Amazing technological breakthrough possible @S-Logix

Office Address

  • #5, First Floor, 4th Street Dr. Subbarayan Nagar Kodambakkam, Chennai-600 024 Landmark : Samiyar Madam
  • +91- 81240 01111

Social List

Research Topics on Explainable Deep Neural Networks

Research Topics on Explainable Deep Neural Networks

   Deep neural networks achieved an extraordinary performance in various difficult tasks and produced accurate results in different application domains such as speech recognition, image recognition, language translation, and other complex problems. Some of the reprovals towards deep learning are requirements for a huge amount of labeled data, computational resources, more training time, high power and energy concerns. Explainable deep neural networks(xDNN) is one of the approaches to handle and control the reprovals of deep neural networks. xDNN provides a high level of explainability and high accuracy. xDNN is noniterative and nonparametric.

   xDNN outperforms other deep learning architectures in terms of training time, computational resources, and produces an explainable classifier. xDNN is a new deep learning architecture that utilizes a feed-forward neural network as a classifier to form several layers in a very clear user understandable manner. xDNN offers computationally very efficient implementation and is clearly understandable to users. Some of the explainable deep learning methods are attribution-based, non-attribution-based methods, and uncertainty quantification. Future advancements in xDNN are tree-based architecture, synthetic data generation, and local optimization.

   • Explainable deep neural networks (xDNN) have demonstrated the ability to outperform the existing deep learning methods by offering an explainable classifier with little computational resources and less training time than the other models.

   • xDNN significance addresses the deficiencies of conventional deep neural networks involving the lack of explainability, computational burden, and interpretability, as it is difficult to analyze which modalities or features are driving the predictions.

   • xDNN offers a new deep learning architecture that combines reasoning and learning in a synergy.

   • The goal of attribution-based methods is to determine the contribution of each input feature to the target output. It covers most visualization methods in computer vision, which explain directly in the domain of input images by localizing regions that contribute most to the decision.

   • Non-attribution-based methods explain concepts, training data, and intrinsic attention mechanisms.

   • xDNN is highly parallelizable and suitable for evolving forms of applications. The most recent application of explainable deep learning is efficient and robust pattern recognition.