With the popularity of federated learning, federated domain generalization (FedDG) has attracted more and more attentions. Existing works of federated learning indicate that the generalization performance of the global model can be improved when the global model is obtained by aggregating local models according to a suitable weights. However, the existing methods to calculate weights do not fully utilize the data influences on the global model update, which gives us an opportunity to improve the generalization performance of the global model further. In this paper, we propose the method DI (data influences), which utilizes the data influences on the global model update to calculate dynamical weights of local model in each round of training. Specifically, the first component data influences calculator (DIC) of DI calculates the local weights of local model from the influences of each data on the global model update and we introduce the influences function to complete the calculation process. The second component data influences adjuster (DIA) of DI calculates the global weights (which are used in the aggregation process of the global model) from local weights. Extensive experiments indicate that our method improves the generalization performance of models significantly. In particular, our method improves model accuracy on benchmark datasets PACS, OfficeHome, and Office-31 by 1.79%, 1.61%, and 2.39% on average, respectively. Source code is publicly available at github.

Accurately forecasting long-term future wind power is critical to achieve safe power grid integration. This problem is quite challenging due to wind power's high volatility and randomness. In this paper, we propose a novel time series forecasting method, namely Deep Conditional Generative Spatio-Temporal model (DCGST), and its high accuracy is achieved by tackling two critical issues simultaneously: a proper handling of the non-stationarity of multiple wind power time series, and a fine-grained modeling of their complicated yet dynamic spatio-temporal dependencies. Specifically, we first formally define theSpatio-Temporal Concept Drift(STCD) problem of wind power, and then we propose a novel deep conditional generative model to learn probabilistic distributions of future wind power values under STCD. Three different tailored neural networks are designed for distributions parameterization, including a graph-based prior network, an attention-based recognition network, and a stochastic seq2seq-based generation network. They are able to encode the dynamic spatio-temporal dependencies of multiple wind power time series and infer one-to-many mappings for future wind power generation. Compared to existing methods, DCGST can learn better spatio-temporal representations of wind power data and learn better uncertainties of data distribution to generate future values. Comprehensive experiments on real-world datasets including the largest public turbine-level wind power dataset verify the effectiveness, efficiency, generality and scalability of our method.

With the popularity of artificial intelligence and cloud computing, many neural network models can be placed on the cloud server as an open service, such as Google Goggles and the online face recognition system of Baidu. The data owner sends his data to the cloud server to get the prediction result of data. Obviously, the cloud service provider can access model parameters and private data if there is no additional protection mechanism. On the one hand, if the adversary can access private data, they can freely use the artificial intelligence model and Big Data technologies to analyze the data owner. On the other hand, when the adversary can access model parameters, the interest of model owner would be harmed. Thus, preserving model parameters (model privacy) and private data (data privacy) becomes the key for applying neural network models as open cloud services. In this article, to protect the model privacy and data privacy in neural network prediction even when a cloud service provider colludes with the data owner or the model owner, we first propose a new system model with two no-colluding cloud servers and a corresponding security model. Then, we propose a new non-interactive outsourcing scheme, which can protect model privacy together with data privacy. Our scheme is able to resist collusive attacks of one server and the data owner as well as collusive attacks of one server and the model owner. At last, the security analyses indicate that our scheme just needs no collusion between cloud servers. The performance analyses indicate that our scheme is very lightweight for the data owner, and it is about tens of milliseconds for a neural network model with 1000 parameters.

Sequential Recommendation (SR) predicts the next interaction behavior via modeling the interaction between the user and the item over a time sequence. A series of works applied Self-Supervised Learning (SSL) in SR to obtain better user representations. Although these efforts proved effective, they only focused on the information of the user itself and ignores self-supervised signals from other users. Due to the widely observed homogeneity in recommender systems, these signals from other users are also vital for user representation. To this end, we propose a novel framework, Self-Supervised Learning based on Similar users for Sequential Recommendation (SSLSRec). We present a contrastive learning objective in SSLSRec to consider augmented views from the same user and similar users as positive samples. Moreover, we propose novel Insert and Substitute augmentation methods to construct more reasonable augmentation views for user sequences. Extensive experiments demonstrate the effectiveness of SSLSRec.

Identifying structural hole spanners that benefit from acting as bridges between communities is a core study in social network analysis. Existing methods for identification mainly focus on measuring the ability of users to control information propagation by bridging holes, while ignoring the impact of reinforcement of the holes themselves on the benefits of bridging spanners. A recent sociological study shows that the more reinforced a hole is, the more likely it is to bring high benefits to its spanners. In this paper, we propose a node embedding-based method ReHSe for identifying reinforced structural hole spanners in social networks. Specifically, an integrated embedding method is devised to extract features encoding reinforcement properties of nodes into a low-dimensional space. Further, to improve the robustness and accuracy of identification, an incremental learning strategy based on a reserved set is employed to train a scoring network in this subspace, to find top-k reinforced hole spanners. Extensive experimental results show that the performance of hole spanners identified by the proposed method outperforms several existing methods.

Attention Deficit Hyperactivity Disorder (ADHD) is a common psychiatric disorder among young children. However, there is no accurate and efficient method to diagnose ADHD up to now, due to the complexity of the pathological mechanism and clinical symptoms. This paper aims to present a spatio-temporal method for classification of ADHD and Typical Developing Children (TDC) using Resting-State functional Magnetic (rs-fMRI). To extract the most discriminative features in both space and time dimensions, 3-Dimensional Convolutional Neural Network (3D-CNN) and Gated Recurrent Unit (GRU) were respectively used to process 3D spatial and 1D temporal information in rs-fMRI. Before GRU, 1D filters with different scales were employed to capture significant features of different time intervals from temporal input. To evaluate proposed method, the 5-fold cross validation was employed using ADHD-200 global competition dataset. As a result, the average accuracy, sensitivity, specificity were 71.65 %, 68.00 % and 73.80 %, respectively. Experiment results show that our method performs better than existing methods, and our model not only has good generalization ability, but maintains a balance between sensitivity and specificity. We believe that our method can be used to build a more accurate automatic assistant diagnosis tool of ADHD.

With anonymous sessions, session-based recommendation aims to forecast user's next action. It has been a difficult endeavor due to the limited information and lack of user profiles. Recent advances have demonstrated that graph structure is more suited to model complex item transitions than chronological order alone. Most existing GNN-based models mainly concentrate on the current session, mining more intra-session sequential pattern data. Other models that leverage neighbor session information or item co-occurrence to obtain global collaborative signals are too sensitive to noise and are insufficient to infer user preference. In this paper, we propose a novel Intent-Aware Graph Neural Networks (IA-GNN) for session-based recommendation. In IA-GNN, we leverage two encoders to learn item embeddings:(1) Local Transition Encoder (LTE) based on session graph to learn complicated sequence dependencies, and (2) Intent Match Encoder (IME) with the help of intent-aware graph to obtain collaborative signals from the perspective of user intent. Furthermore, a tailored position enhanced soft attention mechanism joins the two levels of item representations to generate user preference. Extensive experiments on three real-world benchmark datasets demonstrate that our model is superior to the state-of-the-art models.

The study of time series forecasting is significant and useful in a variety of scenarios. However, due to the high degree of randomness and the complex contextual factors, it remains a difficult challenge. While several works based on machine learning and deep neural network have been proposed in recent years to address these challenges, most of them mine sequence features based on discrete points and overlook the fact that shape similarity plays an important role in inferring the future values. In this paper, we propose a seq2seq model with Shape Attention and Dilated Convolution (SADC) to tackle this problem. SADC contains two important phases: (1) Embedding with multi-scale dilated convolution. We first define the shape as a set of discrete points in a fixed-length window. The features hidden in the shape are then learned using multiple dilated convolutions with different kernels. (2) Inferring with shape attention. During this phase, we first present the shape attention, which aims to provide support information for inferring future values by generating the embedding vector of each prediction window based on shape similarity. The PreNet network is then built to predict the values using the embedding vector for each prediction window. The experimental results conducted on two datasets show that the performance of SADC model outperforms the state-of-the-art models on time series forecasting.

Due to the rapid development of the sharing economy, shared bikes have become one of the most popular and convenient traveling tools in intelligent transport systems. Aiming to save the time spent on waiting for or searching bikes at bike stations, the operators of bike-sharing systems need to dynamically dispatch bikes. Predicting the number of bikes for each station can help to optimize the repository of bikes. The usage of bikes is affected by several uncertain factors, so bike number prediction becomes a challenging and difficult problem. To manage this problem, we propose an algorithm called bimodal Gaussian inhomogeneous Poisson (BGIP) to predict the number of bikes. The BGIP includes three steps. First, the inhomogeneous Poisson process is adopted to describe the process that people arrive at a bike station to pick up or return bikes. Second, the bimodal Gaussian function is used to describe the intensity function of inhomogeneous Poisson process. In order to dynamically uncover the changing trend in the usage state of bikes, we propose a method to measure the influences of external factors on the usage of bikes. Third, the number of bikes is predicted by calculating the mean usage of bikes on the basis of checking-out and checking-in sequences. Experiments demonstrated that our algorithm outperformed the baseline algorithms in solving the bike prediction problem: accurately predicting the number of bikes and determining whether there is at least one bike available at a bike station.