Federated Learning (FL) is a popular distributed machine learning method that enables the development of a robust global model through decentralized computation and periodic model aggregation, without requiring direct access to clients' data. However, data heterogeneity poses a significant challenge in FL, and the global long-tail distribution exacerbates this issue. While substantial research has focused on mitigating performance degradation caused by long-tailed distributions, existing methods typically concentrate on addressing discrepancies between local and global class distributions, often overlooking the fact that these discrepancies stem from variations in the data itself. To address this, we propose a novel approach, Federated Context Optimization and Feature Information Decoupling (FedDR), which generates partition strategies for each sample to extract and leverage long-tail, global, personalized, and label-text information within its features to enhance the representational distinction of tail classes. Specifically, we first design a Feature Information Decoupling module that separates global, personalized, and long-tail information within the features and incorporates this information into the loss function to strengthen the global model's focus on personalized information in tail samples. Furthermore, to exploit the textual label information embedded in the samples, we integrate a cross-modal model, CoOp, which utilizes open-vocabulary prior knowledge, and implement dynamic knowledge distillation between the client model and CoOp to enhance the client model's feature representation capability. Extensive experimental results on multiple benchmarks demonstrate that the proposed FedDR outperforms state-of-the-art methods in the federated long-tailed learning setting.

The emergence of novel generative modeling paradigms, particularly audio language models, has significantly advanced the field of song generation. Although state-of-the-art models are capable of synthesizing both vocals and accompaniment tracks up to several minutes long concurrently, research about partial adjustments or editing of existing songs is still underexplored, which allows for more flexible and effective production. In this paper, we present SongEditor, the first song editing paradigm that introduces the editing capabilities into language-modeling song generation approaches, facilitating both segment-wise and track-wise modifications. SongEditor offers the flexibility to adjust lyrics, vocals, and accompaniments, as well as synthesizing songs from scratch. The core components of SongEditor include a music tokenizer, an autoregressive language model, and a diffusion generator, enabling generating an entire section, masked lyrics, or even separated vocals and background music. Extensive experiments demonstrate that the proposed SongEditor achieves exceptional performance in end-to-end song editing, as evidenced by both objective and subjective metrics.

The scaling of large language models (LLMs) has revolutionized their capabilities in various tasks, yet this growth must be matched with efficient computational strategies. The Mixture-of-Experts (MoE) architecture stands out for its ability to scale model size without significantly increasing training costs. Despite their advantages, current MoE models often display parameter inefficiency. For instance, a pre-trained MoE-based LLM with 52 billion parameters might perform comparably to a standard model with 6.7 billion. Being a crucial part of MoE, current routers in different layers independently assign tokens without leveraging historical routing information, potentially leading to suboptimal token-expert combinations and the parameter inefficiency problem. To alleviate this issue, we introduce the Layerwise Recurrent Router for Mixture-of-Experts (RMoE). RMoE leverages a Gated Recurrent Unit (GRU) to establish dependencies between routing decisions across consecutive layers. Such layerwise recurrence can be efficiently parallelly computed for input tokens and introduces negotiable costs. Our extensive empirical evaluations demonstrate that RMoE-based language models consistently outperform a spectrum of baseline models. Furthermore, RMoE integrates a novel computation stage orthogonal to existing methods, allowing seamless compatibility with other MoE architectures. Our analyses attribute RMoE's gains to its effective cross-layer information sharing, which also improves expert selection and diversity.

Ultra-Wideband (UWB) is widely used to mitigate drift in visual-inertial odometry (VIO) systems. Consistency is crucial for ensuring the estimation accuracy of a UWB-aided VIO system. An inconsistent estimator can degrade localization performance, where the inconsistency primarily arises from two main factors: (1) the estimator fails to preserve the correct system observability, and (2) UWB anchor positions are assumed to be known, leading to improper neglect of calibration uncertainty. In this paper, we propose a consistent and tightly-coupled visual-inertial-ranging odometry (CVIRO) system based on the Lie group. Our method incorporates the UWB anchor state into the system state, explicitly accounting for UWB calibration uncertainty and enabling the joint and consistent estimation of both robot and anchor states. Further-more, observability consistency is ensured by leveraging the invariant error properties of the Lie group. We analytically prove that the CVIRO algorithm naturally maintains the system’s correct unobservable subspace, thereby preserving estimation consistency. Extensive simulations and experiments demonstrate that CVIRO achieves superior localization accuracy and consistency compared to existing methods.

This paper presents a novel robust online calibration framework for Ultra-Wideband (UWB) anchors in UWB-aided Visual-Inertial Navigation Systems (VINS). Accurate anchor positioning, a process known as calibration, is crucial for integrating UWB ranging measurements into state estimation. While several prior works have demonstrated satisfactory results by using robot-aided systems to autonomously calibrate UWB systems, there are still some limitations: 1) these approaches assume accurate robot localization during the initialization step, ignoring localization errors that can compromise calibration robustness, and 2) the calibration results are highly sensitive to the initial guess of the UWB anchors’ positions, reducing the practical applicability of these methods in real-world scenarios. Our approach addresses these challenges by explicitly incorporating the impact of robot localization uncertainties into the calibration process, ensuring robust initialization. To further enhance the robustness of the calibration results against initialization errors, we propose a tightly-coupled Schmidt Kalman Filter (SKF)-based online refinement method, making the system suitable for practical applications. Simulations and real-world experiments validate the improved accuracy and robustness of our approach.

Asynchronous federated learning (AFL) allows participants to immediately submit trained models without waiting for other participants, building upon the federated learning (FL). Due to privacy concerns, FL is more susceptible to financial fraud, compounded by the gradient delay issues introduced by asynchronous submissions, making defense against financial fraud more challenging. Motivated by the above finding, we propose a secure and privacy-preserving AFL defense method for image datasets with an implanted backdoor via filtering redundant neurons (BDAFL), enhancing its resilience against such attacks without compromising privacy. We utilize gradient compensation to mitigate the impact of delays introduced by asynchrony. To counter financial fraud, we employ an anomaly detection algorithm based on neurons’ weights and ensemble distillation to eliminate the affected neurons implanted with the backdoor, rendering the attack ineffective. Extensive experiments demonstrate the effectiveness and superiority of our approach.

This paper develops a new Distributed approach for solving the inverse problem of a Differentiable Dynamic Game (D3G), which enables robots to learn multi-robot coordination from given demonstrations. We formulate multi-robot coordination as the Nash equilibrium of a parameterized dynamic game, where the behavior of each robot is dictated by an objective function that also depends on the behavior of its neighboring robots. The coordination thus can be adapted by tuning the parameters of the objective and the local dynamics of each robot. The proposed algorithm enables each robot to automatically tune such parameters in a distributed and coordinated fashion — only using the data of its neighbors without global information. Its key novelty is the development of a distributed solver for a diff-KKT condition that can enhance scalability and reduce the computational load for gradient computation. We test the proposed algorithm in simulation with heterogeneous robots given different task configurations. The results demonstrate its effectiveness and generalizability for learning multi-robot coordination from demonstrations.

The rapid proliferation of Internet of Things (IoT) devices has brought about unprecedented convenience to people’s daily lives. However, this growth has also created opportunities for hackers to launch large-scale botnet attacks using these devices. As a result, it is critical to deploy real-time traffic classifiers on edge gateways to detect network intrusions and improve near-source protection capabilities. To this end, we propose iDetector, a novel real-time intrusion detection solution for IoT networks that is simple in structure and easy to reproduce. iDetector samples network conversations in real-time using a sliding sampling window and generates traffic samples that integrate multiple features. This allows the samples to accurately capture the patterns of each type of traffic. We propose the nonlinear feature transformation (NFT) algorithm based on the prior distribution of traffic features to increase the information entropy of the samples and thereby improve the classification performance. To enable deployment on edge gateways, we propose EdgeNet, a lightweight deep neural network model that utilizes depthwise separable convolution and self-attention mechanism to enhance classification performance while reducing the number of model parameters. Experimental evaluations show that our solution outperforms state-of-the-art deep learning-based solutions in terms of classification performance and has faster classification speed on resource-constrained edge gateways.

Federated Learning (FL) has emerged as a privacy-preserving paradigm enabling collaborative model training among distributed clients. However, current FL methods operate under the closed-world assumption, i.e., all local training data originates from a global labeled dataset balanced across classes, which is often invalid for practical scenarios. In contrast, in many open-world settings, data have been observed to exhibit heavy-tailed distributions, particularly in the realm of mobile computing and Internet of Things (IoT). Heavy-tailed data can have a significant negative impact on the performance of learning algorithms due to amplifying the heterogeneity in the FL environment. To this end, we introduce a novel framework to counter biased training caused by diverse and imbalanced classes. This framework includes a balance-aware reward aggregation mechanism addressing local majority and global minority class disparities. Rewards are assigned based on client class prevalence for fair aggregation. A calibration module supplements global aggregation to manage conflicts from inconsistent data distribution among clients. Using reward aggregation and calibration, we effectively mitigate heavy-tailed distribution effects, enhancing FL model performance. This framework seamlessly integrates with leading FL methods, demonstrated through extensive experiments on benchmark and real-world datasets.