The Matrix Decomposition techniques have been a vital computational approach to analyzing the hierarchy of functional connectivity in the human brain. However, there are still four shortcomings of these methodologies: 1). Large training samples; 2). Manually tuning hyperparameters; 3). Time-consuming and require extensive computational source; 4). It cannot guarantee convergence to a unique fixed point. Therefore, we propose a novel deep matrix factorization technique called Deep Linear Matrix Approximate Reconstruction (DELMAR) to bridge the abovementioned gaps. The advantages of the proposed method are: at first, proposed DELMAR can estimate the important hyperparameters automatically; furthermore, DELMAR employs the matrix backpropagation to reduce the potential accumulative errors; finally, an orthogonal projection is introduced to update all variables of DELMAR rather than directly calculating the inverse matrices. The validation experiments of three peer methods and DELMAR using real functional MRI signal of the human brain demonstrates that our proposed method can efficiently identify the spatial feature in fMRI signal even faster and more accurately than other peer methods. Moreover, the theoretical analyses indicate that DELMAR can converge to the unique fixed point and even enable the accurate approximation of original input as DNNs.
Deep Neural Networks (DNNs) have already become a crucial computational approach to revealing the spatial patterns in the human brain; however, there are three major shortcomings in utilizing DNNs to detect the spatial patterns in functional Magnetic Resonance Signals: 1). It is a fully connected architecture that increases the complexity of network structures that is difficult to optimize and vulnerable to overfitting; 2). The requirement of large training samples results in erasing the individual/minor patterns in feature extraction; 3). The hyperparameters are required to be tuned manually, which is time-consuming. Therefore, we propose a novel deep nonlinear matrix factorization named Deep Matrix Approximately Nonlinear Decomposition (DEMAND) in this work to take advantage of the shallow linear model, e.g., Sparse Dictionary Learning (SDL) and DNNs. At first, the proposed DEMAND employs a non-fully connected and multilayer-stacked architecture that is easier to be optimized compared with canonical DNNs; furthermore, due to the efficient architecture, training DEMAND can avoid overfitting and enables the recognition of individual/minor features based on a small dataset such as an individual data; finally, a novel rank estimator technique is introduced to tune all hyperparameters of DEMAND automatically. Moreover, the proposed DEMAND is validated by four other peer methodologies via real functional Magnetic Resonance Imaging data in the human brain. In short, the validation results demonstrate that DEMAND can reveal the reproducible meta, canonical, and sub-spatial features of the human brain more efficiently than other peer methodologies.
In this work, to efficiently help escape the stationary and saddle points, we propose, analyze, and generalize a stochastic strategy performed as an operator for a first-order gradient descent algorithm in order to increase the target accuracy and reduce time consumption. Unlike existing algorithms, the proposed stochastic the strategy does not require any batches and sampling techniques, enabling efficient implementation and maintaining the initial first-order optimizer's convergence rate, but provides an incomparable improvement of target accuracy when optimizing the target functions. In short, the proposed strategy is generalized, applied to Adam, and validated via the decomposition of biomedical signals using Deep Matrix Fitting and another four peer optimizers. The validation results show that the proposed random strategy can be easily generalized for first-order optimizers and efficiently improve the target accuracy.
Federated learning (FL) has been recognized as a promising distributed learning paradigm to support intelligent applications at the wireless edge, where a global model is trained iteratively through the collaboration of the edge devices without sharing their data. However, due to the relatively large communication cost between the devices and parameter server (PS), direct computing based on the information from the devices may not be resource efficient. This paper studies the joint communication and learning design for the over-the-air computation (AirComp)-based two-tier wireless FL scheme, where the lead devices first collect the local gradients from their nearby subordinate devices, and then send the merged results to the PS for the second round of aggregation. We establish a convergence result for the proposed scheme and derive the upper bound on the optimality gap between the expected and optimal global loss values. Next, based on the device distance and data importance, we propose a hierarchical clustering method to build the two-tier structure. Then, with only the instantaneous channel state information (CSI), we formulate the optimality gap minimization problem and solve it by using an efficient alternating minimization method. Numerical results show that the proposed scheme outperforms the baseline ones.
This paper studies the joint device selection and power control scheme for wireless federated learning (FL), considering both the downlink and uplink communications between the parameter server (PS) and the terminal devices. In each round of model training, the PS first broadcasts the global model to the terminal devices in an analog fashion, and then the terminal devices perform local training and upload the updated model parameters to the PS via over-the-air computation (AirComp). First, we propose an AirComp-based adaptive reweighing scheme for the aggregation of local updated models, where the model aggregation weights are directly determined by the uplink transmit power values of the selected devices and which enables the joint learning and communication optimization simply by the device selection and power control. Furthermore, we provide a convergence analysis for the proposed wireless FL algorithm and the upper bound on the expected optimality gap between the expected and optimal global loss values is derived. With instantaneous channel state information (CSI), we formulate the optimality gap minimization problems under both the individual and sum uplink transmit power constraints, respectively, which are shown to be solved by the semidefinite programming (SDR) technique. Numerical results reveal that our proposed wireless FL algorithm achieves close to the best performance by using the ideal FedAvg scheme with error-free model exchange and full device participation.
Orbital angular momentum (OAM) radars are able to estimate the azimuth angle and the rotation velocity of multiple targets without relative motion or beam scanning. Moreover, OAM wireless communications can achieve high spectral efficiency (SE) by utilizing a set of information-bearing modes on the same frequency channel. Benefitting from the above advantages, in this paper, we design a novel radar-centric joint OAM radar-communication (RadCom) scheme based on uniform circular arrays (UCAs), which modulates information signals on the existing OAM radar waveform. In details, we first propose an OAM-based three-dimensional (3-D) super-resolution position estimation and rotation velocity detection method, which can accurately estimate the 3-D position and rotation velocity of multiple targets. Then, we derive the posterior Cramer-Rao bound (PCRB) of the OAM-based estimates and, finally, we analyze the transmission rate of the integrated communication system. To achieve the best trade-off between imaging and communication, the transmitted integrated OAM beams are optimized by means of an exhaustive search method. Both mathematical analysis and simulation results show that the proposed radar-centric joint OAM RadCom scheme can accurately estimate the 3-D position and rotation velocity of multiple targets while ensuring the transmission rate of the communication receiver, which can be regarded as an effective supplement to existing joint RadCom schemes.
Intelligent reflecting surface (IRS) is envisioned to change the paradigm of wireless communications from "adapting to wireless channels" to "changing wireless channels". However, current IRS configuration schemes, consisting of sub-channel estimation and passive beamforming in sequence, conform to the conventional model-based design philosophies and are difficult to be realized practically in the complex radio environment. To create the smart radio environment, we propose a model-free design of IRS control that is independent of the sub-channel channel state information (CSI) and requires the minimum interaction between IRS and the wireless communication system. We firstly model the control of IRS as a Markov decision process (MDP) and apply deep reinforcement learning (DRL) to perform real-time coarse phase control of IRS. Then, we apply extremum seeking control (ESC) as the fine phase control of IRS. Finally, by updating the frame structure, we integrate DRL and ESC in the model-free control of IRS to improve its adaptivity to different channel dynamics. Numerical results show the superiority of our proposed joint DRL and ESC scheme and verify its effectiveness in model-free IRS control without sub-channel CSI.
Recently, semantic search has been successfully applied to e-commerce product search and the learned semantic space(s) for query and product encoding are expected to generalize to unseen queries or products. Yet, whether generalization can conveniently emerge has not been thoroughly studied in the domain thus far. In this paper, we examine several general-domain and domain-specific pre-trained Roberta variants and discover that general-domain fine-tuning does not help generalization, which aligns with the discovery of prior art. Proper domain-specific fine-tuning with clickstream data can lead to better model generalization, based on a bucketed analysis of a publicly available manual annotated query-product pair da
Cross-market recommendation aims to recommend products to users in a resource-scarce target market by leveraging user behaviors from similar rich-resource markets, which is crucial for E-commerce companies but receives less research attention. In this paper, we present our detailed solution adopted in the cross-market recommendation contest, i.e., WSDM CUP 2022. To better utilize collaborative signals and similarities between target and source markets, we carefully consider multiple features as well as stacking learning models consisting of deep graph recommendation models (Graph Neural Network, DeepWalk, etc.) and traditional recommendation models (ItemCF, UserCF, Swing, etc.). Furthermore, We adopt tree-based ensembling methods, e.g., LightGBM, which show superior performance in prediction task to generate final results. We conduct comprehensive experiments on the XMRec dataset, verifying the effectiveness of our model. The proposed solution of our team WSDM_Coggle_ is selected as the second place submission.
Recent studies have proven that deep neural networks are vulnerable to backdoor attacks. Specifically, by mixing a small number of poisoned samples into the training set, the behavior of the trained model can be maliciously controlled. Existing attack methods construct such adversaries by randomly selecting some clean data from the benign set and then embedding a trigger into them. However, this selection strategy ignores the fact that each poisoned sample contributes inequally to the backdoor injection, which reduces the efficiency of poisoning. In this paper, we formulate improving the poisoned data efficiency by the selection as an optimization problem and propose a Filtering-and-Updating Strategy (FUS) to solve it. The experimental results on CIFAR-10 and ImageNet-10 indicate that the proposed method is effective: the same attack success rate can be achieved with only 47% to 75% of the poisoned sample volume compared to the random selection strategy. More importantly, the adversaries selected according to one setting can generalize well to other settings, exhibiting strong transferability.