memory search, extremum (max/min) computation, and sorting—significantly reducing CPU involvement. It achieves O(log n) time complexity for search and extremum operations—distribution-agnostic—and O(n log n) for in-memory sorting. Two FPGA prototypes are implemented: a lightweight version compatible with standard memory interfaces and a fully customized Cayley-tree memory architecture. Experimental results demonstrate that, compared to state-of-the-art IMC designs, our approach delivers both asymptotically lower time complexity and superior energy efficiency for search and sorting workloads.

0 citationsRead paper

Recent publications

Latest Papers

Cellular Automata based Resource Efficient Maximally Equidistributed Pseudo-Random Number Generators

Mar 20, 2026

This work proposes a lightweight, composite cellular automaton (CA)-based pseudorandom number generator (PRNG) that overcomes the longstanding trade-off between maximal period length and high-dimensional equidistribution in linear CA-based designs. By integrating near-word-length linear maximum-length CAs through a temporal interleaving strategy, the proposed generator achieves—for the first time—simultaneous maximality in both period and equidistribution while maintaining computational efficiency. Theoretical analysis and empirical evaluations demonstrate that the generator passes nearly all standard statistical randomness tests, exhibiting performance and speed comparable to the Mersenne Twister while significantly enhancing the statistical quality and practical applicability of CA-based PRNGs.

0 citationsRead paper

Algebraic Characterization of Reversible First Degree Cellular Automata over $\mathbb{Z}_d$

Mar 05, 2026

This study addresses the problem of determining global reversibility for a class of one-dimensional, three-neighborhood, d-state first-degree cellular automata (FDCA) under zero boundary conditions, for arbitrary lattice sizes. By analyzing the eight parameters defining the local rule, the work proposes that verifying just three algebraic conditions is sufficient to guarantee reversibility for any lattice size, thereby enabling constant-time reversibility checking and rule synthesis. Built upon algebraic condition analysis, finite cellular automaton modeling, and exploration of the parameterized rule space, this approach establishes the first general framework for generating and verifying reversible FDCA rules applicable to any number of states d. The framework significantly enhances both the design efficiency and theoretical tractability of reversible cellular automata.

0 citationsRead paper

FedNET: Federated Learning for Proactive Traffic Management and Network Capacity Planning

Nov 10, 2025

To address the dual challenges of early identification of high-risk links and simultaneous privacy preservation and global modeling in large-scale communication networks, this paper proposes a federated learning-based distributed multi-step traffic forecasting framework. The method enables privacy-preserving node-level time-series modeling without sharing raw node traffic data, and infers link-level load evolution via source-destination routing mapping and traffic superposition, thereby supporting proactive congestion预警 and capacity planning. This work innovatively introduces federated learning to multi-step network traffic forecasting—marking the first decentralized approach to link risk assessment. Experiments demonstrate strong performance: short-term forecasts achieve R² > 0.92, while long-term forecasts attain R² = 0.45–0.55—comparable to centralized training. Moreover, on real-world topologies, the framework accurately identifies high-risk links up to three days in advance.

0 citationsRead paper

From data to design: Random forest regression model for predicting mechanical properties of alloy steel

Nov 04, 2025

This study addresses the challenge of accurately predicting mechanical properties—namely elongation, tensile strength, and yield strength—of alloy steels. We propose an ensemble learning prediction framework based on random forest regression. The model leverages chemical composition (Fe, Cr, Ni, Mn, Si, Cu, C, etc.) and cold-rolling reduction ratio as input features, integrated with systematic feature engineering, five-fold cross-validation, residual analysis, and learning curve diagnostics for robust modeling and optimization. Compared to conventional empirical formulas and single-model approaches, the proposed framework significantly enhances nonlinear relationship modeling capability and prediction robustness. On the test set, it achieves R² scores of 0.92–0.96 and reduces root-mean-square error (RMSE) by over 35%. These results demonstrate its practical utility in alloy design and process optimization, underscoring strong potential for industrial deployment.

1 citationsRead paper

In-Memory Sorting-Searching with Cayley Tree

Jun 24, 2025

To address high CPU utilization and severe memory-access bottlenecks in data-intensive computing, this paper proposes a novel in-memory computing (IMC) architecture based on the Cayley tree. By reconfiguring conventional memory cells into logic-enabled processing nodes, the architecture natively supports in-memory search, extremum (max/min) computation, and sorting—significantly reducing CPU involvement. It achieves O(log n) time complexity for search and extremum operations—distribution-agnostic—and O(n log n) for in-memory sorting. Two FPGA prototypes are implemented: a lightweight version compatible with standard memory interfaces and a fully customized Cayley-tree memory architecture. Experimental results demonstrate that, compared to state-of-the-art IMC designs, our approach delivers both asymptotically lower time complexity and superior energy efficiency for search and sorting workloads.

0 citationsRead paper