Modern Neural Network Architecture: GNN & ZNN
Guest Lecture @ SRM TRP Engineering College
The Department of Computer Science and Engineering and Department of Artificial Intelligence & Data Science, in association with Office of International Affairs proudly organize a “Modern Neural Network Architecture: GNN & ZNN”
Date: 5th February 2026
Time: 10:30 AM – 12:30 PM and 02:00 PM – 04:00 PM
Venue: Fifth Floor, Seminar Hall
Resource Person:
Dr. Predrag S. Stanimirović
Professor, Faculty of Science and Mathematics
Department of Computer Science
University of Nis, Serbia
Key Takeaways:
- Importance of Matrix Theory in Modern Computing
Matrix algebra forms the mathematical backbone of signal processing, control systems, machine learning, and neural networks.
Concepts like rank, inverse, eigenvalues, norms, and trace are fundamental for both theoretical analysis and practical algorithms
- Symbolic Linear Algebra Enhances Analytical Understanding
Symbolic computation allows exact solutions instead of numerical approximations.
Operations such as row reduction, rank determination, eigenvalues, eigenvectors, and norms can be handled symbolically.
Useful for verifying theoretical properties before numerical implementation
- Gradient Neural Networks (GNN)
GNNs use gradient descent on an energy (error) function based on Frobenius norm.
Suitable for time-invariant (constant) matrix problems.
Convergence depends on:
Choice of learning rate (γ)
Initial conditions
Effective for solving linear matrix equations and least squares problems
- Zeroing Neural Networks (ZNN)
ZNNs are designed specifically for time-varying problems.
Core idea: force the error function to zero exponentially using differential equations.
Guarantees global exponential convergence under suitable activation functions.
More efficient than GNNs for real-time and online applications
Who can attend: Faculty members and students of CSE & AI&DS
Let’s explore how next-generation AI technologies like modern Neural Network are transforming real time and online applications
