arXiv:2504.17929v1 Announce Type: new
Abstract: Explainable artificial intelligence (XAI) enhances AI system transparency by framing interpretability as an optimization problem. However, this approach often necessitates numerous iterations of computationally intensive operations, limiting its applicability in real-time scenarios. While recent research has focused on XAI hardware acceleration on FPGAs and TPU, these methods do not fully address energy efficiency in real-time settings. To address this limitation, we propose XAIedge, a novel framework that leverages approximate computing techniques into XAI algorithms, including integrated gradients, model distillation, and Shapley analysis. XAIedge translates these algorithms into approximate matrix computations and exploits the synergy between convolution, Fourier transform, and approximate computing paradigms. This approach enables efficient hardware acceleration on TPU-based edge devices, facilitating faster real-time outcome interpretations. Our comprehensive evaluation demonstrates that XAIedge achieves a $2times$ improvement in energy efficiency compared to existing accurate XAI hardware acceleration techniques while maintaining comparable accuracy. These results highlight the potential of XAIedge to significantly advance the deployment of explainable AI in energy-constrained real-time applications.
Abstract: The concept of explainable artificial intelligence (XAI) has gained significant attention in recent years. XAI aims to enhance the transparency of AI systems by providing interpretability and insight into their decision-making processes. However, the existing approach to XAI often involves computationally intensive operations, making it challenging to apply in real-time scenarios.
In this article, the authors propose XAIedge, a novel framework that addresses the limitation of existing XAI methods by incorporating approximate computing techniques. By translating XAI algorithms, such as integrated gradients, model distillation, and Shapley analysis, into approximate matrix computations, XAIedge achieves efficient hardware acceleration on edge devices powered by Tensor Processing Units (TPUs).
The authors highlight the multi-disciplinary nature of their approach, which combines concepts from XAI, hardware acceleration, and approximate computing paradigms. By leveraging the synergy between convolution, Fourier transform, and approximate computing, XAIedge achieves faster real-time outcome interpretations while maintaining comparable accuracy to existing XAI hardware acceleration techniques.
The article emphasizes the significance of energy efficiency in real-time settings, where energy-constrained applications demand optimal resource utilization. XAIedge addresses this concern by introducing approximate computing techniques that result in a times$ improvement in energy efficiency compared to accurate XAI hardware acceleration techniques. This improvement opens up opportunities for the deployment of explainable AI in energy-constrained real-time applications.
Overall, XAIedge presents a promising solution to the challenges faced in deploying XAI in real-time scenarios. By incorporating approximate computing techniques and leveraging the power of TPUs, XAIedge not only enhances the interpretability of AI systems but also addresses the energy efficiency requirements of resource-constrained applications. The multi-disciplinary nature of XAIedge showcases the potential for collaboration between different fields to advance the development and deployment of AI technologies.