Research Output

Publications

Peer-reviewed journal articles by DeepMathAI research group members — each with a graphical abstract and summary.

12
Published
0
Accepted
12
Total

2026 – 20272 articles

October 2026 – September 2027

PublishedApplied Math2026

A Parallel Moving Point Projection Algorithm for Generalized Split Convex Feasibility Problems with Applications to Image Restoration

Thammasiri, P., Ungchittrakool, K., Plubtieng, S., & Wangkeeree, R.

A parallel moving point projection algorithm is developed for generalized split convex feasibility problems in real Hilbert spaces. Closed-form projections onto two adaptive half-spaces avoid solving constrained quadratic subproblems at each iteration. The method is proved to converge strongly and is evaluated on synthetic problems and color image restoration with blur and Gaussian noise.

Symmetry (MDPI)
Vol. 18, Issue 11, 1793
DOI: 10.3390/sym18111793
PublishedOptimization2027

Fixed-Time Stability and Noise Robustness of a Polynomial-Gain Gradient Neural Network for Absolute Value Equations

Grace, T., Tananimit, N., Belay, Y. A., & Wangkeeree, R.

A gradient neural network with a polynomial gain is proposed for solving absolute value equations. The paper establishes fixed-time stability, guaranteeing convergence within a time bound independent of the initial state, and analyzes the robustness of the network dynamics under noise.

Carpathian Journal of Mathematics
Vol. 43, Issue 1, pp. 233–248
View Article

2025 – 202610 articles

October 2025 – September 2026

PublishedApplied Math2026

Explicit Run Length Evaluation Under Asymmetric One-Sided and Symmetric Two-Sided Control-Limit Schemes for a Cubic Trend AR Model and Its Application

Karoon, K., & Areepong, Y.

An explicit formula for the Average Run Length (ARL) of the Extended EWMA control chart is derived for an autoregressive model with cubic trend and exponential white noise, under both asymmetric one-sided and symmetric two-sided control limits. The explicit formula agrees with trapezoidal, Simpson, and Boole numerical integral equation approximations at about 100% accuracy without repeated numerical integration, and the Extended EWMA chart detects small and moderate process shifts faster and more consistently than the standard EWMA chart.

Symmetry (MDPI)
Vol. 18, Issue 10, 1620
DOI: 10.3390/sym18101620
PublishedMedical AI2026

An Anatomically Guided and Optimization-Refined Radiomics Framework for Opportunistic Osteoporosis Assessment from Lumbar Spine MRI

Mahatthanatrakul, A., Klinsuwan, T., Wangkeeree, R., & Laoruengthana, A.

An anatomically guided radiomics pipeline that opportunistically screens for osteoporosis from routine sagittal lumbar-spine MRI. The framework automatically localizes and segments the L1–L4 vertebral bodies, extracts a 546-column radiomic feature vector (GLCM, LBP, wavelet, Gabor, Hu moments, and first-order statistics), and applies optimization-refined SVM classification and SVR regression to estimate bone mineral density and T-score — enabling early osteoporosis detection from scans acquired for other clinical reasons.

Diagnostics (MDPI)
Vol. 16, Issue 14, 2241, 2026
DOI: 10.3390/diagnostics16142241
PublishedOptimization2026

A Projection Recurrent Neural Network Method for Solving Absolute Value Equations Associated with Second-Order Cones

Wangkeeree, R., Wangkeeree, R., Nazemi, A., Belay, Y.A., & Ungchittrakool, K.

This paper proposes a projection recurrent neural network for solving absolute value equations (AVE) constrained within second-order cones. The network is proven to be globally convergent to the exact solution using Lyapunov stability theory, offering a continuous-time dynamical approach that outperforms classical iterative solvers.

Journal of Computational and Applied Mathematics (Elsevier)
Vol. 485, 15 October 2026, 117567
DOI: 10.1016/j.cam.2026.117567
PublishedOptimization2026

A Projection Neural Network With Delays and Optimization Approaches for Solving Absolute Value Equations

Prathom, K., Wangkeeree, R., Belay, Y.A., & Hongsri, A.

A novel projection neural network incorporating time delays is developed to solve absolute value equations. The delay dynamics capture real-world computational latencies, and the model is combined with optimization techniques to ensure global stability. Experiments demonstrate superior convergence speed compared to delay-free counterparts.

IEEE Access
Vol. 14, pp. 50141–50155, March 2026
DOI: 10.1109/ACCESS.2026.3679347
PublishedApplied Math2026

A relaxed iterative method for approximating solutions of pseudomonotone hierarchical variational inequality problems in Hilbert spaces

Gebremeskel, K.G., Tewele, T.G., Wangkeeree, R., Belay, Y.A., & Meche, T.H.

This work introduces a relaxed iterative algorithm for solving hierarchical variational inequality problems in infinite-dimensional Hilbert spaces under pseudomonotonicity assumptions. Weak and strong convergence theorems are rigorously established, with numerical experiments validating the theoretical results.

The Journal of Analysis (Springer)
08 May 2026
DOI: 10.1007/s41478-026-01096-8
PublishedApplied Math2027

Continuity and Lipschitz Properties of Approximate Weak Solution Maps via Nonlinear Scalarization

Preechasilp, P., & Wangkeeree, R.

This paper investigates the stability of solution maps for approximate weak vector optimization problems. Using a nonlinear scalarization technique, Hölder continuity and Lipschitz properties of the solution maps are derived under parametric perturbations, contributing to the theoretical foundations of robust optimization.

Carpathian Journal of Mathematics
Vol. 43, Issue 1, pp. 69–91
View Article
PublishedMedical AI2026

Multi-view machine learning with an optic disc localization for glaucoma diagnosis

Siying, P., Muangphara, T., Photun, A., Suppalap, S., Klinsuwan, T., Phruancharoen, C., Treeyawedkul, S., Chira-adisai, T., Supattanawong, Y., & Wangkeeree, R.

An AI-powered glaucoma screening system that combines optic disc localization with multi-view machine learning. The model fuses features from multiple retinal fundus image perspectives to achieve robust classification of glaucoma vs. normal cases, offering a clinically applicable diagnostic tool validated on real patient data.

Applied Sciences (MDPI)
Vol. 16, Issue 7, 3158
DOI: 10.3390/app16073158
PublishedApplied Math2026

Inertial Forward–Backward–Forward Algorithm with Moving Point Projection for Monotone Inclusions and Image Restoration

Thammasiri, P., Berinde, V., Plubtieng, S., Ungchittrakool, K., & Wangkeeree, R.

A three-step inertial Forward–Backward–Forward (FBF) splitting algorithm with a moving projection operator is introduced for solving monotone inclusion problems. The inertial extrapolation significantly accelerates convergence. Applications to image restoration under various noise models confirm the method's competitive performance against state-of-the-art algorithms.

Symmetry (MDPI)
Vol. 18, Issue 5, 782
DOI: 10.3390/sym18050782
PublishedOptimization2027

Multi-view Laplacian twin support vector machine with pinball loss function

Damminsed, V., & Wangkeeree, R.

This paper presents a Multi-view Laplacian Twin SVM that integrates pinball loss for improved robustness to noise and outliers. The Laplacian graph regularization captures geometric structure across multiple data views, while the twin hyperplane formulation reduces computational complexity. The model achieves state-of-the-art classification accuracy on benchmark datasets.

Carpathian Journal of Mathematics
Vol. 43, Issue 1, pp. 139–165
View Article
PublishedOptimization2026

Compact Gradient-Based Neural Network for Stochastic Support Vector Regression with Probabilistic Constraints

Tananimit, N., Grace, T., Belay, Y. A., & Wangkeeree, R.

A compact gradient-based neural network is designed to solve stochastic support vector regression (SSVR) with probabilistic constraints. Using a smoothed Fischer–Burmeister function, the nonsmooth KKT conditions are reformulated, yielding a numerically stable gradient flow. The proposed architecture achieves a 25% reduction in network complexity with faster training times while outperforming standard SVR implementations on UCI benchmarks.

Journal of Nonlinear and Variational Analysis
Vol. 10, Issue 6
DOI: 10.23952/jnva.10.2026.6.04