Related Experiment Video
Updated: May 9, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
8.9K
HybridBranchNetV2: Towards reliable artificial intelligence in image classification using reinforcement learning
Ebrahim Parcham1, Mansoor Fateh1, Vahid Abolghasemi2
1Faculty of Computer Engineering, Shahrood University of Technology, Shahrood, Iran.
Plos One
|February 10, 2025
Summary
HybridBranchNetV2 enhances artificial intelligence (AI) adaptability in dynamic environments. This novel hybrid architecture integrates reinforcement learning and graph-based methods for improved object recognition and classification accuracy.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Current artificial intelligence (AI) algorithms exhibit limitations in adapting to dynamic real-world scenarios.
- Challenges include complex classification tasks and object relationship extraction due to non-adaptive behaviors.
Purpose of the Study:
- Introduce HybridBranchNetV2, an optimized hybrid architecture to enhance AI adaptability.
- Address limitations in current AI models for dynamic environments.
Main Methods:
- Integrate reinforcement learning for adaptive feature extraction.
- Employ graph-based techniques for analyzing object relationships in complex environments.
- Dynamically adjust feature extraction based on environmental feedback.
Main Results:
- Achieved an average accuracy of 91.75% across four challenging datasets.
- Demonstrated significant improvements: 14% on Visual Genome and ImageNet 1K, 6% on CIFAR and ImageNet, 1% on Flowers.
- Enhanced classification accuracy and computational efficiency.
Conclusions:
- HybridBranchNetV2 offers superior adaptability and performance in complex AI tasks.
- The model is suitable for real-time applications with reduced risk of overfitting.
- The framework significantly improves adaptability, performance, and computational efficiency.
