NEURAL NETWORK OPTIMIZATION TECHNIQUES FOR REAL-TIME AUTONOMOUS SYSTEMS
Keywords:
Neural Network Optimization; Real-Time Systems; Autonomous Vehicles; Model Compression; Edge AI; Quantization; Pruning; Robotics; Deep Learning; Inference AccelerationAbstract
Real-time autonomous systems—including autonomous vehicles, drones, and robotic manipulators—depend on neural networks capable of performing inference within strict latency, energy, and reliability constraints. As models grow more complex, the challenge lies in optimizing computation without compromising accuracy or safety. This paper investigates modern optimization techniques such as model pruning, quantization, knowledge distillation, edge-efficient architectures, and hardware–software co-optimization. Using case studies and performance evaluations from recent publications, we highlight how various optimizations reduce inference latency by up to 85% and energy consumption by up to 70%. The results demonstrate that intelligently optimized neural models enable safe, efficient, and scalable deployment of real-time autonomous systems.
