Personalized On-Device E-Health Analytics With Decentralized Block Coordinate Descent

Ye, Guanhua; Yin, Hongzhi; Chen, Tong; Xu, Miao; Nguyen, Quoc Viet Hung; Song, Jiangning

doi:10.1109/jbhi.2022.3140455

Cited by 8 publications

(8 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…✓ Supported × Not supported or unavailable 1 In comparison with retraining the model from its initial state. 2 All remaining clients possess a usable model (ensemble) before unlearning finishes. the knowledge of the quitting client has blended into all client models in the learning process, these frameworks have to execute a complicated inexact unlearning procedure (FedUnl) or retrain all models from the ground up (DSGD).…”

Section: Iic Unlearning Effectivenessmentioning

confidence: 99%

“…The surge of edge computing and big data brings people personalized services in various domains like product recommendation 1 and personalized healthcare analysis 2 . In those services, users' edge devices (e.g., smartphones and smartwatches) play an important role in collecting user data and generating analytical feedback [3][4][5] , while providing a security and timeliness advantage compared with the outgoing centralized services.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Heterogeneous Decentralized Machine Unlearning with Seed Model Distillation

Chen

Nguyen

et al. 2023

Preprint

View full text Add to dashboard Cite

As some recent information security legislation endowed users with unconditional rights to be forgotten by any trained machine learning model, personalized service providers must put unlearning functionality into their consideration. The most straightforward method to unlearn users’ contribution is to retrain the model from the initial state, which is not realistic in high throughput applications with frequent unlearning requests. Though some machine unlearning frameworks have been proposed to speed up the retraining process, they fail to match decentralized learning scenarios. In this paper, we design a decentralized unlearning framework called HDUS, which uses distilled seed models to construct erasable ensembles in all clients. Moreover, the framework is compatible with heterogeneous on-device models, representing stronger scalability in real-world applications. Extensive experiments on three real-world datasets show that our HDUS achieves state-of-the-art performance.

show abstract

Section: Iic Unlearning Effectivenessmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Heterogeneous Decentralized Machine Unlearning with Seed Model Distillation

Chen

Nguyen

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Such methods also work for the decentralized on-device learning framework (He et al, 2018;Koloskova et al, 2019;. However, these frameworks either reveal the sensitive user data (Ye et al, 2022) or uncover the model parameters (He et al, 2018).…”

Section: Neighbors In Collaborationmentioning

confidence: 99%

“…In this centralized setting, the user devices are only used for transmitting locally collected data to the server and displaying could-generated results. Consequently, the only way to obtain analytic results is to wait for the output from the global model, which can be devastating when monitoring fatal diseases under network delay or connection interruptions (Ye, Yin, Chen, Xu, Nguyen, and Song, 2022). Additionally, since the central server stores overall personal user data, the security of sensitive information and the reliability of the learned models are highly vulnerable to adversarial attacks (e.g., attribute inference attacks (Mosallanezhad, Beigi, and Liu, 2019;Zhang, Yin, Chen, Huang, Nguyen, and Cui, 2022) and false data injection (Koh and Liang, 2017)).…”

Section: Introductionmentioning

confidence: 99%

Heterogeneous Collaborative Learning for Personalized Healthcare Analytics via Messenger Distillation

Ye¹,

Chen²,

Li³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Personalized predictive analytics are changing the way people live, such as recommendation, ehealth, and chatbots. With the prominent development of edge devices, predictive models based on deep neural networks (DNNs) can be conveniently hosted on-device to deliver accurate results, which also provides better privacy and timeliness. Due to the inevitable data sparsity on a single device, collaborative learning (e.g., federated learning) is regarded as an integral approach for learning performant personalized deep models. However, traditional decentralized learning protocols communicate the knowledge by sharing the model parameters (i.e., weights or gradients). Consequently, such methods strictly require homogeneity of all participant models, thus being unable to handle applications where heterogeneous deep models with different architectures are built for the same task but adapt to a variety of device capacities. Furthermore, to support model personalization, a common practice is to pick a curated subset of similar users (i.e., neighbors) for knowledge aggregation, but the neighbor selection process potentially threatens user privacy with the need for exchanging personal information or model parameters that are highly sensitive.In this paper, we propose a Similarity-based Decentralized Knowledge Distillation (SD-Dist) framework for collaboratively learning heterogeneous deep models on decentralized devices. By introducing a preloaded reference dataset, SD-Dist enables all participant devices to identify similar users and distil knowledge from them without any assumptions on a fixed model architecture. In addition, none of these operations will reveal any sensitive information like personal data and model parameters. Extensive experimental results on three real-life datasets show that SD-Dist can achieve competitive performance with less compute resources, while ensuring model heterogeneity and privacy. As revealed in our experiments, our framework also enhances the resultant models' robustness when users' data is sparse and diverse.

show abstract

“…Centralized POI recommendation highly relies on the stability of the server and internet connectivity. Once the server is impaired or overcrowded, or network quality cannot be guaranteed, the recommendation services are unable to ensure timeliness and may even go oline [37,52]. Considering many tourist attractions are located in remote areas with limited telecom infrastructures, weakening the resilience of cloud-based POI recommendation.…”

Section: Introductionmentioning

confidence: 99%

Decentralized Collaborative Learning Framework for Next POI Recommendation

Long

Chen

Hùng

et al. 2023

ACM Trans. Inf. Syst.

View full text Add to dashboard Cite

Next Point-of-Interest (POI) recommendation has become an indispensable functionality in Location-based Social Networks (LBSNs) due to its effectiveness in helping people decide the next POI to visit. However, accurate recommendation requires a vast amount of historical check-in data, thus threatening user privacy as the location-sensitive data needs to be handled by cloud servers. Although there have been several on-device frameworks for privacy-preserving POI recommendations, they are still resource-intensive when it comes to storage and computation, and show limited robustness to the high sparsity of user-POI interactions. On this basis, we propose a novel d ecentralized c ollaborative l earning framework for POI r ecommendation (DCLR), which allows users to train their personalized models locally in a collaborative manner. DCLR significantly reduces the local models’ dependence on the cloud for training, and can be used to expand arbitrary centralized recommendation models. To counteract the sparsity of on-device user data when learning each local model, we design two self-supervision signals to pretrain the POI representations on the server with geographical and categorical correlations of POIs. To facilitate collaborative learning, we innovatively propose to incorporate knowledge from either geographically or semantically similar users into each local model with attentive aggregation and mutual information maximization. The collaborative learning process makes use of communications between devices while requiring only minor engagement from the central server for identifying user groups, and is compatible with common privacy preservation mechanisms like differential privacy. We evaluate DCLR with two real-world datasets, where the results show that DCLR outperforms state-of-the-art on-device frameworks and yields competitive results compared with centralized counterparts.

show abstract

Personalized On-Device E-Health Analytics With Decentralized Block Coordinate Descent

Cited by 8 publications

References 31 publications

Heterogeneous Decentralized Machine Unlearning with Seed Model Distillation

Heterogeneous Decentralized Machine Unlearning with Seed Model Distillation

Heterogeneous Collaborative Learning for Personalized Healthcare Analytics via Messenger Distillation

Decentralized Collaborative Learning Framework for Next POI Recommendation

Contact Info

Product

Resources

About