CISMN: A Chaos-Integrated Synaptic-Memory Network with Multi-Compartment Chaotic Dynamics for Robust Nonlinear Regression DOI Creative Commons
Yaser Shahbazi, Mohsen Mokhtari Kashavar, Abbas Ghaffari

et al.

Mathematics, Journal Year: 2025, Volume and Issue: 13(9), P. 1513 - 1513

Published: May 4, 2025

Modeling complex, non-stationary dynamics remains challenging for deterministic neural networks. We present the Chaos-Integrated Synaptic-Memory Network (CISMN), which embeds controlled chaos across four modules—Chaotic Memory Cells, Chaotic Plasticity Layers, Synapse and a Attention Mechanism—supplemented by logistic-map learning-rate schedule. Rigorous stability analyses (Lyapunov exponents, boundedness proofs) gradient-preservation guarantees underpin our design. In experiments, CISMN-1 on synthetic acoustical regression dataset (541 samples, 22 features) achieved R2 = 0.791 RMSE 0.059, outpacing physics-informed attention-augmented baselines. CISMN-4 PMLB sonar benchmark (208 60 bands) attained 0.424 0.380, surpassing LSTM, memristive, reservoir models. Across seven standard tasks with 5-fold cross-validation, CISMN led diabetes (R2 0.483 ± 0.073) excelled in high-dimensional, low-sample regimes. Ablations reveal scalability–efficiency trade-off: lightweight variants train <10 s >95% peak accuracy, while deeper configurations yield marginal gains. sustains gradient norms (~2300) versus LSTM collapse (<3), fixed-seed protocols ensure <1.2% MAE variation. Interpretability (feature-attribution entropy ≈ 2.58 bits), motivating future hybrid explanation methods. recasts as computational asset robust, generalizable modeling scientific, financial, engineering domains.

Language: Английский

Chaos-based approaches to digital data security: Analysis of incommensurate fractional-order Arneodo chaotic system and engineering application on Nvidia Jetson AGX Orin DOI
Akif Akgül, Mustafa YAZ, Berkay Emi̇n

et al.

Integration, Journal Year: 2025, Volume and Issue: unknown, P. 102355 - 102355

Published: Jan. 1, 2025

Language: Английский

Citations

3
Dynamic Analysis and Implementation of FPGA for a New 4D Fractional-Order Memristive Hopfield Neural Network DOI Creative Commons
Fei Yu,

Shankou Zhang,

Dan Su

et al.

Fractal and Fractional, Journal Year: 2025, Volume and Issue: 9(2), P. 115 - 115

Published: Feb. 13, 2025

Memristor-based fractional-order chaotic systems can record information from the past, present, and future, describe real world more accurately than integer-order systems. This paper proposes a novel memristor model verifies its characteristics through pinched loop (PHL) method. Subsequently, new memristive Hopfield neural network (4D-FOMHNN) is introduced to simulate induced current, accompanied by Caputo’s definition of fractional order. An Adomian decomposition method (ADM) employed for system solution. By varying parameters order 4D-FOMHNN, rich dynamic behaviors including transient chaos, coexistence attractors are observed using methods such as bifurcation diagrams Lyapunov exponent analysis. Finally, proposed FOMHNN implemented on field-programmable gate array (FPGA), oscilloscope observation results consistent with MATLAB numerical simulation results, which further validate theoretical analysis provide basis application in field encryption.

Language: Английский

Citations

3
Fractional-order bi-Hopfield neuron coupled via a multistable memristor: Complex neuronal dynamic analysis and implementation with microcontroller DOI
Victor Kamdoum Tamba,

Arsene Loic Mbanda Biamou,

Viet–Thanh Pham

et al.

AEU - International Journal of Electronics and Communications, Journal Year: 2025, Volume and Issue: unknown, P. 155661 - 155661

Published: Jan. 1, 2025

Language: Английский

Citations

1
CISMN: A Chaos-Integrated Synaptic-Memory Network with Multi-Compartment Chaotic Dynamics for Robust Nonlinear Regression DOI Creative Commons
Yaser Shahbazi, Mohsen Mokhtari Kashavar, Abbas Ghaffari

et al.

Mathematics, Journal Year: 2025, Volume and Issue: 13(9), P. 1513 - 1513

Published: May 4, 2025

Modeling complex, non-stationary dynamics remains challenging for deterministic neural networks. We present the Chaos-Integrated Synaptic-Memory Network (CISMN), which embeds controlled chaos across four modules—Chaotic Memory Cells, Chaotic Plasticity Layers, Synapse and a Attention Mechanism—supplemented by logistic-map learning-rate schedule. Rigorous stability analyses (Lyapunov exponents, boundedness proofs) gradient-preservation guarantees underpin our design. In experiments, CISMN-1 on synthetic acoustical regression dataset (541 samples, 22 features) achieved R2 = 0.791 RMSE 0.059, outpacing physics-informed attention-augmented baselines. CISMN-4 PMLB sonar benchmark (208 60 bands) attained 0.424 0.380, surpassing LSTM, memristive, reservoir models. Across seven standard tasks with 5-fold cross-validation, CISMN led diabetes (R2 0.483 ± 0.073) excelled in high-dimensional, low-sample regimes. Ablations reveal scalability–efficiency trade-off: lightweight variants train <10 s >95% peak accuracy, while deeper configurations yield marginal gains. sustains gradient norms (~2300) versus LSTM collapse (<3), fixed-seed protocols ensure <1.2% MAE variation. Interpretability (feature-attribution entropy ≈ 2.58 bits), motivating future hybrid explanation methods. recasts as computational asset robust, generalizable modeling scientific, financial, engineering domains.

Language: Английский

Citations

0