In Defense Of Nature-inspired Algorithms: A Strategic Optimization Perspective For Modern Computer Vision
The prevalence of gradient-based deep learning has reformed computer vision optimization techniques, leading to the belief that backpropagation and its variations are adequate for contemporary vision systems. That presumption is contested in this paper. This paper argue that nature-inspired algorithms (NIAs) constitute strategic outer-loop optimization frameworks within hierarchical deep learning architectures. Although gradient descent is still essential for learning differentiable