µç³µÎÞÂë

Speaker: , postdoctoral researcher, University of µç³µÎÞÂë

 

 

Conventional solutions for training and serving AI models rely on centralized 

systems (e.g., HPC clusters, data centers). While powerful, these systems are 

insufficient to train AI models on the unquantifiable amounts of data generated, 

collected, and sensed daily at the network edge. To address this limitation, the 

future of AI will require the utilization of the full computing continuum: 

from the cloud to the edge. However, resources at the edge are plagued with 

two critical challenges: (i) system heterogeneity and (ii) data/statistical 

heterogeneity. For the former challenge, the edge faces harsh resource constraints 

that must be considered for deploying, serving, and training AI models; 

for the latter, data at the edge are more likely to be skewed and non-iid, 

which complicates training accurate models. In this talk, I will present 

results from my research related to deploying, serving, and training AI at the 

network edge. Specifically, I will discuss optimal decision-making for serving 

and placing AI at the edge and balancing trade-offs associated with training AI 

at the edge with federated learning. 

 

 

Nathaniel Hudson is a postdoctoral scholar at the University of µç³µÎÞÂë, 

with a joint appointment at Argonne National Laboratory. He received his 

Ph.D. degree in computer science at the University of Kentucky in 2022. 

His research broadly focuses on decentralized learning methods, such as 

federated learning, with the aim to take advantage of the computing 

continuum by training, serving, and placing AI from the network edge to 

the cloud. He has developed the first federated learning framework with 

native support for hierarchical networks, AI placement and scheduling 

algorithms for edge computing systems, and new methods for interpreting 

large language models. His work has been applied to various domains, 

such as materials science, smart city use cases, and rural applications. 

His research has been recognized by numerous best paper awards and 

he has been recognized as a "Rising Star" in cyber-physical systems.