The increasing of pre-trained models has significantly facilitated the performance on limited data tasks with transfer learning. However, progress learning mainly focuses optimizing weights models, which ignores structure mismatch between model and target task. This paper aims to improve from another angle - in addition tuning weights, we tune order better match To this end, propose TransTailor, targeting at pruning for improved Different traditional pipelines, prune fine-tune according...

Modern Federated Learning (FL) has become increasingly essential for handling highly heterogeneous mobile devices. Current approaches adopt a partial model aggregation paradigm that leads to sub-optimal accuracy and higher training overhead. In this paper, we challenge the prevailing notion of partial-model propose novel "full-weight aggregation" method named Moss, which aggregates all weights within models preserve comprehensive knowledge. Evaluation across various applications demonstrates...

Cognitive radio networks allow opportunistic spectrum access and can significantly improve spectral efficiency. To achieve higher sensing accuracy, cognitive systems often require cooperation among secondary users. One of the most important aspects in collaborative is data fusion algorithm which combines results from users to produce final channel status hypothesis. However, plenty factors may affect performance certain rule, for example, individual node's number involved nodes, like. If...

Harvesting energy from ambient vibrations with piezoelectric transducers is an alternative solution for the surging needs of self-powered devices, such as IoT condition and structural monitoring or biomedical implants. A limiting factor transducer narrow bandwidth due to high mechanical quality factor. Therefore, output power drops significantly when excitation frequency deviates resonant in a real environment. One introduce time delays into active harvesting-interface concepts Synchronous...

Crowdsourcing Federated learning (CFL) is a new crowdsourcing development paradigm for the Deep Neural Network (DNN) models, also called "software 2.0". In practice, privacy of CFL can be compromised by many attacks, such as free-rider adversarial gradient leakage and inference attacks. Conventional defensive techniques have low efficiency because they deploy heavy encryption or rely on Trusted Execution Environments (TEEs). To improve protecting from these this paper proposes FedSlice to...

Federated learning (FL) has emerged as an effective solution to decentralized and privacy-preserving machine for mobile clients. While traditional FL demonstrated its superiority, it ignores the non-iid (independently identically distributed) situation, which widely exists in scenarios. Failing handle situations could cause problems such performance decreasing possible attacks. Previous studies focus on "symptoms" directly, they try improve accuracy or detect attacks by adding extra steps...

Providing machine learning services is becoming profit business for IT companies. It estimated that the AI-related will bring trillions of dollars to global economy. When selling services, companies should consider two important aspects: security DNN model and inference latency. The models are expensive train represent precious intellectual property. latency because modern usually deployed time-sensitive tasks affects user's experience. Existing solutions cannot achieve a good balance...

This paper presents a piezoelectric energy harvesting interface circuit with technique to broaden the bandwidth by introducing two time delays into conventional synchronous-electrical-charge-extraction (SECE) scheme. The adaptively adjusts these for maximum power output at off-resonant frequencies. is increased 71% while consumption of itself low level 0.85μW. chip fabricated in 0.35μm CMOS process technology.

One of the most crucial functionalities cognitive radio networks is spectrum sensing. Completing this task in an accurate manner requires opportunistic access. Traditionally, sensing has been performed through energy detection by each individual secondary user. In order to increase accuracy, measurements are aggregated using different fusion functions. However, even though collaborative can accuracy under benign settings, it prone falsification attacks, where malicious users report fake...

The rapid digitization of various services has led to the emergence super apps, providing an array utilities under one application. A common usage scenario such platforms, as Alipay, involves shared accounts by multiple users, typically within a family. However, this use poses unique challenge security protocols designed prevent unauthorized access, they can misinterpret legitimate multi-user behavior fraudulent activity. In paper, we explore complexities involved in accurately discerning...

This paper presents a novel system design of an adaptive time-controlled energy harvesting interface circuit for broadening the bandwidth piezoelectric transducers. The proposed monitors excitation frequency and delays switching activity accordingly to achieve optimal impedance matching. theory calculating time delay values are described in detail. Simulation result shows that is able generate nearly as much power theoretical limitation over large range frequency, increases from 3.7 Hz more...

Federated learning (FL) has emerged as an effective solution to decentralized and privacy-preserving machine for mobile clients. While traditional FL demonstrated its superiority, it ignores the non-iid (independently identically distributed) situation, which widely exists in scenarios. Failing handle situations could cause problems such performance decreasing possible attacks. Previous studies focus on "symptoms" directly, they try improve accuracy or detect attacks by adding extra steps...

In the area of performance analysis wireless networks, one critical issue is hidden terminal problem, which considered as severest reasons for degradation network performance. this paper, we incorporate reactive jamming scheme with distributed coordination function (DCF) in IEEE 802.11 based local networks (WLANs) to improve throughput presence terminals. proposed protocol, schedule access point (AP) broadcast signal reactively hinder simultaneous transmission Both analytical and numerical...

Fine-tuning is a typical mechanism to achieve model adaptation for mobile/web users, where trained by the cloud further retrained fit target user task. While traditional fine-tuning has been proved effective, it only utilizes local data adaptation, failing take advantage of valuable knowledge from other users. In this paper, we attempt extend local-user multi-user fed-tuning with help Federated Learning (FL). Following new paradigm, propose EEFT, framework aiming Efficient and Effective...

Indoor localization is important to many location based applications and services. Many indoor methods have been proposed they can be roughly categorized into two groups: One on the distance estimation between a target point Access Points (APs); other Received Signal Strength (RSS) fingerprint map. However, approaches all assume that locations of (APs) are known beforehand. In this paper, we consider scenario APs not known, use measured RSS (Received Strength) at some location-known...

In the context of hepatocellular carcinoma (HCC) treatment, this study introduces a biomimetic "bomb" approach to target HCC by utilizing macrophages (MAs) as ultrasound-responsive carriers. This advanced drug delivery system (MAs-DOX/PFP/Ms, MAs-DPM) encapsulates doxorubicin (DOX) and perfluoropentane (PFP) is designed precisely HCC. MAs-DPM exhibits excellent loading capacity biocompatibility, can effectively release DOX enhance ultrasound imaging capabilities under irradiation (UI)....

Trusted Execution Environments (TEE) are used to safeguard on-device models. However, directly employing TEEs secure the entire DNN model is challenging due limited computational speed. Utilizing GPU can accelerate DNN's computation speed but commercial widely-available GPUs usually lack security protection. To this end, scholars introduce TSDP, a method that protects privacy-sensitive weights within and offloads insensitive GPUs. Nevertheless, current methods do not consider presence of...

Trusted Execution Environments (TEE) are used to safeguard on-device models. However, directly employing TEEs secure the entire DNN model is challenging due limited computational speed. Utilizing GPU can accelerate DNN's computation speed but commercial widely-available GPUs usually lack security protection. To this end, scholars introduce TEE-shielded partition (TSDP), a method that protects privacy-sensitive weights within and offloads insensitive GPUs. Nevertheless, current methods do not...

The increasing of pre-trained models has significantly facilitated the performance on limited data tasks with transfer learning. However, progress learning mainly focuses optimizing weights models, which ignores structure mismatch between model and target task. This paper aims to improve from another angle - in addition tuning weights, we tune order better match To this end, propose TransTailor, targeting at pruning for improved Different traditional pipelines, prune fine-tune according...

In this paper, we proposed a method to fuse data from radar and IR sensor in Extend Kalman probability hypothesis density (EK-GMPHD) filter. Firstly the multi-target is estimated with infrared (IR) using EK-GMPHD filter, then filtering results are fused measurements through sequential way, state updated at tracking system. Under false alarms, missed detections dense targets environment, has high reliability when multi-target. Simulation experiments presented demonstrate performance of method.

Background: Normal karyotype acute myeloid leukemia (NK AML) currently does not have a universal measurable residual disease (MRD) target, thus making next generation sequencing (NGS) more attractive as an MRD monitoring tool. Also, there is limited knowledge on selecting consolidation therapy based the status at time of first achievement complete remission (CR1) in NK AML between allogeneic hematopoietic cell transplantation (allo‐HCT) and chemotherapy. Aims: Current study aims to...

Abstract Background: Although some improvements in the management of pancreatic cancer (PC) have been made, no major breakthroughs terms biomarker discovery or effective treatment emerged. Here, we applied artificial intelligence (AI)-based methods to develop a model diagnose PC and predict survival outcome. Methods: Multiple bioinformatics methods, including RankProd, were performed identify differentially expressed genes (DEGs) PC. A Back Propagation (BP) was constructed, followed by...

Summary Majority of the deep learning techniques in seismic image analysis focus on solving one task at a time and ignore richness presence many other structures vicinity their correlation with interest hand. These approaches work best identification simple shallow areas survey where signal-to-noise ratio is high struggle deeper as signal becomes weaker. In addition, it challenge to acquire right data quality labels train models for some fundamental challenges geoscience. this paper, we...