In this work, we present SilentSense, a framework to authenticate users silently and transparently by exploiting the user touch behavior biometrics leveraging integrated sensors capture micro-movement of device caused user's screen-touch actions. By tracking fine-detailed actions user, build "touch-based biometrics" model owner extracting some principle features, then verify whether current is or guest/attacker. When using smartphone, unique operating pattern detected learnt collecting...

Much research has been conducted to securely outsource multiple parties' data aggregation an untrusted aggregator without disclosing each individual's privately owned data, or enable parties jointly aggregate their while preserving privacy. However, those works either require secure pair-wise communication channels suffer from high complexity. In this paper, we consider how external can learn some algebraic statistics (e.g., sum, product) over participants' the We assume all are subject...

Many cloud platforms emerge to meet urgent requirements for large-volume personal image store, sharing and search. Though most would agree that images contain rich sensitive information (e.g., people, location event) people's privacy concerns hinder their participation into untrusted services, today's provide little support protection. Facing large-scale from multiple users, it is extremely challenging the maintain index structure schedule parallel computation without learning anything about...

Recent studies show that WiFi interference has been a major problem for low power urban sensing technology ZigBee networks. Existing approaches dealing with such interferences often modify either the nodes or nodes. However, massive deployment of and uncooperative users call innovative cross-technology coexistence without intervening legacy systems. In this work we investigate when is interested signal.Mitigating short duration (called flash) in long data smog) challenging, especially cannot...

Facing a large number of personal photos and limited resource mobile devices, cloud plays an important role in photo storing, sharing searching. Meanwhile, some recent reputation damage stalk events caused by leakage increase people's concern about privacy. Though most would agree that search function privacy are both valuable, few system supports them simultaneously. The center such ideal is privacy-preserving outsourced image similarity measurement, which extremely challenging when the...

With the increased popularity of smartphones, various security threats and privacy leakages targeting them are discovered investigated. In this work, we present SilentSense, a framework to authenticate users silently transparently by exploiting dynamics mined from user touch behavior biometrics micro-movement device caused user's screen-touch actions. We build “touch-based biometrics” model owner extracting some principle features, then verify whether current is or guest/attacker. When using...

Vertical Federated Learning (VFL) enables multiple data owners, each holding a different subset of features about largely overlapping sets sample(s), to jointly train useful global model. Feature selection (FS) is important VFL. It still an open research problem as existing FS works designed for VFL either assumes prior knowledge on the number noisy or post-training threshold be selected, making them unsuitable practical applications. To bridge this gap, we propose Stochastic Dual-Gate based...

Many proximity-based mobile social networks are developed to facilitate connections between any two people, or help a user find people with matched profile within certain distance. A challenging task in these applications is protect the privacy of participants' profiles and personal interests. In this paper, we design novel mechanisms, when given preference-profile submitted by user, that search persons matching-profile decentralized multi-hop networks. Our mechanisms also establish secure...

Many proximity-based mobile social networks are developed to facilitate connections between any two people, or help a user find people with matched profile within certain distance. A challenging task in these applications is protect the privacy of participants' profiles and communications. In this paper, we design novel mechanisms, when given preference-profile submitted by user, that search persons matching-profile decentralized networks. Meanwhile, our mechanisms establish secure...

This paper presents a localization model employing convolutional neural network (CNN) and Gaussian process regression (GPR) based on Wi-Fi received signal strength indication (RSSI) fingerprinting data. In the proposed scheme, CNN is trained by training dataset. The adapts to complex scenes with multipath effects or many access points (APs). More specifically, pre-processing algorithm makes RSSI vector which formed considerable values from different APs readable algorithm. improves...

Federated learning allows large amounts of mobile clients to jointly construct a global model without sending their private data central server. A fundamental issue in this framework is the susceptibility erroneous training data. This problem especially challenging due invisibility clients' local and process, as well resource constraints. In paper, we aim solve by introducing first FL debugging framework, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Federated Learning (FL) has emerged as a privacy-preserving distributed machine learning paradigm. To motivate data owners to contribute towards FL, research on FL incentive mechanisms is gaining great interest. Existing monetary generally share the same model with all participants regardless of their contributions. Such an assumption can be unfair who contributed more and promote undesirable free-riding, especially when final utility value participants. In this paper, we propose...

Received Signal Strength (RSS) maps provide fundamental information for mobile users, aiding the development of conflict graph and improving communication quality to cope with complex unstable wireless channels. In this paper, we present CARM: a scheme that exploits crowd-sensing construct outdoor RSS using smartphone measurements. An alternative yet impractical approach in literature is appeal professionals customized devices. Our work distinguishes itself from previous studies by...

Federated learning (FL) enables large amounts of participants to construct a global model, while storing training data privately at each client device. A fundamental issue in this framework is the susceptibility erroneous data. This problem especially challenging due invisibility clients' local and process, as well resource constraints number mobile edge devices. In paper, we try tackle by introducing first FL debugging framework, FLDebugger, for mitigating test error caused The pro-posed...

Federated Learning (FL) enables multiple partici-pants to collaboratively train a model in privacy-preserving way. The performance of the FL heavily depends on quality participants' local data, which makes measuring contributions participants an essential task for various purposes, e.g., participant selection and reward allocation. Shapley value is widely adopted by previous work contribution assessment, which, however, requires repeatedly leave-one-out retraining thus incurs prohibitive...

The existing work on distributed secure multi-party computation, e.g., set operations, dot product, ranking, focus the privacy protection aspects, while verifiability of user inputs and outcomes are neglected. Most works assume that involved parties will follow protocol honestly. In practice, a malicious adversary can easily forge his/her input values to achieve incorrect or simply lie about computation results cheat other parities. this work, we problem verifiable preserving multiparty...

It is necessary to improve the performance of object detection algorithm in resource-constrained embedded devices by lightweight improvement. In order further recognition accuracy for small target objects, this paper integrates 5 × deep detachable convolution kernel on basis MobileNetV2-SSDLite model, extracts features two special convolutional layers addition detecting target, and designs a new network—Lightweight Microscopic Detection Network (LMS-DN). The network can be implemented such...

Mobile-centric AI applications put forward high requirements for resource-efficiency of model inference. Input filtering is a promising approach to eliminate the redundancy in input so as reduce cost Previous efforts have tailored effective solutions many applications, but left two essential questions unanswered: (1) theoretical filterability an inference workload guide application techniques, thereby avoiding trial-and-error resource-constrained mobile applications; (2) robust...

In this work, we design and develop Montage for real-time multi-user formation tracking localization by off-the-shelf smartphones. achieves submeter-level accuracy integrating temporal spatial constraints from user movement vectorestimation distance measuring. Montage, designed a suite of novel techniques to surmount variety challenges in tracking, without infrastructure fingerprints, any priori user-specific (e.g., stride-length phoneplacement) or site-specific digitalized map) knowledge:...

ABSTRACTEye-controlled human-computer interaction (ECHCI) attracts attention for its human-centered, natural and direct operation characteristics. The most common ECHCI trigger motion is "gazing," which causes Midas Touch problems, lowering usability experience. This paper innovatively proposed using combinations based on blinking as the of interactive objects (IOs). Trigger motions IO design parameters were explored platform consisting Tobii eye-tracker computer vision kits. In experiment,...