Abstract (Eng): | Modern AI models, such as ChatGPT, exhibit a dual challenge of being both computationally and memory intensive. While these models have demonstrated remarkable capabilities, deploying them in real-life scenarios often demands the ability to run them on edge devices such as mobile phones or Internet of Things (IoT) devices. Edge computing facilitates on-device processing, reducing latency and dependence on centralized servers. However, the limited hardware resources of edge devices necessitate neural network compression techniques to enable the deployment of large models.
This research emphasizes the critical importance of neural network compression, particularly for applications requiring real-time audio-based monitoring on edge devices. In the context of audio-based applications, opting for local deployment becomes increasingly crucial due to security and privacy concerns, avoiding the transmission of sensitive sound data to external servers. In order to achieve neural network compression, various methods of quantization and pruning were used.
A smaller, compressed AI model not only facilitates edge deployment but also contributes to energy efficiency, thereby aligning with the global imperative to combat climate change. This research aims to provide a holistic approach to edge deployment, considering efficiency, privacy, and environmental sustainability in the development of audio-based monitoring systems for real-time edge applications. |
---|