In recent years, the rise in energy consumption across infrastructure has highlighted the need for more energy-efficient technologies. This is particularly critical in edge computing environments, where resources and power are limited. Consequently, there is increasing interest in improving the energy efficiency of resource-intensive tasks on edge servers. Edge servers commonly use Log-Structured Merge-tree-based Key-Value Store (LSM-KVS), to manage continuous data streams from edge devices. A key operation in LSM-KVS, known as compaction, merges key-value pairs in a CPU-intensive and energy-demanding process. Additionally, delays during compaction can cause write stalls, blocking I/O operations and degrading performance. This creates a significant challenge in balancing energy consumption and system performance. To address these challenges, we propose ECO-KVS, a solution that improves both energy efficiency and performance in LSM-KVS by offloading compaction tasks across edge servers in an edge federation. ECO-KVS leverages a real-time learning model to predict compaction time and energy consumption, reducing write stalls and enhancing overall energy efficiency. Implemented on RocksDB, ECO-KVS achieves up to 21% higher throughput compared to the baseline RocksDB and improves the performance-to-energy efficiency ratio by up to 18 % compared to EdgePilot, a state-of-the-art solution for edge environments.

Federated voiceprint recognition enables decentralized biometric authentication with enhanced adaptability to heterogeneous clients. Conventional federated learning (FL) methods often face significant challenges due to data heterogeneity among clients, resulting in suboptimal performance. This paper presents FedM2M (Federated Meta-Learning with Model Memory and Momentum Increment), a novel framework designed to enhance model adaptability and generalization in heterogeneous audio environments. Through the integration of model-agnostic meta-learning for adaptive local training, coupled with gradient increment optimization and momentum-based convergence stabilization, FedM2M effectively addresses the challenges of non-independent and identically distributed (non-IID) data distributions. Extensive empirical evaluation across benchmark datasets, including Librispeech, VoxCeleb, and Zhvoice, demonstrates that FedM2M consistently outperforms state-of-the-art FL algorithms. Ablation analysis further confirms FedM2M's superior accuracy and robustness, particularly in non-IID scenarios, establishing its efficacy as a robust solution for real-world voiceprint recognition applications.

Effective coordination in multi-agent systems remains challenging in dynamic and partially observable environments, where agents must reason over evolving interdependencies and limited communication bandwidth. We propose ADAPT, a unified framework for multi-agent coordination that integrates message compression, dependency estimation, and a novel auction-based dynamic prioritization mechanism. In ADAPT, agents exchange compact messages and compute dependency scores to determine how much their behavior depends on others. A distributed auction protocol then assigns priority positions, guiding autoregressive decision-making in a manner aligned with inter-agent influence. This enables flexible, influence-aware coordination without centralized control or extensive communication rounds. Experiments on SMACv2 and GRF show that ADAPT achieves higher win rates, faster convergence, and lower communication cost compared to state-of-the-art baselines. Further analyses confirm its scalability to large teams, compatibility with value decomposition, and runtime efficiency. These results show that ADAPT enables scalable, efficient, and modular multi-agent coordination.

Multi-agent reinforcement learning has achieved substantial progress under the centralized training with decentralized execution framework. However, most existing methods assume deterministic and noise-free local observations, limiting applicability to real-world environments characterized by stochastic partial observability. This paper introduces GraphSem, a semantic-graph communication framework designed to enhance agent coordination under observation uncertainty and randomized initial conditions. Graph-Sem employs Transformer-based encoders to abstract higher-level features from observations, selectively transmits these features via dynamic communication weighting, and fuses inter-agent information through an attention-guided graph convolutional network. To approximate aspects of real-world sensing challenges, we introduce controlled stochasticity to both observations and initial states during training. Experiments on perturbed variants of SMAC and Traffic Junction benchmarks show that GraphSem outperforms state-of-the-art baselines across diverse coordination tasks, with improvements of up to 30.4% in average win rates. Ablation studies suggest that semantic encoding, graph-based message fusion, and adaptive communication mechanisms collectively contribute to enhanced robustness, sample efficiency, and performance under stochastic conditions.

Accurately predicting the probability distribution of quantized latent representations is a critical challenge for entropy models in learned video compression (LVC). Existing mainstream LVC methods typically adopt ready-made entropy models based on image compression, which fail to fully exploit the information of spatial-temporal correlation. To address this issue, we propose a spatial correlation priors and hierarchical temporal attention (SCP-HTA) model, which exploits the spatial correlation information from the current video frames and refine the temporal information from the context. First, we extract the spatial correlation of the current frame to guide the generation of masks, enabling the frame to leverage more information during encoding and decoding process. Additionally, to obtain more accurate temporal information, we introduce a hierarchical temporal attention module at channel level when we generate the context. Experimental results demonstrate that the proposed SCP-HTA model achieve 15.76% bitrate saving in PSNR and 61.82% in MS-SSIM on average across all test datasets when compared with VTM-13.2 (LDP).

Learned video compression (LVC) methods typically align spatial-temporal transformation features with optical flow to perform motion compensation. However, existing LVC methods typically rely on a single reference feature to provide local detail features. This ignores global structural information, resulting in limited capabilities when dealing with fast motion or occlusion scenarios. In addition, with the encoding of P-frames, errors accumulate, leading to a degradation in reconstruction quality. To address these issues, we propose the Neighbor-Aware Feature-Driven Motion Compensation (NAFD-MC) that utilizes the spatial-temporal correlations of the neighboring features to explore the global structural information and the local detail information. Furthermore, we introduce the Synergy Filtering Module (SFM) to enhance inter-frame consistency and alleviate the error accumulation. Experimental results demonstrate that our method outperforms the H.266/VVC reference software VTM-13.2 in public benchmark datasets.

Multi-agent systems in non-stationary environments face challenges due to rapidly changing dynamics, leading to quick obsolescence of experiences in the replay buffer. To address this, we propose the Adaptive Experience Replay with Attention-Based Sequence Embedding (AERAS) framework, which integrates sequence embedding with an attention mechanism to prioritize experiences based on their relevance. By assigning adaptive weights, AERAS emphasizes relevant experiences while diminishing the impact of outdated ones, enhancing efficiency and learning performance in multi-agent reinforcement learning. Evaluations on the StarCraft II Multi-Agent Challenge and Google Research Football environments show that AERAS consistently outperforms state-of-the-art methods, achieving faster convergence and higher win rates. Ablation studies confirm the essential roles of sequence embedding and attention mechanisms in boosting AERAS's robustness and adaptability, underscoring its effectiveness in managing non-stationary environments within multi-agent systems.

Monitoring vital signs, such as heart rate (HR) and respiratory rate (RR) is critical for veterinary medicine. The existing contact based systems are difficult to use on animals for longer periods as they may cause movement restriction. Contactless solutions have recently gained more popularity due to their ease-of-use. However, there are very few validated systems for animals. In this study, we propose a contactless vital signs monitoring system, CageDot for animals. Our system provides continuous real-time monitoring of HR and RR during hospitalization. The CageDot is placed under the animal cage and detects the heart vibrations using a geophone sensor. The CageDot also has a signal quality control algorithm to address the problem of obtaining high-quality cardiac data in real-life noisy environments, such as hospitals. Compared to the existing works that use controlled environments, this is especially significant. The algorithm includes several steps that include background noise/movement removal, subject movement detection, heartbeat extraction, and vital signs estimation. The experimental results on 16 hospitalized dogs and cats show that this system can achieve high accuracy for vital signs monitoring with a mean absolute error (MAE) of 4.71 for HR (3.8% error rate) and 1.28 (8% error rate) for RR.

Heterogeneous federated learning (HFL) enables collaborative learning across clients with diverse model architectures and data distributions while preserving privacy. However, existing HFL approaches often struggle to effectively address the challenges posed by model diversity, leading to suboptimal performance and limited generalization ability. This paper pro-poses Fed2PKD, a novel HFL framework that tackles these challenges through a two-pronged knowledge distillation approach. Fed2PKD combines prototypical contrastive knowledge distillation to align client embeddings with global class prototypes and semi-supervised global knowledge distillation to capture global data characteristics. Experimental results on three benchmarks (MNIST, CIFAR10, and CIFAR100) demonstrate that Fed2PKD significantly outperforms existing state-of-the-art HFL methods, achieving average improvements of up to 30.53%, 13.89%, and 5.80 % in global model accuracy, respectively. Furthermore, Fed2PKD enables personalized models for each client, adapting to their specific data distributions and model architectures while benefiting from global knowledge sharing. Theoretical analysis provides convergence guarantees for Fed2PKD under realistic assumptions. Fed2PKD represents a significant step forward in HFL, unlocking the potential for privacy-preserving collaborative learning in real-world scenarios with model and data diversity.

Distributed Learning is pivotal for training extensive deep neural networks across multiple nodes, leveraging parallel computation to hasten the learning process. However, it faces challenges in communication efficiency and resource utilization. Asynchronous Quantized Stochastic Gradient Descent (AQSGD) addresses communication bottlenecks by updating quantized model parameters, thereby expediting training and reducing bandwidth usage. Yet, current stochastic quantization methods may inadequately capture varied gradient distributions, leading to accumulated biases and amplified quantization errors. These issues are amplified as the number of distributed nodes grows. This study proposes a novel Stochastic Quantization with Multivariate Gaussians (SQMG) for distributed machine learning. SQMG employs a multivariate Gaussian model to represent the relationships in the gradient updates for quantization. The SQMG approach allows for constructing an optimized quantization target space, coupled with an iterative mapping scheme that effectively projects the parameters onto this space while minimizing quantization errors. Experiments on DNN and CNN models for MNIST and CIFAR-10 show that SQMG increases accuracy by 0.92% and 1.54% for DNN and CNN models, respectively, compared to conventional quantization methods. The results validate SQMG’s ability to reduce quantization errors and improve model accuracy in distributed learning systems.

This paper presents the design and evaluation of an engagement-free and contactless vital signs and occupancy monitoring system called BedDot. While many existing works demonstrated contactless vital signs estimation, they do not address the practical challenge of environment noises, online bed occupancy detection and data quality assessment in the realworld environment. This work presents a robust signal quality assessment algorithm consisting of three parts: bed occupancy detection, movement detection, and heartbeat detection, to identify high-quality data. It also presents a series of innovative vital signs estimation algorithms that leverage the advanced signal processing and Bayesian theorem for contactless heart rate (HR), respiration rate (RR), and inter-beat interval (IBI) estimation. The experimental results demonstrate that BedDot achieves over 99% accuracy for bed occupancy detection, and MAE of 1.38 BPM, 1.54 BPM, and 24.84 ms for HR, RR, and IBI estimation, respectively, compared with an FDA-approved device. The BedDot system has been extensively tested with data collected from 75 subjects for more than 80 hours under different conditions, demonstrating its generalizability across different people and environments.

Multi-agent reinforcement learning commonly uses a global team reward signal to represent overall collaborative performance. Value decomposition breaks this global reward into estimated individual value functions per agent, enabling efficient training. However, in sparse reward environments, agents struggle to assess if their actions achieve the team goal, slowing convergence. This impedes the algorithm’s convergence rate and overall efficacy. We present IPERS, an Individual Prioritized Experience Replay algorithm with Subgoals for Sparse Reward Multi-Agent Reinforcement Learning. IPERS integrates joint action decomposition and prioritized experience replay, maintaining invariance between global and individual loss gradients. Subgoals serve as intermediate goals that break down complex tasks into simpler steps with dense feedback and provide helpful intrinsic rewards that guide agents. This facilitates learning coordinated policies in challenging collaborative environments with sparse rewards. Experimental evaluations of IPERS in both the SMAC and GRF environments demonstrate rapid adaptation to diverse multi-agent tasks and significant improvements in win rate and convergence performance relative to state-of-the-art algorithms.

Collecting Wi-Fi fingerprints is essential for Wi-Fi-based indoor localization techniques. However, this process can be time-consuming and labor-intensive due to the spatial and tempo-ral variations of Wi-Fi signals caused by environmental factors, interference, and fading. Moreover, the variability of signals emit-ted by different access points can hinder localization accuracy, especially in complex indoor environments. To overcome these challenges, we propose the Attention Mechanism-based Transfer Learning Indoor Localization (AMTL-Loc) framework, which transfers a pre-trained model from a source space to a target space and adapts it using minimal data by extracting redundant information from Wi-Fi fingerprints. Our experimental evalu-ations show that the AMTL-Loc framework can significantly reduce the fingerprint collection workload in diverse indoor environments while maintaining high localization accuracy compared to existing state-of-the-art indoor localization methods. Therefore, our framework offers a promising solution to enhance the efficiency and accuracy of Wi-Fi-based indoor localization techniques.

Stochastic computing (SC) is a low-cost computation paradigm that can replace conventional binary arithmetic to provide a low hardware footprint with high scalability. However, since the SC bitstream length grows with the precision of the represented data, regardless of its lower power consumption, the convolutional SC-based neural networks may not be efficient in hardware area and energy. This work proposes a novel SC accelerator, PSC-Conv, to implement the convolutional layer using a new binary-interfaced stochastic computing architecture. PSC-Conv exploits kernel-wise parallelism in CNNs, reducing hardware footprint and energy consumption. Experimental re-sults show that the proposed implementation excels among several state-of-the-art SC-based implementations regarding area and power efficiency. We also compared the implementations of three modern CNNs, including LeNet-5, MobileNet, and ResNet-50. Experimental results demonstrate that, on average, PSC-Conv can achieve 5.02x speedup and 87.9% energy reduction compared with the binary implementation.

Runoff prediction is essential for flood forecasting, irrigation planning, and sustainable water resource management. However, accurate predictions can be challenging due to the involvement of multiple variables. This paper presents a novel Graph Convolution-based Spatial-temporal Attention LSTM Multi-Task learning (GC-SALM) model for accurate runoff predictions. Our approach combines a multilayer neural network and an attention mechanism for enhanced generalization performance. The GC-SALM model employs spatial attention and graph convolutional networks to discern local and global spatial patterns, while temporal attention and LSTM are utilized to capture temporal characteristics within extended sequences. Experimental results reveal that the proposed model outperforms six state-of-the-art methods in runoff prediction and flow calibration, emphasizing its potential for real-world hydrological applications.

Federated learning (FL) enables model training on decentralized devices while preserving data privacy. However, data heterogeneity poses a significant challenge to FL, and various approaches have been proposed to address it. Existing research has mainly focused on either enhancing global models or customizing personalized models for clients. This paper proposes a novel approach, FedCRC, that decouples the machine learning model into a representation extractor and predictor. This enables us to enhance both generalization and personalization, thereby addressing the challenge of data heterogeneity in FL. The approach employs a stable global predictor to unify the representation learning criterion during the training of the representation extractor. Additionally, a personalized predictor is trained for each client to achieve a personalized model tailored to the local data distribution. Our FedCRC algorithm was evaluated on multiple benchmark datasets with varying distributions, covering diverse settings. Extensive experimental results demonstrate the effectiveness of our method.

Asynchronous federated learning is a distributed machine learning paradigm that may alleviate the impact of straggler nodes and improve the efficiency of federated training. However, some nodes can become sluggish, and node dropout may frequently happen for various reasons, such as network connection constraints, energy deficits, and system faults. Consequently, the global model may deviate from the desired convergence direction and lead to suboptimal results. This work proposes an asynchronous federated learning framework, FedDGIC, to mitigate the impact of the node dropout problem. The proposed framework can improve training efficiency by utilizing a dynamic grouping algorithm with gradient compensation. Experiments are performed in a real federated learning environment using two datasets, i.e., MNIST and CIFAR-10. Compared with three state-of-the-art methods, the proposed FedDGIC can significantly improve training efficiency and provide reliable asynchronous federated learning.

Multi-agent systems have recently received significant attention from researchers in many scientific fields. The value factorization method is popular for scaling up cooperative reinforcement learning in multi-agent environments. However, the approximation of the joint value function may introduce a significant disparity between the estimated and actual joint reward value function, leading to a local optimum for cooperative multi-agent deep reinforcement learning. In addition, as the number of agents increases, the input space grows exponentially, negatively impacting the convergence performance of multi-agent algorithms. This work proposes an efficient multi-agent rein-forcement learning algorithm, QDN, to enhance the convergence performance in cooperative multi-agent tasks. The proposed QDN scheme utilizes a competitive network to enable the agents to learn the value of the environmental state without the influence of actions. Hence, the error between the estimated joint reward value function and the actual joint reward value function can be significantly reduced, preventing the emergence of sub-optimal actions. Meanwhile, the proposed QDN algorithm utilizes the parametric noise on the network weights to introduce random-ness in the network's weights so that the agents can explore the environments and states effectively, thereby improving the convergence performance of the QDN algorithm. We evaluate the proposed QDN scheme using the SMAC challenges with various map difficulties. Experimental results show that the QDN algorithm excels in the convergence speed and the success rate in all scenarios compared to some state-of-the-art methods. Further experiments using four additional multi-agent tasks demonstrate that the QDN algorithm is robust in various multi-agent tasks and can significantly improve the training convergence performance compared with the state-of-the-art methods.