Mathematical Optimization of Backpropagation Stability Using Gradient-Convergent Strategies for Enhanced Neural Network Performance Pushpendra Singh Danghi, R. Kasthuri, S. Basheer Ahamed, A. Madhavi, K. Ranjith Singh, Gundavarapu Mallikarjuna Rao 2025 International Conference on Metaverse and Current Trends in Computing Icmctc 2025, 2025 To address this, the proposed work advocates for a new Gradient-Convergent Stability Optimization Framework (GCSOF) that improves the stability and performance of backpropagation in artificial neural networks. The conventional backpropagation techniques are known to have problems inherent in gradient instability, slow convergence, and sensitivity in adversarial perturbation. To solve this problem, we design three core mechanisms integrated by GCSOF, Dynamic Hessian Scaling (DHS), Adaptive Gradient Flow Regulation (AGFR), and Entropy-Controlled Learning Rate (ECLR). The second-order derivative analysis is used by DHS to predict instability zones and a dynamic adjustment of learning rates with eigenvalue spreads resulting in controlled gradient magnitudes. AGFR presents a feedback control system that controls gradient updates that align with the optimal paths of curvature. In the meantime, ECLR uses Shannon entropy to adapt learning rates to control how rich information should be in layers to make the model more adaptable in the course of complex learning. The proposed mathematical model is proved to accelerate the convergence rates, suppress the oscillations, and hinder the gradient vanishing in deep networks. The experimental results show that GCSOF outperforms conventional backpropagation algorithms when it comes to stability metrics, accuracy convergence, and the inherent resistance to adversarial noise. This is an innovative framework of mathematically grounded methods for improving the backpropagation efficiency, which leads to the development of robust and high-performing neural network architectures for solving complex learning tasks. Future research will continue to develop GCSOF into high-performance distributed networks.
Mechanical Behavior of Handmade Epoxy-Based Composites Mohammed Al-Farouni, Sanjeev Kumar Joshi, Rajiv Gandhi N., R. Kasthuri, Ankita Joshi E3s Web of Conferences, 2024 The development of products in a variety of industries is now being increasingly influenced by material advancements. Many experts are looking for basic materials that are strong, light, and inexpensive. Solids are often rather thick, whereas light materials are weaker. To attain significant strength while lowering weight, we use composite materials. This work deals with the mixed effects of composites made by hand using an epoxy tar and a hardener along with different fiberes of chopped mate (KC), Kenaf (KA), and Kevlar (KB). Mechanical studies such as tensile (UTS), flexural (FL), impact (IM), and hardness (BHN) were conducted after creating specimens according to standard measurements.
Energy Efficiency in 5G Networks Using Gaussian Mixture Models Zaid Alsalami, R.Kasthuri, Sowmiya S, Pooja Rani, Deepak Kumar Gupta, I. V. Gowtham Kumar Reddy 2024 IEEE International Conference on Communication Computing and Signal Processing Iicccs 2024, 2024 In this research, a new approach for improving energy efficiency of the networks in the 5G context is presented: utilizing the Gaussian Mixture Models. The process includes acquiring data in the first step, training the model in the second step, and using the model to make real-time decisions to minimize resources and inefficient energy consumption in the last step. This includes the collection of data on context and complete flow details of the course of traffic passage as well as actions committed by users. The GMM is then trained using the Expectation-Maximization (EM) algorithm for developing the model and reinforcement learning for real-time decision-making on the developed model. In case studies analyzed, there is evidence of the enhancement of energy consumption by 14% on average absolute reduction in energy per traffic unit, by 11% of the average absolute increase in network throughput and by 49% of the mean absolute reduction of latency in comparison with baseline conditions.
Simulation of Hemoglobin and Oxyhemoglobin Dynamics Using a Robust Computational Technique R. Kasthuri Communications on Applied Nonlinear Analysis, 2024 In this study a robust computational technique is used for a better understand of the dynamics of hemoglobin and oxyhemoglobin in human blood.[1] Results obtained using this technique helps to calculate SpO2 levels. The major goal of this research is to create a mathematical model that can explain how the concentrations of hemoglobin and oxyhemoglobin change over time. A graphical representation has been arrived with the solution of the proposed model using R-K method and MATLAB tool.
Fuzzy inventory model without shortages using GMI approach P Vasanthi, S Ranganayaki, R Kasthuri Journal of Physics Conference Series, 2022 The aim of this research work is to come out with an inventory model without shortages involving set-up cost and holding cost as imprecise variables. The uncertainty is chosen as fuzzy number in triangularized form and the graded mean integration method is applied to defuzzify the parameters. The effect of defuzzification is explained by analyzing for different input values through numerical examples.
A fuzzy purchase model with backorders - GMI method of defuzzification R Kasthuri, P Vasanthi, S Ranganayaki, K Kavithamani Journal of Physics Conference Series, 2022 A fuzzy EOQ model with backorders is considered in which the costs like setup, holding and the penalty price are assumed as triangular FN (fuzzy numbers). Graded mean integration approach which is more simple and accurate is employed to defuzzify the cost functions. The model is illustrated with numerical examples to compute the optimal values.
Retraction: Triangular fuzzy numbers model with cost parameters P. Vasanthi, S. Ranganayaki, R. Kasthuri Journal of Physics Conference Series, 2021 The scope of our research work is to analyzea purchasing model in which the uncontrolled variables are fuzzified. The fuzzy cost parameters are defuzzified and an optimal solution of the ordering quantity, maximum inventory and annual cost are estimated. Numerical example and sensitivity analysis has been done to explain the mathematical model.
Retraction: Optimization of Fuzzy Model for Signed Distance Method S Ranganayaki, R Kasthuri, P Vasanthi Journal of Physics Conference Series, 2021 The research involves the estimation of minimum total cost of an inventory under both stable and imprecise environment. Demand related to unit cost is assumed here. Cost parameters and decision variables are uncertain in nature that are defuzzified by signed distance method. The KKT conditions are applied to optimize the objective function.. Numerical example is contributed to describe the comparison between crisp and fuzzy solutions.
Inventory model with demand dependent on unit price under fuzzy parameters and decision variables S. Ranganayaki, R. Kasthuri, P. Vasanthi International Journal of Recent Technology and Engineering, 2019 An EOQ model with demand dependent on unit price is considered and a new approach of finding optimal demand value is done from the optimal unit cost price after defuzzification. Here the cost parameters like setup cost, holding cost and shortage cost and also the decision variables like unit price, lot size and the maximum inventory are taken under fuzzy environment. Triangular fuzzy numbers are used to fuzzify these input parameters and unknown variables. For the proposed model an optimal solution has been determined using Karush Kuhn-Tucker conditions method. Graded Mean Integration (GMI) method is used for defuzzification. Numerical solutions are obtained and sensitivity analysis is done for the chosen model
Fuzzy eoq model with shortages using kuhns-tucker conditions Dr.P. Vasanthi, Dr.S. Ranganayaki, R. Kasthuri International Journal of Engineering and Advanced Technology, 2019 The work involves purchase inventory model with shortages under fuzzy environment. An EOQ model is formulated in which the input parameters like order cost, demand rate, carrying cost and penalty cost and the decision variables like the maximum invsentory level and the lot size are fuzzified using triangular fuzzy membership function. An optimum solution of the model is arrived by using Kuhn-Tucker conditions. The crisp values of the proposed model is obtained by defuzzifying the assumed model using Graded mean Integration (GMI) method. Finally the solutions are tabulated and an analsysis of the crisp and fuzzy values of the total cost has been done in this paper
