Completion Date
Spring 4-14-2024
Document Type
Thesis
Degree Name
Master of Science (MS)
Program or Discipline Name
Computer and Information Sciences
Secondary Program or Discipline Name
Cybersecurity Operations and Control Management
First Advisor
Sangwhan Cha
Second Advisor
Majid Shalan
Abstract
In the rapidly evolving field of artificial intelligence (AI), deep learning models' interpretability
and reliability are severely hindered by their complexity and opacity. Enhancing the
transparency and interpretability of AI systems for humans is the primary objective of the
emerging field of explainable AI (XAI). The attention mechanisms at the heart of XAI's work
are based on human cognitive processes. Neural networks can now dynamically focus on
relevant parts of the input data thanks to these mechanisms, which enhances interpretability
and performance. This report covers in-depth talks of attention mechanisms in neural networks
within XAI, as well as an analysis of the theoretical foundations, architectural applications, and
empirical evidence showing how well they work to improve model transparency. The report
provides a comprehensive analysis of the role of attention mechanisms in AI models to address
ethical concerns, comply with regulatory requirements, and foster a deeper understanding and
trust in AI systems. The report contributes to the discussion about bringing AI closer to human
values and cognitive processes so that its advancements are impactful and responsible by
conducting a thorough analysis.
Recommended Citation
Kotipalli, B. (2024). The Role of Attention Mechanisms in Enhancing Transparency and Interpretability of Neural Network Models in Explainable AI. Retrieved from https://digitalcommons.harrisburgu.edu/dandt/2