arXiv subjects

Jiaxun Lu

Publications and source records attributed to Jiaxun Lu.

At least 19 recordsLinked to original sources

arXiv2026

Unified AI Gateway: A Framework for Joint Model Routing and KV Cache Management

Large language model (LLM) inference increasingly spans models that differ in size, capability, price, and provider. This shift creates two costs for developers. One is the integration cost of choosing among and switching between many models. The other is the inference cost of rebuilding a KV cache when it is unavailable or incompatible with the selected model. We define and analyze the Unified AI Gateway as a system setting for an edge-deployed AI traffic hub. It coordinates model routing, KV cache management, and compute placement across end devices, edge resources, and cloud model services. At request time, the gateway jointly selects a target model, an execution site, and a KV cache action under task-quality, latency, cost, and resource constraints. In parallel, background cache-management actions optimize KV cache placement, replication, retrieval, and lifecycle decisions for subsequent requests. We synthesize existing evidence on KV cache reuse, compression, cross-model mapping, distributed storage, and transfer, and discuss the remaining challenges of integrating these capabilities into one system. Across eight typical workload profiles, our workload-level analytical simulation reports TTFT speedups of 1.25$\times$--13.28$\times$ and input-cost benefits of 1.20$\times$--6.16$\times$.

cs.DC↗

arXiv2026

ENEC: A Lossless AI Model Compression Method Enabling Fast Inference on Ascend NPUs

The rapid scaling of Large Language Models presents significant challenges for their deployment and inference, particularly on resource-constrained specialized AI hardware accelerators such as Huawei's Ascend NPUs, where weight data transfer has become a critical performance bottleneck. While lossless compression can preserve model accuracy and reduce data volume, existing lossless compression algorithms exhibit extremely low throughput when ported to the Ascend NPU architecture. In this paper, we propose ENEC, a novel lossless compression method specifically customized for AI model weights and optimized for Ascend Neural Processing Units. ENEC adopts a block-based fixed-length encoding scheme and incorporates a series of NPU-specific optimizations: bit-width quantization with hierarchical halving bit-packing, vectorized branch-free integer transformation, and dependency-decoupled intra-segment scan for efficient prefix-sum computation. Experimental results demonstrate that ENEC outperforms existing state-of-the-art NPU compressors in both compression ratio and throughput. Compared to leading GPU solutions, ENEC achieves a 3.43X higher throughput than DietGPU and a 1.12X better compression ratio than nvCOMP. By reducing weight transmission overhead, ENEC significantly improves end-to-end inference performance, achieving up to a 6.3X speedup. On Ascend NPUs, ENEC is the first open-source lossless compression algorithm for model weights that achieves performance comparable to state-of-the-art GPU compressors, offering an effective solution for deploying large-scale AI models.

cs.AR↗

arXiv2022

Towards Effective Clustered Federated Learning: A Peer-to-peer Framework with Adaptive Neighbor Matching

In federated learning (FL), clients may have diverse objectives, and merging all clients' knowledge into one global model will cause negative transfer to local performance. Thus, clustered FL is proposed to group similar clients into clusters and maintain several global models. In the literature, centralized clustered FL algorithms require the assumption of the number of clusters and hence are not effective enough to explore the latent relationships among clients. In this paper, without assuming the number of clusters, we propose a peer-to-peer (P2P) FL algorithm named PANM. In PANM, clients communicate with peers to adaptively form an effective clustered topology. Specifically, we present two novel metrics for measuring client similarity and a two-stage neighbor matching algorithm based Monte Carlo method and Expectation Maximization under the Gaussian Mixture Model assumption. We have conducted theoretical analyses of PANM on the probability of neighbor estimation and the error gap to the clustered optimum. We have also implemented extensive experiments under both synthetic and real-world clustered heterogeneity. Theoretical analysis and empirical experiments show that the proposed algorithm is superior to the P2P FL counterparts, and it achieves better performance than the centralized cluster FL method. PANM is effective even under extremely low communication budgets.

cs.AI↗

arXiv2022

Unified Group Fairness on Federated Learning

Federated learning (FL) has emerged as an important machine learning paradigm where a global model is trained based on the private data from distributed clients. However, most of existing FL algorithms cannot guarantee the performance fairness towards different groups because of data distribution shift over groups. In this paper, we formulate the problem of unified group fairness on FL, where the groups can be formed by clients (including existing clients and newly added clients) and sensitive attribute(s). To solve this problem, we first propose a general fair federated framework. Then we construct a unified group fairness risk from the view of federated uncertainty set with theoretical analyses to guarantee unified group fairness on FL. We also develop an efficient federated optimization algorithm named Federated Mirror Descent Ascent with Momentum Acceleration (FMDA-M) with convergence guarantee. We validate the advantages of the FMDA-M algorithm with various kinds of distribution shift settings in experiments, and the results show that FMDA-M algorithm outperforms the existing fair FL algorithms on unified group fairness.

cs.LG↗

arXiv2021

How global observation works in Federated Learning: Integrating vertical training into Horizontal Federated Learning

Federated learning (FL) has recently emerged as a transformative paradigm that jointly train a model with distributed data sets in IoT while avoiding the need for central data collection. Due to the limited observation range, such data sets can only reflect local information, which limits the quality of trained models. In practice, the global information and local observations would require a joint consideration for learning to make a reasonable policy. However, in horizontal FL, the central agency only acts as a model aggregator without utilizing its global observation to further improve the model. This could significantly degrade the performance in some missions such as traffic flow prediction in network systems, where the global information may enhance the accuracy. Meanwhile, the global feature may not be directly transmitted to agents for data security. How to utilize the global observation residing in the central agency while protecting its safety thus rises up as an important problem in FL. In this paper, we develop a vertical-horizontal federated learning (VHFL) process, where the global feature is shared with the agents in a procedure similar to that of vertical FL without any extra communication rounds. By considering the delay and packet loss, we will analyze VHFL convergence and validate its performance by experiments. It is shown that the proposed VHFL could enhance the accuracy compared with horizontal FL while still protecting the security of global data.

cs.LG↗

arXiv2021

Convergence Analysis and System Design for Federated Learning over Wireless Networks

Federated learning (FL) has recently emerged as an important and promising learning scheme in IoT, enabling devices to jointly learn a model without sharing their raw data sets. However, as the training data in FL is not collected and stored centrally, FL training requires frequent model exchange, which is largely affected by the wireless communication network. Therein, limited bandwidth and random package loss restrict interactions in training. Meanwhile, the insufficient message synchronization among distributed clients could also affect FL convergence. In this paper, we analyze the convergence rate of FL training considering the joint impact of communication network and training settings. Further by considering the training costs in terms of time and power, the optimal scheduling problems for communication networks are formulated. The developed theoretical results can be used to assist the system parameter selections and explain the principle of how the wireless communication system could influence the distributed training process and network scheduling.

cs.LG↗

arXiv2020

Intelligent networking with Mobile Edge Computing: Vision and Challenges for Dynamic Network Scheduling

Mobile edge computing (MEC) has been considered as a promising technique for internet of things (IoT). By deploying edge servers at the proximity of devices, it is expected to provide services and process data at a relatively low delay by intelligent networking. However, the vast edge servers may face great challenges in terms of cooperation and resource allocation. Furthermore, intelligent networking requires online implementation in distributed mode. In such kinds of systems, the network scheduling can not follow any previously known rule due to complicated application environment. Then statistical learning rises up as a promising technique for network scheduling, where edges dynamically learn environmental elements with cooperations. It is expected such learning based methods may relieve deficiency of model limitations, which enhance their practical use in dynamic network scheduling. In this paper, we investigate the vision and challenges of the intelligent IoT networking with mobile edge computing. From the systematic viewpoint, some major research opportunities are enumerated with respect to statistical learning.

cs.NI↗

arXiv2019

Towards Big data processing in IoT: Path Planning and Resource Management of UAV Base Stations in Mobile-Edge Computing System

Heavy data load and wide cover range have always been crucial problems for online data processing in internet of things (IoT). Recently, mobile-edge computing (MEC) and unmanned aerial vehicle base stations (UAV-BSs) have emerged as promising techniques in IoT. In this paper, we propose a three-layer online data processing network based on MEC technique. On the bottom layer, raw data are generated by widely distributed sensors, which reflects local information. Upon them, unmanned aerial vehicle base stations (UAV-BSs) are deployed as moving MEC servers, which collect data and conduct initial steps of data processing. On top of them, a center cloud receives processed results and conducts further evaluation. As this is an online data processing system, the edge nodes should stabilize delay to ensure data freshness. Furthermore, limited onboard energy poses constraints to edge processing capability. To smartly manage network resources for saving energy and stabilizing delay, we develop an online determination policy based on Lyapunov Optimization. In cases of low data rate, it tends to reduce edge processor frequency for saving energy. In the presence of high data rate, it will smartly allocate bandwidth for edge data offloading. Meanwhile, hovering UAV-BSs bring a large and flexible service coverage, which results in the problem of effective path planning. In this paper, we apply deep reinforcement learning and develop an online path planning algorithm. Taking observations of around environment as input, a CNN network is trained to predict the reward of each action. By simulations, we validate its effectiveness in enhancing service coverage. The result will contribute to big data processing in future IoT.

cs.NI↗

arXiv2019

Towards Big data processing in IoT: network management for online edge data processing

Heavy data load and wide cover range have always been crucial problems for internet of things (IoT). However, in mobile-edge computing (MEC) network, the huge data can be partly processed at the edge. In this paper, a MEC-based big data analysis network is discussed. The raw data generated by distributed network terminals are collected and processed by edge servers. The edge servers split out a large sum of redundant data and transmit extracted information to the center cloud for further analysis. However, for consideration of limited edge computation ability, part of the raw data in huge data sources may be directly transmitted to the cloud. To manage limited resources online, we propose an algorithm based on Lyapunov optimization to jointly optimize the policy of edge processor frequency, transmission power and bandwidth allocation. The algorithm aims at stabilizing data processing delay and saving energy without knowing probability distributions of data sources. The proposed network management algorithm may contribute to big data processing in future IoT.

cs.NI↗

arXiv2018

Model change detection with application to machine learning

Model change detection is studied, in which there are two sets of samples that are independently and identically distributed (i.i.d.) according to a pre-change probabilistic model with parameter $θ$, and a post-change model with parameter $θ'$, respectively. The goal is to detect whether the change in the model is significant, i.e., whether the difference between the pre-change parameter and the post-change parameter $\|θ-θ'\|_2$ is larger than a pre-determined threshold $ρ$. The problem is considered in a Neyman-Pearson setting, where the goal is to maximize the probability of detection under a false alarm constraint. Since the generalized likelihood ratio test (GLRT) is difficult to compute in this problem, we construct an empirical difference test (EDT), which approximates the GLRT and has low computational complexity. Moreover, we provide an approximation method to set the threshold of the EDT to meet the false alarm constraint. Experiments with linear regression and logistic regression are conducted to validate the proposed algorithms.

stat.ML↗

arXiv2018

Minor probability events detection in big data: An integrated approach with Bayesian testing and MIM

The minor probability events detection is a crucial problem in Big data. Such events tend to include rarely occurring phenomenons which should be detected and monitored carefully. Given the prior probabilities of separate events and the conditional distributions of observations on the events, the Bayesian detection can be applied to estimate events behind the observations. It has been proved that Bayesian detection has the smallest overall testing error in average sense. However, when detecting an event with very small prior probability, the conditional Bayesian detection would result in high miss testing rate. To overcome such a problem, a modified detection approach is proposed based on Bayesian detection and message importance measure, which can reduce miss testing rate in conditions of detecting events with minor probability. The result can help to dig minor probability events in big data.

eess.SP↗

arXiv2018

Active and Adaptive Sequential learning

A framework is introduced for actively and adaptively solving a sequence of machine learning problems, which are changing in bounded manner from one time step to the next. An algorithm is developed that actively queries the labels of the most informative samples from an unlabeled data pool, and that adapts to the change by utilizing the information acquired in the previous steps. Our analysis shows that the proposed active learning algorithm based on stochastic gradient descent achieves a near-optimal excess risk performance for maximum likelihood estimation. Furthermore, an estimator of the change in the learning problems using the active learning samples is constructed, which provides an adaptive sample size selection rule that guarantees the excess risk is bounded for sufficiently large number of time steps. Experiments with synthetic and real data are presented to validate our algorithm and theoretical results.

cs.LG↗

arXiv2017

Semi-centralized control for multi-robot formation and theoretical lower bound

Multi-robot formation control enables robots to cooperate as a working group in completing complex tasks, which has been widely used in both civilian and military scenarios. Before moving to reach a given formation, each robot should choose a position from the formation so that the whole system cost is minimized. To solve the problem, we formulate an optimization problem in terms of the total moving distance and give a solution by the Hungarian method. To analyze the deviation of the achieved formation from the ideal one, we obtain the lower bound of formation bias with respect to system's parameters based on notions in information theory. As an extension, we discuss methods of transformation between different formations. Some theoretical results are obtained to give a guidance of the system design.

cs.RO↗

arXiv2017

Beyond Empirical Models: Pattern Formation Driven Placement of UAV Base Stations

This work considers the placement of unmanned aerial vehicle base stations (UAV-BSs) with criterion of minimum UAV-recall-frequency (UAV-RF), indicating the energy efficiency of mobile UAVs networks. Several different power consumptions, including signal transmit power, on-board circuit power and the power for UAVs mobility, and the ground user density are taken into account. Instead of conventional empirical stochastic models, this paper utilizes a pattern formation system to track the instable and non-ergodic time-varying nature of user density. We show that for a single time-slot, the optimal placement is achieved when the transmit power of UAV-BSs equals their on-board circuit power. Then, for multiple time-slot duration, we prove that the optimal placement updating problem is a nonlinear dynamic programming coupled with an integer linear programming. Since the original problem is NP-hard and can not be solved with conventional recursive methods, we propose a sequential-Markov-greedy-decision method to achieve near minimal UAV-RF in polynomial time. Further, we prove that the increment of UAV-RF caused by inaccurate predicted user density is proportional to the generalization error of learned patterns. Here, in regions with large area, high-rise buildings or low user density, large sample sets are required for effective pattern formation.

cs.IT↗

arXiv2017

Subcarrier grouping with environmental sensing for MIMO-OFDM systems over correlated double-selective fading channels

With the increase of physical antenna and subcarrier numbers in MIMO-OFDM systems, channel side information feedback amount and signal precoding complexity overburden will consume much more system resource, even become intolerable. To solve this problem, previous works mainly focused on fixed subcarrier grouping size and precoded MIMO signals in the same group with unitary channel state information (CSI). It could reduce the system overburden, but such a process would lead to system capacity loss due to the channel mismatch in precoding procedure. In this paper, we consider a MIMOOFDM system over double-selective i.i.d. Rayleigh channels and investigate the quantitative relation between group size and capacity loss theoretically. By exploiting our developed theoretical results, we also propose an adaptive subcarrier grouping algorithm, which not only enables to have a good control of system service quality but also to reduce system overburden significantly. Numerical results are shown to provide valuable insights on the system design of MIMO-OFDM systems and indicate that the proposed subcarrier grouping scheme is extremely efficient in some common scenarios.

cs.IT↗

arXiv2017

Energy-efficient 3D UAV-BS Placement Versus Mobile Users' Density and Circuit Power

Properly 3D placement of unmanned aerial vehicle mounted base stations (UAV-BSs) can effectively prolong the life-time of the mobile ad hoc network, since UAVs are usually powered by batteries. This paper involves the on-board circuit consumption power and considers the optimal placement that minimizes the UAV-recall-frequency (UAV-RF), which is defined to characterize the life-time of this kind of network. Theoretical results show that the optimal vertical and horizontal dimensions of UAV can be decoupled. That is, the optimal hovering altitude is proportional to the coverage radius of UAVs, and the slope is only determined by environment. Dense scattering environment may greatly enlarge the needed hovering altitude. Also, the optimal coverage radius is achieved when the transmit power equals to on-board circuit power, and hence limiting on-board circuit power can effectively enlarge life-time of system. In addition, our proposed 3D placement method only require the statistics of mobile users' density and environment parameters, and hence it's a typical on-line method and can be easily implemented. Also, it can be utilized in scenarios with varying users' density.

cs.IT↗

arXiv2017

Directivity-Beamwidth Tradeoff of Massive MIMO Uplink Beamforming for High Speed Train Communication

High-mobility adaption and massive Multiple-input Multiple-output (MIMO) application are two primary evolving objectives for the next generation high speed train (HST) wireless communication system. In this paper, we consider how to design a location-aware beamforming for the massive MIMO system in the high traffic density HST network. We first analyze the tradeoff between beam directivity and beamwidth, based on which we present the sensitivity analysis of positioning accuracy. Then, in order to guarantee a high efficient transmission, we derive an optimal problem to maximize the beam directivity under the restriction of diverse positioning accuracies. After that, we present a low-complexity beamforming design by utilizing location information, which requires neither eigen-decomposing (ED) the uplink channel covariance matrix (CCM) nor ED the downlink CCM (DCCM). Finally, we study the beamforming scheme in future high traffic density HST network, where a two HSTs encountering scenario is emphasized. By utilizing the real-time location information, we propose an optimal adaptive beamforming scheme to maximize the achievable rate region under limited channel source constraint. Numerical simulation indicates that a massive MIMO system with less than a certain positioning error can guarantee a required performance with satisfying transmission efficiency in the high traffic density HST scenario and the achievable rate region when two HSTs encounter is greatly improved as well.

cs.IT↗

arXiv2017

Toward Traffic Patterns in High-speed Railway Communication Systems: Power Allocation and Antenna Selection

In high-speed railway (HSR) communication systems, distributed antenna is usually employed to support frequent handover and enhance the signal to noise ratio to user equipments. In this case, dynamic time-domain power allocation and antenna selection (PAWAS) could be jointly optimized to improve the system performances. This paper consider this problem in such a simple way where dynamic switching between multiple-input-multiple-output (MIMO) and single-input-multiple-output (SIMO) is allowed and exclusively utilized, while the channel states and traffic demand are taken into account. The channel states includes sparse and rich scattering terrains, and the traffic patterns includes delay-sensitive and delay-insensitive as well as hybrid. Some important results are obtained in theory. In sparse scattering terrains, for delay-sensitive traffic, the PAWAS can be viewed as the generalization of channel-inversion associated with transmit antenna selection. On the contrary, for delay-insensitive traffic, the power allocation with MIMO can be viewed as channel-inversion, but with SIMO, it is traditional water-filling. For the hybrid traffic, the PAWAS can be partitioned as delay-sensitive and delay-insensitive parts by some specific strategies. In rich scattering terrains, the corresponding PAWAS is derived by some amendments in sparse scattering terrains and similar results are then presented.

cs.IT↗