Liquid Time-Constant Neural Networks with Runge Kutta and Euler Solvers for High-Accuracy Intrusion Detection
Intrusion detection systems, Liquid Time Consatant Neural networks, Runge-Kutta method, Euler method, Discrete and Continuous time neural dynamics.
AuthorsAbstractA Liquid time constant neural network-based intrusion detection sytem (LTCNN-IDS), is an innovative approach for optimizing the feature extraction process to enhance the performance of intrusion detection systems. To simulate the continuous-time dynamics of LTCNNs, the authors combined intrinsic differential equations that regulate neuronal state evolution. While existing work has mostly used discrete time-based Recurrent neural network variations such as Long short-term memory, this study investigates the use of basic first-order solvers such as Euler's method, and the fourth order solver, the Runge-Kutta method (RK4) in the LTCNN architecture. The Euler method is one of the simplest and most intuitive method for numerically solving ordinary differential equations (ODEs). It is particularly useful in neural models such as LTCNNs, where the dynamics of neurons are described by differential equations. This allows the simulation of continuous-time neural dynamics in discrete-time steps, where the RK4 solver uses a higher-order approximation of latent dynamics, resulting in more accurate and stable updates throughout the temporal integration. Including RK4 in the network architecture enables end-to-end training of latent ODE-based systems, combining the interpretability of dynamical systems with the flexibility of deep learning. The performance of the proposed technique is compared with that of LSTM. The results exhibited an accuracy exceeding 0.99.
•••••••••••••••••••••••••••••••• ejprd.org - Published by Riset Publication Services LLC
EJPRD
Copyright ©2026 by Riset Publication Services LLC