Federated learning has emerged as an effective paradigm to achieve privacy-preserving collaborative among different parties. Compared traditional centralized that requires collecting data from each party, in federated learning, only the locally trained models or computed gradients are exchanged, without exposing any information. As a result, it is able protect privacy some extent. In recent years, become more and prevalent there have been many surveys for summarizing related methods this hot...

Deep learning models have been deployed to a wide range of edge devices. Since the data distribution on devices may differ from cloud where model was trained, it is typically desirable customize for each device improve accuracy. However, such customization hard because collecting usually prohibited due privacy concerns. In this paper, we propose PMC, privacy-preserving framework effectively CNN without raw data. Instead, introduce method extract statistical information edge, which contains...

Federated learning (FL) has become a prevalent distributed machine paradigm with improved privacy. After learning, the resulting federated model should be further personalized to each different client. While several methods have been proposed achieve personalization, they are typically limited single local device, which may incur bias or overfitting since data in device is extremely limited. In this paper, we attempt realize personalization beyond The motivation that during FL process, there...

Federated learning has emerged as an effective paradigm to achieve privacy-preserving collaborative among different parties. Compared traditional centralized that requires collecting data from each party, in federated learning, only the locally trained models or computed gradients are exchanged, without exposing any information. As a result, it is able protect privacy some extent. In recent years, become more and prevalent there have been many surveys for summarizing related methods this hot...

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...

The knowledge of a deep learning model may be transferred to student model, leading intellectual property infringement or vulnerability propagation. Detecting such reuse is nontrivial because the suspect models not white-box accessible and/or serve different tasks. In this paper, we propose ModelDiff, testing-based approach similarity comparison. Instead directly comparing weights, activations, outputs two models, compare their behavioral patterns on same set test inputs. Specifically,...

Scholar clustering has garnered increasing attention due to the explosive growth of scholar data. Although researchers have proposed many algorithms cluster scholars, they typically focus on scholars from intrinsic view (scholars' contents). These may lead inaccurate and biased results because ignore extrinsic (scholar's specialty) changeability scholars' interest in each view. In this paper, we propose a multi-view topic model (MSCT), which integrates complementary information both views...

Transfer learning is a popular software reuse technique in the deep community that enables developers to build custom models (students) based on sophisticated pretrained (teachers). However, like vulnerability inheritance traditional reuse, some defects teacher model may also be inherited by students, such as well-known adversarial vulnerabilities and backdoors. Reducing challenging since student unaware of how trained and/or attacked. In this paper, we propose ReMoS, relevant slicing reduce...

Pressure-induced enhancement in optoelectronic properties of black-TiO2 nanocrystals are presented. The underlying mechanisms elucidated by comparing its behavior with that white-TiO2. disordered layer contributes to...

Mobile social applications have been widely used by Internet users. Users can efficiently acquire many kinds of information and share their statuses various platforms. However, when a user intends to through the user's applications, set access permission only before is posted. Once posted, it completely beyond control. Specifically, during recommendation process (for friend or information) in cannot control who get his/her information. If one accessed another malicious user, privacy be...

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...

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...

Recently, diffusion-based deep generative models (e.g., Stable Diffusion) have shown impressive results in text-to-image synthesis. However, current often require multiple passes of prompt engineering by humans order to produce satisfactory for real-world applications. We propose BeautifulPrompt, a model high-quality prompts from very simple raw descriptions, which enables generate more beautiful images. In our work, we first fine-tuned the BeautifulPrompt over low-quality and collecting...

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...

In this paper, we propose a general network adaptation framework, namely WealthAdapt, to effectively adapt large for small data tasks, with the assistance of wealth related data. While many existing algorithms have proposed techniques resource-constrained systems, they typically implement based on dataset and do not perform well when facing tasks. Because poor feature expression ability, it may result in incorrect filter selection overfitting during fine-tuning process. first expand target...

Artificial intelligence (AI) powered drug development has received remarkable attention in recent years. It addresses the limitations of traditional experimental methods that are costly and time-consuming. While there have been many surveys attempting to summarize related research, they only focus on general AI or specific aspects such as natural language processing graph neural network. Considering rapid advance computer vision, using molecular image enable appears be a more intuitive...