1 August 2020 in Knowledge-Based Systems
18 November 2024 in IEEE Transactions on Consumer Electronics
With the rising demand for intelligent services and privacy protection in consumer artificial intelligence (AI), federated edge learning has emerged as a beacon for privacy-preserving distributed machine learning. This non-centralized approach offers distinct data privacy advantages, making it especially suitable for applications with strict privacy policies, such as consumer-centric healthcare systems. However, consumer electronics/devices in federated edge learning might unlawfully violate model ownership by duplicating or distributing models. Another concern is that a central server, if compromised, can become a conduit for model leakage, whether for profit motives or due to external intrusions, thus exacerbating the model ownership violation. To our knowledge, few works have focused on model ownership protection for federated edge learning in the consumer-centric healthcare domain. In this paper, we introduce a federated model ownership protection (FedMOP) approach. FedMOP is a model ownership safeguard and a formidable fortress built upon two core innovations. First, FedMOP can swiftly embed and replace watermarks during local training, allowing for the accurate identification and subsequent flagging of deceitful consumer devices. Second, FedMOP ensures that the server only accesses encrypted versions of the federated model, keeping the model a secret. Extensive evaluations demonstrate the robustness and effectiveness of FedMOP. We hope that FedMOP can light the way for surmounting similar ethical challenges in the broader realm of consumer-edge AI.
1 April 2021 in Information Sciences
11 June 2024 in IEEE Transactions on Consumer Electronics
The Consumer Internet of Things (CIoT) integrates the advantage of Internet of Things (IoT) technologies to provide convenience in consumers’ daily lives. With the rapid development of the CIoT, data collected from consumer smart devices has increased exponentially. In the CIoT, web pages, as internet information carriers, offer spammers opportunities to conduct security attacks, which could harm the CIoT systems. Inspired by this challenge, this paper introduces an intelligent feature extraction method, page2vec, and a new classification algorithm, RFiRF, to detect web spam in the CIoT. page2vec is based on a score propagation model, which calculates each page’s goodness and badness scores through the links of a web graph. It is observed that different scoring functions can produce different web page features. Based on this observation, 20 to 30 scoring functions are designed and incorporated into page2vec. This way, page2vec can automatically extract web page features from a web graph. Unfortunately, page2vec also brings a high dimensionality problem when constructing binary classifiers for spam detection. To address this problem, a new classification algorithm, called Random Forest in Random Forest (RFiRF), is proposed. RFiRF replaces a random forest’s meta-classifier (decision tree) with a random forest. It divides the sparse data space into dense sub-spaces by randomly sampling training instances and features. Experiments on two benchmark datasets show that 1) page2vec features are much more predictive than the raw link-based and node2vec features, and 2) RFiRF outperforms some classification algorithms in most cases when facing the high dimensionality problem. We hope this paper can give peers valuable insights into the CIoT security.
10 April 2024 in Journal of Information Security and Applications
5 August 2022 in Research Square
As one kind of distributed machine learning technique, federated learning enables multiple clients to build a model across decentralized datacollaboratively without explicitly aggregating the data. Due to its abilityto break data silos, federated learning has received increasing attentionin many fields, including finance, healthcare, and education. However,the invisibility of clients’ training data and the local training process result in some security issues. Recently, many works have beenproposed to research the security attacks and defenses in federatedlearning, but there has been no special survey on poisoning attacks onfederated learning and the corresponding defenses. In this paper, weinvestigate the most advanced schemes on federated learning poisoningattacks and defenses and point out the future directions in these areas.