Search

We propose a novel framework that leverages large language models (LLMs) to guide the rank selection in tensor network models for higher-order data analysis. By utilising the intrinsic reasoning capabilities and domain knowledge of LLMs, our approach offers enhanced interpretability of the rank choices and can effectively optimise the objective function. This framework enables users without specialised domain expertise to utilise tensor network decompositions and understand the underlying rationale within the rank selection process. Experimental results validate our method on financial higher-order datasets, demonstrating interpretable reasoning, strong generalisation to unseen test data, and its potential for self-enhancement over successive iterations. This work is placed at the intersection of large language models and higher-order data analysis.

Most of 3D single object trackers (SOT) in point clouds follow the two-stream multi-stage 3D Siamese or motion tracking paradigms, which process the template and search area point clouds with two parallel branches, built on supervised point cloud backbones. In this work, beyond typical 3D Siamese or motion tracking, we propose a neat and compact one-stream transformer 3D SOT paradigm from the novel perspective, termed as \textbf{EasyTrack}, which consists of three special designs: 1) A 3D point clouds tracking feature pre-training module is developed to exploit the masked autoencoding for learning 3D point clouds tracking representations. 2) A unified 3D tracking feature learning and fusion network is proposed to simultaneously learns target-aware 3D features, and extensively captures mutual correlation through the flexible self-attention mechanism. 3) A target location network in the dense bird's eye view (BEV) feature space is constructed for target classification and regression. Moreover, we develop an enhanced version named EasyTrack++, which designs the center points interaction (CPI) strategy to reduce the ambiguous targets caused by the noise point cloud background information. The proposed EasyTrack and EasyTrack++ set a new state-of-the-art performance ($\textbf{18\%}$, $\textbf{40\%}$ and $\textbf{3\%}$ success gains) in KITTI, NuScenes, and Waymo while runing at \textbf{52.6fps} with few parameters (\textbf{1.3M}). The code will be available at https://github.com/KnightApple427/Easytrack.

A novel tensor decomposition framework, termed Tensor Star (TS) decomposition, is proposed which represents a new type of tensor network decomposition based on tensor contractions. This is achieved by connecting the core tensors in a ring shape, whereby the core tensors act as skip connections between the factor tensors and allow for direct correlation characterisation between any two arbitrary dimensions. Uniquely, this makes it possible to decompose an order-$N$ tensor into $N$ order-$3$ factor tensors $\{\mathcal{G}_{k}\}_{k=1}^{N}$ and $N$ order-$4$ core tensors $\{\mathcal{C}_{k}\}_{k=1}^{N}$, which are arranged in a star shape. Unlike the class of Tensor Train (TT) decompositions, these factor tensors are not directly connected to one another. The so obtained core tensors also enable consecutive factor tensors to have different latent ranks. In this way, the TS decomposition alleviates the "curse of dimensionality" and controls the "curse of ranks", exhibiting a storage complexity which scales linearly with the number of dimensions and as the fourth power of the ranks.

Mobility entropy is proposed to measure predictability of human movements, based on which, the upper and lower bound of prediction accuracy is deduced, but corresponding mathematical expressions of prediction accuracy keeps yet open. In this work, we try to analyze and model prediction accuracy in terms of entropy based on the 2-order Markov chain model empirical results on a large scale CDR data set, which demonstrates the observation that users with the same level of entropy achieve different levels of accuracy\cite{Empirical}. After dividing entropy into intervals, we fit the probability density distributions of accuracy in each entropy interval with Gaussian distribution and then we estimate the corresponding mean and standard deviation of these distributions. After observing that the parameters vary with increasing entropy, we then model the relationship between parameters and entropy using least squares method. The mean can be modelled as a linear function, while the standard deviation can be modelled as a Gaussian distribution. Thus based on the above analysis, the probability density function of accuracy given entropy can be expressed by functional Gaussian distribution. The insights from our work is the first step to model the correlation prediction accuracy and predictability entropy, thus shed light on the further work in this direction.

Handover rate is one of the most import metrics to instruct mobility management and resource management in wireless cellular networks. In the literature, the mathematical expression of handover rate has been derived for homogeneous cellular network by both regular hexagon coverage model and stochastic geometry model, but there has not been any reliable result for heterogeneous cellular networks (HCNs). Recently, stochastic geometry modeling has been shown to model well the real deployment of HCNs and has been extensively used to analyze HCNs. In this paper, we give an effective handover analysis for HCNs by stochastic geometry modeling, derive the mathematical expression of handover rate by employing an infinitesimal method for a generalized multi-tier scenario, discuss the result by deriving some meaningful corollaries, and validate the analysis by computer simulation with multiple walking models. By our analysis, we find that in HCNs the handover rate is related to many factors like the base stations' densities and transmitting powers, user's velocity distribution, bias factor, pass loss factor and etc. Although our analysis focuses on the scenario of multi-tier HCNs, the analytical framework can be easily extended for more complex scenarios, and may shed some light for future study., Comment: 20 pages

The spatial structure of transmitters in wireless networks plays a key role in evaluating the mutual interference and hence the performance. Although the Poisson point process (PPP) has been widely used to model the spatial configuration of wireless networks, it is not suitable for networks with repulsion. The Ginibre point process (GPP) is one of the main examples of determinantal point processes that can be used to model random phenomena where repulsion is observed. Considering the accuracy, tractability and practicability tradeoffs, we introduce and promote the $\beta$-GPP, an intermediate class between the PPP and the GPP, as a model for wireless networks when the nodes exhibit repulsion. To show that the model leads to analytically tractable results in several cases of interest, we derive the mean and variance of the interference using two different approaches: the Palm measure approach and the reduced second moment approach, and then provide approximations of the interference distribution by three known probability density functions. Besides, to show that the model is relevant for cellular systems, we derive the coverage probability of the typical user and also find that the fitted $\beta$-GPP can closely model the deployment of actual base stations in terms of the coverage probability and other statistics., Comment: 29 pages, 11 figures

Though cooperative relaying is believed to be a promising technology to improve the energy efficiency of cellular networks, the relays' static power consumption might worsen the energy efficiency therefore can not be neglected. In this paper, we focus on whether and how the energy efficiency of cellular networks can be improved via relays. Based on the spatial Poisson point process, an analytical model is proposed to evaluate the energy efficiency of relay-assisted cellular networks. With the aid of the technical tools of stochastic geometry, we derive the distributions of signal-to-interference-plus-noise ratios (SINRs) and mean achievable rates of both non-cooperative users and cooperative users. The energy efficiency measured by "bps/Hz/W" is expressed subsequently. These established expressions are amenable to numerical evaluation and corroborated by simulation results., Comment: 6 pages, 5 figures, accepted by IEEE Globecom'12. arXiv admin note: text overlap with arXiv:1108.1257 by other authors

In this letter, we consider a joint macro-relay network with densely deployed relay stations (RSs) and dynamically varied traffic load measured by the number of users. An energy-efficient strategy is proposed by intelligently adjusting the RS working modes (active or sleeping) according to the traffic variation. Explicit expressions related to the network energy efficiency are derived based on stochastic geometry theory. Simulation results demonstrate that the derived analytic results are reasonable and the proposed strategy can significantly improve the network energy efficiency., Comment: This paper has been withdrawn by the author because this work needs to be further improved

In cognitive radio networks, channel aggregation (CA) and channel fragmentation (CF) techniques have been proposed to enhance the spectrum utilization. While most of the literature studies CA and CF independently, in this paper we combine CA and CF innovatively and present a new spectrum sharing strategy named CAF (Channel Aggregation and Fragmentation). We elaborate on the proposed CAF strategy and derive the balance equation by a continuous time Markov chain (CTMC) model. Then various system performance metrics including blocking probability, dropping probability, spectrum utilization and throughput of the secondary network are evaluated. Both analytical and simulation results show that our strategy lowers the blocking and dropping probabilities and enhances the spectrum utilization and throughput effectively. Moreover, by tuning the bandwidth requirement of each secondary user, different system performance can be achieved., Comment: 6 pages, 7 figures