To see the other types of publications on this topic, follow the link: Asynchronous heterogeneous temporal integrator.
Journal articles on the topic 'Asynchronous heterogeneous temporal integrator'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 16 journal articles for your research on the topic 'Asynchronous heterogeneous temporal integrator.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse journal articles on a wide variety of disciplines and organise your bibliography correctly.
1
Fekak, Fatima-Ezzahra, Michael Brun, Anthony Gravouil, and Bruno Depale. "A new heterogeneous asynchronous explicit–implicit time integrator for nonsmooth dynamics." Computational Mechanics 60, no. 1 (March 11, 2017): 1–21. http://dx.doi.org/10.1007/s00466-017-1397-0.
Full text
2
Zafati, Eliass, and Julie Al Hout. "Reflection error analysis for wave propagation problems solved by a heterogeneous asynchronous time integrator." International Journal for Numerical Methods in Engineering 115, no. 6 (May 9, 2018): 651–94. http://dx.doi.org/10.1002/nme.5820.
Full text
3
Li, Longyuan, Jihai Zhang, Junchi Yan, Yaohui Jin, Yunhao Zhang, Yanjie Duan, and Guangjian Tian. "Synergetic Learning of Heterogeneous Temporal Sequences for Multi-Horizon Probabilistic Forecasting." Proceedings of the AAAI Conference on Artificial Intelligence 35, no. 10 (May 18, 2021): 8420–28. http://dx.doi.org/10.1609/aaai.v35i10.17023.
Full textAbstract:
Time-series is ubiquitous across applications, such as transportation, finance and healthcare. Time-series is often influenced by external factors, especially in the form of asynchronous events, making forecasting difficult. However, existing models are mainly designated for either synchronous time-series or asynchronous event sequence, and can hardly provide a synthetic way to capture the relation between them. We propose Variational Synergetic Multi-Horizon Network (VSMHN), a novel deep conditional generative model. To learn complex correlations across heterogeneous sequences, a tailored encoder is devised to combine the advances in deep point processes models and variational recurrent neural networks. In addition, an aligned time coding and an auxiliary transition scheme are carefully devised for batched training on unaligned sequences. Our model can be trained effectively using stochastic variational inference and generates probabilistic predictions with Monte-Carlo simulation. Furthermore, our model produces accurate, sharp and more realistic probabilistic forecasts. We also show that modeling asynchronous event sequences is crucial for multi-horizon time-series forecasting.
4
Agrawal, Shaashwat, Aditi Chowdhuri, Sagnik Sarkar, Ramani Selvanambi, and Thippa Reddy Gadekallu. "Temporal Weighted Averaging for Asynchronous Federated Intrusion Detection Systems." Computational Intelligence and Neuroscience 2021 (December 17, 2021): 1–10. http://dx.doi.org/10.1155/2021/5844728.
Full textAbstract:
Federated learning (FL) is an emerging subdomain of machine learning (ML) in a distributed and heterogeneous setup. It provides efficient training architecture, sufficient data, and privacy-preserving communication for boosting the performance and feasibility of ML algorithms. In this environment, the resultant global model produced by averaging various trained client models is vital. During each round of FL, model parameters are transferred from each client device to the server while the server waits for all models before it can average them. In a realistic scenario, waiting for all clients to communicate their model parameters, where client models are trained on low-power Internet of Things (IoT) devices, can result in a deadlock. In this paper, a novel temporal model averaging algorithm is proposed for asynchronous federated learning (AFL). Our approach uses a dynamic expectation function that computes the number of client models expected in each round and a weighted averaging algorithm for continuous modification of the global model. This ensures that the federated architecture is not stuck in a deadlock all the while increasing the throughput of the server and clients. To implicate the importance of asynchronicity in cybersecurity, the proposed algorithm is tested using NSL-KDD intrusion detection system datasets. The performance accuracy of the global model is about 99.5% on the dataset, outperforming traditional FL models in anomaly detection. In terms of asynchronicity, we get an increased throughput of almost 10.17% for every 30 timesteps.
5
Zhu, Yixin, Dongfen Li, Wenqiang Guo, and Fengli Zhang. "Effect of Heterogeneity of Vertex Activation on Epidemic Spreading in Temporal Networks." Mathematical Problems in Engineering 2014 (2014): 1–7. http://dx.doi.org/10.1155/2014/409510.
Full textAbstract:
Development of sensor technologies and the prevalence of electronic communication services provide us with a huge amount of data on human communication behavior, including face-to-face conversations, e-mail exchanges, phone calls, message exchanges, and other types of interactions in various online forums. These indirect or direct interactions form potential bridges of the virus spread. For a long time, the study of virus spread is based on the aggregate static network. However, the interaction patterns containing diverse temporal properties may affect dynamic processes as much as the network topology does. Some empirical studies show that the activation time and duration of vertices and links are highly heterogeneous, which means intense activity may be followed by longer intervals of inactivity. We take heterogeneous distribution of the node interactivation time as the research background to build an asynchronous communication model. The two sides of the communication do not have to be active at the same time. One derives the threshold of virus spreading on the communication mode and analyzes the reason the heterogeneous distribution of the vertex interactivation time suppresses the spread of virus. At last, the analysis and results from the model are verified on the BA network.
6
Zang, Yu, Zhe Xue, Shilong Ou, Lingyang Chu, Junping Du, and Yunfei Long. "Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction." Proceedings of the AAAI Conference on Artificial Intelligence 38, no. 15 (March 24, 2024): 16642–50. http://dx.doi.org/10.1609/aaai.v38i15.29603.
Full textAbstract:
Asynchronous federated learning (AFL) is a distributed machine learning technique that allows multiple devices to collaboratively train deep learning models without sharing local data. However, AFL suffers from low efficiency due to poor client model training quality and slow server model convergence speed, which are a result of the heterogeneous nature of both data and devices. To address these issues, we propose Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction (FedAC). Our framework consists of three key components. The first component is client weight evaluation based on temporal gradient, which evaluates the client weight based on the similarity between the client and server update directions. The second component is adaptive server update with prospective weighted momentum, which uses an asynchronous buffered update strategy and a prospective weighted momentum with adaptive learning rate to update the global model in server. The last component is client update with fine-grained gradient correction, which introduces a fine-grained gradient correction term to mitigate the client drift and correct the client stochastic gradient. We conduct experiments on real and synthetic datasets, and compare with existing federated learning methods. Experimental results demonstrate effective improvements in model training efficiency and AFL performance by our framework.
7
van der Meulen, Nick, Peter van Baalen, Eric van Heck, and Sipko Mülder. "No teleworker is an island: The impact of temporal and spatial separation along with media use on knowledge sharing networks." Journal of Information Technology 34, no. 3 (February 20, 2019): 243–62. http://dx.doi.org/10.1177/0268396218816531.
Full textAbstract:
Despite its prevalence, there is a lack of understanding regarding the effect of telework on an organization’s knowledge base. Recognizing the enabling role of electronic communication media, this article therefore addresses the interaction effects of media synchronicity and temporal as well as spatial separation among colleagues on sharing in knowledge networks. Special attention is paid to knowledge awareness (a form of metaknowledge representing “who knows what”) as well as homogeneous and heterogeneous knowledge sources to further explicate the relationship between coworker separation and knowledge sharing. Multiple surveys were placed between two smaller ethnographic investigations and combined with whole network data to form an in-depth study of 64 knowledge workers at a medium-sized European research and advisory organization. The results reveal that spatial separation directly reduces the frequency of knowledge sharing between colleagues, whereas temporal separation affects knowledge sharing through reduced knowledge awareness, resulting in lower job and proactive performance. The use of asynchronous media can serve to mitigate most of the negative effects of spatial separation on knowledge sharing but may also exacerbate the negative effect of temporal separation on teleworkers’ knowledge awareness of colleagues with identical expertise.
8
Rezaei, Mohammad R., Reza Saadati Fard, Milos R. Popovic, Steven A. Prescott, and Milad Lankarany. "Synchrony-Division Neural Multiplexing: An Encoding Model." Entropy 25, no. 4 (March 30, 2023): 589. http://dx.doi.org/10.3390/e25040589.
Full textAbstract:
Cortical neurons receive mixed information from the collective spiking activities of primary sensory neurons in response to a sensory stimulus. A recent study demonstrated an abrupt increase or decrease in stimulus intensity and the stimulus intensity itself can be respectively represented by the synchronous and asynchronous spikes of S1 neurons in rats. This evidence capitalized on the ability of an ensemble of homogeneous neurons to multiplex, a coding strategy that was referred to as synchrony-division multiplexing (SDM). Although neural multiplexing can be conceived by distinct functions of individual neurons in a heterogeneous neural ensemble, the extent to which nearly identical neurons in a homogeneous neural ensemble encode multiple features of a mixed stimulus remains unknown. Here, we present a computational framework to provide a system-level understanding on how an ensemble of homogeneous neurons enable SDM. First, we simulate SDM with an ensemble of homogeneous conductance-based model neurons receiving a mixed stimulus comprising slow and fast features. Using feature-estimation techniques, we show that both features of the stimulus can be inferred from the generated spikes. Second, we utilize linear nonlinear (LNL) cascade models and calculate temporal filters and static nonlinearities of differentially synchronized spikes. We demonstrate that these filters and nonlinearities are distinct for synchronous and asynchronous spikes. Finally, we develop an augmented LNL cascade model as an encoding model for the SDM by combining individual LNLs calculated for each type of spike. The augmented LNL model reveals that a homogeneous neural ensemble model can perform two different functions, namely, temporal- and rate-coding, simultaneously.
9
Di Carvalho, Josie Antonucci, and Stephen A. Wickham. "Does spatiotemporal nutrient variation allow more species to coexist?" Oecologia 194, no. 4 (October 24, 2020): 695–707. http://dx.doi.org/10.1007/s00442-020-04768-9.
Full textAbstract:
AbstractTemporal heterogeneity in nutrient availability is known to increase phytoplankton diversity by allowing more species to coexist under different resource niches. Spatial heterogeneity has also been positively correlated with species diversity. Here we investigated how temporal and spatial differences in nutrient addition together impact biodiversity in metacommunities varying in the degree of connectivity among the patches. We used a microcosm experimental design to test two spatiotemporal ways of supplying nutrients: synchronously (nutrients were added regionally—to all four patches at the same time) and asynchronously (nutrients were added locally—to a different patch each time), combined with two different degrees of connectivity among the patches (low or high connectivity). We used three species of algae and one species of cyanobacteria as the primary producers; and five ciliate and two rotifer species as the grazers. We expected higher diversity in metacommunities receiving an asynchronous nutrient supply, assuming stronger development of heterogeneous patches with this condition rather than with synchronous nutrient supply. This result was expected, however, to be dependent on the degree of connectivity among patches. We found significant effects of nutrient addition in both groups of organisms. Phytoplankton diversity increased until the fourth week (transiently) and zooplankton richness was persistently higher under asynchronous nutrient addition. Our results were consistent with our hypothesis that asynchronicity in nutrient supply would create a more favorable condition for species to co-occur. However, this effect was, in part, transient and was not influenced by the degree of connectivity.
10
Mueen, Ahmed, Mohammad Awedh, and Bassam Zafar. "Multi-obstacle aware smart navigation system for visually impaired people in fog connected IoT-cloud environment." Health Informatics Journal 28, no. 3 (July 2022): 146045822211126. http://dx.doi.org/10.1177/14604582221112609.
Full textAbstract:
Design of smart navigation for visually impaired/blind people is a hindering task. Existing researchers analyzed it in either indoor or outdoor environment and also it’s failed to focus on optimum route selection, latency minimization and multi-obstacle presence. In order to overcome these challenges and to provide precise assistance to visually impaired people, this paper proposes smart navigation system for visually impaired people based on both image and sensor outputs of the smart wearable. The proposed approach involves the upcoming processes: (i) the input query of the visually impaired people (users) is improved by the query processor in order to achieve accurate assistance. (ii) The safest route from source to destination is provided by implementing Environment aware Bald Eagle Search Optimization algorithm in which multiple routes are identified and classified into three different classes from which the safest route is suggested to the users. (iii) The concept of fog computing is leveraged and the optimal fog node is selected in order to minimize the latency. The fog node selection is executed by using Nearest Grey Absolute Decision Making Algorithm based on multiple parameters. (iv) The retrieval of relevant information is performed by means of computing Euclidean distance between the reference and database information. (v) The multi-obstacle detection is carried out by YOLOv3 Tiny in which both the static and dynamic obstacles are classified into small, medium and large obstacles. (vi) The decision upon navigation is provided by implementing Adaptive Asynchronous Advantage Actor-Critic (A3C) algorithm based on fusion of both image and sensor outputs. (vii) Management of heterogeneous is carried out by predicting and pruning the fault data in the sensor output by minimum distance based extended kalman filter for better accuracy and clustering the similar information by implementing Spatial-Temporal Optics Clustering Algorithm to reduce complexity. The proposed model is implemented in NS 3.26 and the results proved that it outperforms other existing works in terms of obstacle detection and task completion time.
11
Gupta, Aniket, Alix Reverdy, Jean-Martial Cohard, Basile Hector, Marc Descloitres, Jean-Pierre Vandervaere, Catherine Coulaud, et al. "Impact of distributed meteorological forcing on simulated snow cover and hydrological fluxes over a mid-elevation alpine micro-scale catchment." Hydrology and Earth System Sciences 27, no. 1 (January 10, 2023): 191–212. http://dx.doi.org/10.5194/hess-27-191-2023.
Full textAbstract:
Abstract. From the micro- to the mesoscale, water and energy budgets of mountainous catchments are largely driven by topographic features such as terrain orientation, slope, steepness, and elevation, together with associated meteorological forcings such as precipitation, solar radiation, and wind speed. Those topographic features govern the snow deposition, melting, and transport, which further impacts the overall water cycle. However, this microscale variability is not well represented in Earth system models due to coarse resolutions. This study explores the impact of precipitation, shortwave radiation, and wind speed on the water budget distribution over a 15.28 ha small, mid-elevation (2000–2200 m) alpine catchment at Col du Lautaret (France). The grass-dominated catchment remains covered with snow for 5 to 6 months per year. The surface–subsurface coupled distributed hydrological model ParFlow-CLM is used at a very high resolution (10 m) to simulate the impacts on the water cycle of meteorological variability at very small spatial and temporal scales. These include 3D simulations of hydrological fluxes with spatially distributed forcing of precipitation, shortwave radiation, and wind speed compared to 3D simulations of hydrological fluxes with non-distributed forcing. Our precipitation distribution method encapsulates the spatial snow distribution along with snow transport. The model simulates the dynamics and spatial variability of snow cover using the Common Land Model (CLM) energy balance module and under different combinations of distributed forcing. The resulting subsurface and surface water transfers are computed by the ParFlow module. Distributed forcing leads to spatially heterogeneous snow cover simulation, which becomes patchy at the end of the melt season and shows a good agreement with the remote sensing images (mean bias error (MBE) = 0.22). This asynchronous melting results in a longer melting period compared to the non-distributed forcing, which does not generate any patchiness. Among the distributed meteorological forcings tested, precipitation distribution, including snow transport, has the greatest impact on spatial snow cover (MBE = 0.06) and runoff. Shortwave radiation distribution has an important impact, reducing evapotranspiration as a function of the slope orientation (decreasing the slope between observed and simulated evapotranspiration from 1.55 to 1.18). For the primarily east-facing catchment studied here, distributing shortwave radiation helps generate realistic timing and spatial heterogeneity in the snowmelt at the expense of an increase in the mean bias error (from 0.06 to 0.22) for all distributed forcing simulations compared to the simulation with only distributed precipitation. Distributing wind speed in the energy balance calculation has a more complex impact on our catchment, as it accelerates snowmelt when meteorological conditions are favorable but does not generate snow patches at the end of our test case. This shows that slope- and aspect-based meteorological distribution can improve the spatio-temporal representation of snow cover and evapotranspiration in complex mountain terrain.
12
Zheng, Ciyan, Long Peng, Jason K. Eshraghian, Xiaoli Wang, Jian Cen, and Herbert Ho-Ching Iu. "Spiking Neuron Implementation Using a Novel Floating Memcapacitor Emulator." International Journal of Bifurcation and Chaos 32, no. 15 (December 15, 2022). http://dx.doi.org/10.1142/s0218127422502248.
Full textAbstract:
Memcapacitors (MCs) are promising candidates for the future design of low-power integrated neuromorphic computing systems, with particular emphasis on dynamical spiking neuron models that exhibit rich temporal behaviors. We present a novel floating flux-controlled MC that is designed using only three current feedback amplifiers, one analog multiplier, one capacitor and one resistor. Compared with existing floating MC emulators, our proposed design has a simpler structure without the need for DC biasing voltage sources, and can operate at higher working frequencies, and therefore enabling rapid prototyping of applied MC circuits for experimental verification of large-scale MC arrays. The consistency of the theoretical analysis, simulation and experimental results confirms the correctness and practicability of this new memcapacitor emulator. To further demonstrate a potential use of our MC, in this work, we apply the MC as the first parameterizable leaky integrator for spiking neuron through simulation and experiments. The intrinsic tunable capacitance of the MC can bring about novel short-term memory dynamics to neuronal circuits by dynamically modifying the membrane time constant on-the-fly, which ultimately resembles long-term potentiation, and can thus offer longer term memory. Our results highlight the potential for integrating heterogeneous spiking neural networks with richer temporal dynamics that rely on MC-based circuits to further the capability of neuromorphic computing.
13
Mintrone, Caterina, Luca Rindi, and Lisandro Benedetti‐Cecchi. "Stabilizing effects of spatially heterogeneous disturbance via reduced spatial synchrony on a rocky shore community." Ecology, January 29, 2024. http://dx.doi.org/10.1002/ecy.4246.
Full textAbstract:
AbstractUnderstanding how synchronous species fluctuations affect community stability is a main research topic in ecology. Yet experimental studies evaluating how changes in disturbance regimes affect the synchrony and stability of populations and communities remain rare. We hypothesized that spatially heterogeneous disturbances of moderate intensity would promote metacommunity stability by decreasing the spatial synchrony of species fluctuations. To test this hypothesis, we exposed rocky shore communities of algae and invertebrates to homogeneous and gradient‐like spatial patterns of disturbance at two levels of intensity for 4 years and used synchrony networks to characterize community responses to these disturbances. The gradient‐like disturbance at low intensity enhanced spatial β diversity compared to the other treatments and produced the most heterogeneous and least synchronized network, which was also the most stable in terms of population and community fluctuations. In contrast, homogeneous disturbance destabilized the community, enhancing spatial synchronization. Intense disturbances always reduced spatial β diversity, indicating that strong perturbations could destabilize communities via biotic homogenization regardless of their spatial structure. Our findings corroborated theoretical predictions, emphasizing the importance of spatially heterogeneous disturbances in promoting stability by amplifying asynchronous spatial and temporal fluctuations in population and community abundance. In contrast to other networks, synchrony networks are vulnerable to the removal of most peripheral nodes, which are less synchronized, but may contribute more to stability than other nodes by dampening large fluctuations in species abundance. Our findings suggest that climate change and direct anthropogenic disturbance can compromise the stability of ecological communities through combined effects on diversity and synchrony, as well as further affecting ecosystems through habitat loss.
14
Bi, Hongjie, Matteo di Volo, and Alessandro Torcini. "Asynchronous and Coherent Dynamics in Balanced Excitatory-Inhibitory Spiking Networks." Frontiers in Systems Neuroscience 15 (December 10, 2021). http://dx.doi.org/10.3389/fnsys.2021.752261.
Full textAbstract:
Dynamic excitatory-inhibitory (E-I) balance is a paradigmatic mechanism invoked to explain the irregular low firing activity observed in the cortex. However, we will show that the E-I balance can be at the origin of other regimes observable in the brain. The analysis is performed by combining extensive simulations of sparse E-I networks composed of N spiking neurons with analytical investigations of low dimensional neural mass models. The bifurcation diagrams, derived for the neural mass model, allow us to classify the possible asynchronous and coherent behaviors emerging in balanced E-I networks with structural heterogeneity for any finite in-degree K. Analytic mean-field (MF) results show that both supra and sub-threshold balanced asynchronous regimes are observable in our system in the limit N >> K >> 1. Due to the heterogeneity, the asynchronous states are characterized at the microscopic level by the splitting of the neurons in to three groups: silent, fluctuation, and mean driven. These features are consistent with experimental observations reported for heterogeneous neural circuits. The coherent rhythms observed in our system can range from periodic and quasi-periodic collective oscillations (COs) to coherent chaos. These rhythms are characterized by regular or irregular temporal fluctuations joined to spatial coherence somehow similar to coherent fluctuations observed in the cortex over multiple spatial scales. The COs can emerge due to two different mechanisms. A first mechanism analogous to the pyramidal-interneuron gamma (PING), usually invoked for the emergence of γ-oscillations. The second mechanism is intimately related to the presence of current fluctuations, which sustain COs characterized by an essentially simultaneous bursting of the two populations. We observe period-doubling cascades involving the PING-like COs finally leading to the appearance of coherent chaos. Fluctuation driven COs are usually observable in our system as quasi-periodic collective motions characterized by two incommensurate frequencies. However, for sufficiently strong current fluctuations these collective rhythms can lock. This represents a novel mechanism of frequency locking in neural populations promoted by intrinsic fluctuations. COs are observable for any finite in-degree K, however, their existence in the limit N >> K >> 1 appears as uncertain.
15
Wang, Xingyu, and Lan Zhou. "The Many Roles of Macrophages in Skeletal Muscle Injury and Repair." Frontiers in Cell and Developmental Biology 10 (July 11, 2022). http://dx.doi.org/10.3389/fcell.2022.952249.
Full textAbstract:
Skeletal muscle is essential to physical activity and energy metabolism. Maintaining intact functions of skeletal muscle is crucial to health and wellbeing. Evolutionarily, skeletal muscle has developed a remarkable capacity to maintain homeostasis and to regenerate after injury, which indispensably relies on the resident muscle stem cells, satellite cells. Satellite cells are largely quiescent in the homeostatic steady state. They are activated in response to muscle injury. Activated satellite cells proliferate and differentiate into myoblasts. Myoblasts fuse to form myotubes which further grow and differentiate into mature myofibers. This process is tightly regulated by muscle microenvironment that consists of multiple cellular and molecular components, including macrophages. Present in both homeostatic and injured muscles, macrophages contain heterogeneous functional subtypes that play diverse roles in maintaining homeostasis and promoting injury repair. The spatial-temporal presence of different functional subtypes of macrophages and their interactions with myogenic cells are vital to the proper regeneration of skeletal muscle after injury. However, this well-coordinated process is often disrupted in a chronic muscle disease, such as muscular dystrophy, leading to asynchronous activation and differentiation of satellite cells and aberrant muscle regeneration. Understanding the precise cellular and molecular processes regulating interactions between macrophages and myogenic cells is critical to the development of therapeutic manipulation of macrophages to promote injury repair. Here, we review the current knowledge of the many roles played by macrophages in the regulation of myogenic cells in homeostatic, regenerating, and dystrophic skeletal muscles.
16
Imran, Mubashir, Hongzhi Yin, Tong Chen, Nguyen Quoc Viet Hung, Alexander Zhou, and Kai Zheng. "ReFRS: Resource-efficient Federated Recommender System for Dynamic and Diversified User Preferences." ACM Transactions on Information Systems, August 29, 2022. http://dx.doi.org/10.1145/3560486.
Full textAbstract:
Owing to its nature of scalability and privacy by design, federated learning (FL) has received increasing interest in decentralized deep learning. FL has also facilitated recent research on upscaling and privatizing personalized recommendation services, using on-device data to learn recommender models locally. These models are then aggregated globally to obtain a more performant model, while maintaining data privacy. Typically, federated recommender systems (FRSs) do not take into account the lack of resources and data availability at the end-devices. In addition, they assume that the interaction data between users and items is i.i.d. and stationary across end-devices (i.e., users), and that all local recommender models can be directly averaged without considering the user’s behavioral diversity. However, in real scenarios, recommendations have to be made on end-devices with sparse interaction data and limited resources. Furthermore, users’ preferences are heterogeneous and they frequently visit new items. This makes their personal preferences highly skewed, and the straightforwardly aggregated model is thus ill-posed for such non-i.i.d. data. In this paper, we propose Resource Efficient Federated Recommender System (ReFRS) to enable decentralized recommendation with dynamic and diversified user preferences. On the device side, ReFRS consists of a lightweight self-supervised local model built upon the variational autoencoder for learning a user’s temporal preference from a sequence of interacted items. On the server side, ReFRS utilizes a scalable semantic sampler to adaptively perform model aggregation within each identified cluster of similar users. The clustering module operates in an asynchronous and dynamic manner to support efficient global model update and cope with shifting user interests. As a result, ReFRS achieves superior performance in terms of both accuracy and scalability, as demonstrated by comparative experiments on real datasets.