Optimizing Activity Recognition in Video Using Evolutionary Computation

Augmented the power of Genetic Algorithms and Evolutionary Computation to optimize the VGG-16 architecture to perform Video Activity Classification that boosts the accuracy scores by over 5% on the test datasets.

Incorporated Evolutionary Computation to perform model optimization and feature selection on the neural network architecture.

The results obtained clearly showcases the power of Evolutionary Computation techniques for optimizing Deep Learning Architectures. PSO based feature selection is an efficient technique and does a great job of capturing relevant features and eradicating irrelevant features. PSO and ES (Evolutionary Strategy) based model selection and hyper-parameter optimization do a good job of generating the best set of hyperparameters, which would allow the model to operate at its full efficacy.

This proves the power and efficacy of Genetic algorithms and evolutionary computations techniques to improve the accuracy and performance of neural network-based architecture.

These techniques can be used in several different problem statements like Object Detection, Image Segmentation, and Edge Detection. Feature Selection techniques involving genetic algorithms can be used for all these applications to get the most of the vast input image data. The optimization used in the proposed work can also be extended to data involving texts and relational databases to solve problems like fraud detection and sentiment analysis.