Kennesaw State researcher brings artificial intelligence to everyday devices with new NSF grant

Kennesaw State Researcher Advances AI for Everyday Devices

Artificial intelligence (AI) is often associated with large data centers and supercomputers. However, a researcher at Kennesaw State University is working to make AI more accessible by embedding it directly into personal devices. This effort is supported by a National Science Foundation (NSF) grant aimed at shifting AI processing away from the cloud and onto mobile phones, drones, and other smaller systems.

The project focuses on enabling AI to operate without relying on an internet connection, which is uncommon in current AI applications. Running AI locally on devices reduces dependency on expensive infrastructure, potentially speeding up access and improving privacy and reliability.

Activation Sparsity: A New Approach to Efficiency

The core of this research is a technique called activation sparsity. Unlike traditional models that activate all neurons in an AI system, activation sparsity selectively activates only a subset of neurons at a time. This means the system predicts which parts of the model it needs in advance and loads only those, reducing memory usage and energy consumption.

Existing methods often shrink AI models by lowering data precision or pruning less important parameters. This research takes a different path by focusing on predicting which neurons will be active, allowing it to be combined with other optimization methods like pruning and quantization for even greater efficiency.

Practical Applications and Open-Source Tools

The research will test tiny machine-learning models that serve as predictors to support larger AI systems. This balance makes AI feasible on constrained devices such as industrial sensors and drones. An open-source simulator is also being developed to help other researchers and students experiment with and improve the technology.

Embedding AI in smaller devices has clear industrial applications, including monitoring factory robots and predicting equipment failures. These use cases illustrate how local AI can enhance operational efficiency and reliability in various sectors.

University Support and Industry Impact

The work is conducted at the Sustainable Smart System Lab on KSU’s Marietta Campus. The university provides crucial support through lab space, administrative assistance, and graduate research help. This environment fosters progress and collaboration on applied AI research.

Interim Dean of the College of Computing and Software Engineering, Yiming Ji, highlighted the significance of this work for KSU’s mission. The research not only tackles technical challenges but also creates opportunities for students and industry partners to engage with AI innovations.

• Running AI on-device reduces latency and dependence on cloud infrastructure.
• Activation sparsity offers a new way to improve AI model efficiency.
• Open-source tools will accelerate research and education in AI hardware optimization.
• Potential applications span manufacturing, robotics, and more.

For professionals interested in AI deployment on edge devices and efficient model design, this project offers a promising direction. Further information on AI training and courses can be found at Complete AI Training.