TY - GEN
T1 - SplitGP
T2 - 42nd IEEE International Conference on Computer Communications, INFOCOM 2023
AU - Han, Dong Jun
AU - Kim, Do Yeon
AU - Choi, Minseok
AU - Brinton, Christopher G.
AU - Moon, Jaekyun
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - A fundamental challenge to providing edge-AI services is the need for a machine learning (ML) model that achieves personalization (i.e., to individual clients) and generalization (i.e., to unseen data) properties concurrently. Existing techniques in federated learning (FL) have encountered a steep tradeoff between these objectives and impose large computational requirements on edge devices during training and inference. In this paper, we propose SplitGP, a new split learning solution that can simultaneously capture generalization and personalization capabilities for efficient inference across resource-constrained clients (e.g., mobile/IoT devices). Our key idea is to split the full ML model into client-side and server-side components, and impose different roles to them: the client-side model is trained to have strong personalization capability optimized to each client's main task, while the server-side model is trained to have strong generalization capability for handling all clients' out-of-distribution tasks. We analytically characterize the convergence behavior of SplitGP, revealing that all client models approach stationary points asymptotically. Further, we analyze the inference time in SplitGP and provide bounds for determining model split ratios. Experimental results show that SplitGP outperforms existing baselines by wide margins in inference time and test accuracy for varying amounts of out-of-distribution samples.
AB - A fundamental challenge to providing edge-AI services is the need for a machine learning (ML) model that achieves personalization (i.e., to individual clients) and generalization (i.e., to unseen data) properties concurrently. Existing techniques in federated learning (FL) have encountered a steep tradeoff between these objectives and impose large computational requirements on edge devices during training and inference. In this paper, we propose SplitGP, a new split learning solution that can simultaneously capture generalization and personalization capabilities for efficient inference across resource-constrained clients (e.g., mobile/IoT devices). Our key idea is to split the full ML model into client-side and server-side components, and impose different roles to them: the client-side model is trained to have strong personalization capability optimized to each client's main task, while the server-side model is trained to have strong generalization capability for handling all clients' out-of-distribution tasks. We analytically characterize the convergence behavior of SplitGP, revealing that all client models approach stationary points asymptotically. Further, we analyze the inference time in SplitGP and provide bounds for determining model split ratios. Experimental results show that SplitGP outperforms existing baselines by wide margins in inference time and test accuracy for varying amounts of out-of-distribution samples.
UR - http://www.scopus.com/inward/record.url?scp=85170508346&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85170508346&partnerID=8YFLogxK
U2 - 10.1109/INFOCOM53939.2023.10229027
DO - 10.1109/INFOCOM53939.2023.10229027
M3 - Conference contribution
AN - SCOPUS:85170508346
T3 - Proceedings - IEEE INFOCOM
BT - INFOCOM 2023 - IEEE Conference on Computer Communications
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 17 May 2023 through 20 May 2023
ER -