The proliferation of Internet of Things (IoT) devices has led to the generation of vast amounts of sensitive and decentralized data. Federated Learning (FL) offers a promising solution by enabling collaborative model training without centralizing raw data. However, existing methods still face challenges such as reduced model accuracy, low robustness, and potential data privacy leakage. To address these issues, we propose a Secure and Robust Federated Learning framework (SRFL) under a dual-server architecture. Specifically, we use two servers to separate the functions of model aggregation and verification, minimizing the threat of single points of failure. We further construct verification models based on perturbed or encrypted client updates to filter malicious updates and defend against poisoning attacks. Meanwhile, we integrate the Cheon-Kim-Kim-Song (CKKS) scheme with random perturbation and ciphertext transformation mechanisms, ensuring that servers cannot access plaintext client updates throughout the training process. Formal security analysis and extensive experiments demonstrate that our scheme consistently achieves high accuracy and robustness under adversarial conditions while preserving data privacy.

In the Internet of Things (IoT) environment, a large number of edge devices collaboratively generate and process data, making Federated Learning (FL) an appealing solution for privacy-preserving model training without sharing raw data. However, traditional FL frameworks suffer from performance degradation due to model poisoning attacks under Non-Independent and Identically Distributed (Non-IID) settings, as well as lack decision-making explainability. To address these challenges, we propose the Robustness-enhanced and explainable Federated Learning (ReFed) framework by combining Local Interpretable Model-agnostic Explanations (LIME) with similarity-based aggregation. Specifically, by fitting a local linear regression model to approximate the decision processes of both the global and local models for improved transparency, and dynamically computing client aggregation weights based on inter-model prediction similarity for enhanced robustness. Experiments demonstrate that ReFed achieves 99.76% accuracy on MNIST and 87.65% accuracy on CIFAR-10, outperforming typical FL methods in adversarial settings.

Non-fungible tokens (NFTs) are unique digital assets that play an increasingly important role in decentralized markets, supporting new forms of ownership, valuation, and exchange. Their inherently multimodal structure, which encompasses visual content, metadata, and trading history, has led to a growing academic interest in modeling NFT pricing and market behavior. However, existing research is limited by the lack of comprehensive datasets that unify these modalities with consistent formatting and longitudinal coverage. To address this gap, we introduce MultiNFT, a large-scale multimodal dataset comprising 50 curated profile picture (PFP) NFT collections, including 523,020 unique assets and 2.38 million transaction records from April 2021 to September 2025. MultiNFT integrates standardized images, structured metadata, and time-series trading data, along with rarity scores and aesthetic features, offering a unified foundation for multimodal learning and NFT analytics. Unlike prior datasets that focus on visual similarity or static snapshots, MultiNFT captures evolving valuation dynamics across market cycles and connects them to trait-level characteristics. We demonstrate the utility of the dataset through three case studies, including within-collection rarity-price analysis, visual feature clustering across collections, and quantifying feature contributions in a comprehensive pricing model. By bridging computer vision, behavioral modeling, and financial forecasting, MultiNFT supports a wide range of interdisciplinary research and practical use cases. The dataset is publicly available and is intended to promote reproducible experimentation and further exploration of the mechanisms driving value in digital asset ecosystems.

With the rapid development of the Internet of Things (IoT), a large amount of multimodal medical data is outsourced to the IoT cloud to reduce local computation and storage costs. To retrieve encrypted multimodal medical data in cloud computing, privacy-preserving cross-modal retrieval (PPCMR) has attracted significant attention. However, existing PPCMR solutions are challenging to meet real-time retrieval requirements for large-scale datasets due to linear index structures and complex encryption algorithms. To solve these issues, we propose a novel PPCMR scheme, named CSCMR, based on canonical correlation analysis (CCA) and symmetric key hidden vector encryption (SHVE). First, we use locality-sensitive hashing and CCA to cluster multimodal medical data according to semantic similarity. Then, we build an index structure for Bloom filter based on the clustering results, achieving constant retrieval complexity. Finally, we encrypt each value in the filter using SHVE algorithm, protecting data privacy while reducing encryption computation overheads. Formal security analysis proves that our scheme can resist selective chosen-plaintext attack. Extensive experiments demonstrate that our scheme is effective and feasible.

In the era of Internet of Things (IoT) and federated learning (FL), where distributed training models are essential, the FL paradigm has come into the spotlight for researchers. However, the inconsistency in the sources of client data and non-independent and identically distributed (Non-IID) heterogeneous characteristics lead to loss in model accuracy. Existing method which attempts to homogenize data distribution among clients based on generative adversarial networks (GANs) incurs high computation overheads on clients in IoT. In this article, we propose a lightweight feature prototype knowledge transfer (FPKT) mechanism. By capturing the essence of data categories, FPKT generates pseudo-features without requiring the original data features, thereby efficiently enhancing model accuracy. We formally prove that FPKT resists chosen plaintext attack (CPA) and experiments demonstrate that our scheme achieves a hundredfold increase in computational efficiency and improves model accuracy by up to 40%.

To address the asynchronous challenges stemming from user resource constraints and intermittent network connections in distributed Internet of Things (IoT) systems, asynchronous federated learning (AFL) has been extensively explored in both academic and industrial domains. However, existing AFL approaches still struggle to effectively tackle the issue of model aggregation information loss due to delayed users, which leads to degraded model accuracy. To alleviate the impact of delayed or unavailable model updates on model aggregation, we propose a novel model update enhancement method to compensate for the loss of model aggregation information. Specifically, we utilize delayed model updates as an update agent for this user and correct the weights of these updates within the delayed rounds threshold to ensure the delayed user’s contribution to the model aggregation. Additionally, we integrate symmetric homomorphic encryption (SHE) into AFL to ensure user privacy while simultaneously minimizing computational overhead. Lastly, we conduct extensive experiments to demonstrate that our scheme improves model accuracy by 6.53% compared to state-of-the-art solutions. Our code is available athttps://github.com/MiaoGroup-XDU/acc-improved-ppafl.

Abstract Conversational recommender systems (CRS) enable traditional recommender systems to interact with users by asking questions about their preferences and recommending items. Conversation recommendation has made significant progress, however, studies on attribute-aware conversational recommendation have overlooked the problem that after obtaining the candidate attribute set, the relevance of the attributes in the candidate attribute set to the user’s current preferred attributes is not further considered, resulting in the existence of many user-uninterested attributes in the candidate attribute set. This seriously affects the dialogue quality and reduces the user’s patience, decreasing recommendation accuracy. To address this problem, this paper proposes a new framework called Attribute Clipping based on Dynamic Graph (ACDG). In ACDG, firstly, the reasoning module uses the restriction property of graph structure to obtain a large set of candidate attributes, and then the attribute clipping module filters out the attributes with high relevance to the current user’s preferred attributes. Thus, ACDG can obtain a better-quality set of candidate attributes. Through extensive experiments on four benchmark CRS datasets, we validate the effectiveness of the method.

We propose an approach to estimate arm and hand dynamics from monocular video by utilizing the relationship between arm and hand. Although monocular full human motion capture technologies have made great progress in recent years, recovering accurate and plausible arm twists and hand gestures from in-the-wild videos still remains a challenge. To solve this problem, our solution is proposed based on the fact that arm poses and hand gestures are highly correlated in most real situations. To fully exploit arm-hand correlation as well as inter-frame information, we carefully design a Spatial-Temporal Parallel Arm-Hand Motion Transformer (PAHMT) to predict the arm and hand dynamics simultaneously. We also introduce new losses to encourage the estimations to be smooth and accurate. Besides, we collect a motion capture dataset including 200K frames of hand gestures and use this data to train our model. By integrating a 2D hand pose estimation model and a 3D human pose estimation model, the proposed method can produce plausible arm and hand dynamics from monocular video. Extensive evaluations demonstrate that the proposed method has advantages over previous state-of-the-art approaches and shows robustness under various challenging scenarios.

In Taobao, the largest e-commerce platform in China, billions of items are provided and typically displayed with their images.For better user experience and business effectiveness, Click Through Rate (CTR) prediction in online advertising system exploits abundant user historical behaviors to identify whether a user is interested in a candidate ad. Enhancing behavior representations with user behavior images will help understand user's visual preference and improve the accuracy of CTR prediction greatly. So we propose to model user preference jointly with user behavior ID features and behavior images. However, training with user behavior images brings tens to hundreds of images in one sample, giving rise to a great challenge in both communication and computation. To handle these challenges, we propose a novel and efficient distributed machine learning paradigm called Advanced Model Server (AMS). With the well-known Parameter Server (PS) framework, each server node handles a separate part of parameters and updates them independently. AMS goes beyond this and is designed to be capable of learning a unified image descriptor model shared by all server nodes which embeds large images into low dimensional high level features before transmitting images to worker nodes. AMS thus dramatically reduces the communication load and enables the arduous joint training process. Based on AMS, the methods of effectively combining the images and ID features are carefully studied, and then we propose a Deep Image CTR Model. Our approach is shown to achieve significant improvements in both online and offline evaluations, and has been deployed in Taobao display advertising system serving the main traffic.

Gender classification is a fundamental and important application in computer vision, and it has become a research hotspot. Real-world applications require gender classification in unconstrained conditions where traditional methods are not appropriate. This paper proposes a Deep Convolutional Neural Network for feature extraction together with fully-connected layers for metric learning. A Siamese network is built for similarity measuring to promote the performance of classification. Extensive experiments on several databases demonstrate that a significant improvement can be obtained for gender classification tasks in both constrained and unconstrained conditions.

This work proposes a novel convolutional neural network architecture which can locate landmarks accurately by learning local responses of facial landmarks. The network consists of a Conditional Variational Auto-Encoder(CVAE) and a Deep Convolutional Neural Network(DCNN). The CVAE is used to learn the response maps of facial landmarks from face images and the DCNN is used to learn accurate landmark locations from the response maps and facial textures. The CVAE consists of a face encoder, which extracts high-level information from raw pixels, and a decoder which outputs local response maps from high-level coding. We derive the CVAE used for catching local responses as an optimization problem, which can be solved through back-propagation. Extensive experiments show that the proposed CVAE can learn better local response maps than Fully Convolutional Network(FCN). Our method outperforms state-of-the-art methods on AFLW(5 points) and the challenging subset of 300-W(68 points), which means our method shows advantages in the condition of complex poses and expressions.