Current neural network models of primate vision focus on replicating overall levels of behavioral accuracy, often neglecting perceptual decisions' rich, dynamic nature. Here, we introduce a novel computational framework to model the dynamics of human behavioral choices by learning to align the temporal dynamics of a recurrent neural network (RNN) to human reaction times (RTs). We describe an approximation that allows us to constrain the number of time steps an RNN takes to solve a task with human RTs. The approach is extensively evaluated against various psychophysics experiments. We also show that the approximation can be used to optimize an ``ideal-observer'' RNN model to achieve an optimal tradeoff between speed and accuracy without human data. The resulting model is found to account well for human RT data. Finally, we use the approximation to train a deep learning implementation of the popular Wong-Wang decision-making model. The model is integrated with a convolutional neural network (CNN) model of visual processing and evaluated using both artificial and natural image stimuli. Overall, we present a novel framework that helps align current vision models with human behavior, bringing us closer to an integrated model of human vision.

In an urban environment, the accurate prediction of congestion levels is a prerequisite for formulating traffic demand management strategies reasonably. Current traffic forecasting studies mostly focus on the road topological network and assume that the spatial linkages of road segments is fixed, thus ignoring associated congestion level changes between road segments. This study introduces a fusion-based graph convolutional network called the F-GCN. The model aims to capture the dynamic correlations of the congestion levels among road segments while constructing the static topology of the road network. In particular, the entropy-based grey relation analysis method is first implemented in the dynamic graph convolutional network (GCN) module to measure the potential correlations among the observed congestion levels. Then, spatio-temporal blocks that integrate GCN layers, spatial attention mechanisms, and long short-term memory layers are built for both the static and dynamic GCN modules. Finally, the F-GCN model is developed by fusing the static GCN and dynamic GCN module. The numerical experiments on Beijing taxies demonstrated that the proposed F-GCN model achieved a significant 5.47%, 5.64%, and 8.14% improvements for the 15-, 30-, and 45-min prediction tasks in the weighted mean absolute percentage error compared to the state-of-the-art baseline models. This improvement highlights the effectiveness of the model in predicting congestion levels, demonstrating its superiority and potential in capturing the dynamic potential correlations among the congestion levels of road segments.

Although there have been many researches and corresponding degrees of freedom (DoF) analysis on multiple pairwise users exchanging data by one-hop relaying through one or more relays, the two-hop relay system still needs further exploring. In this paper, we design a communication scheme for multiple pairs of users in the two-tier heterogeneous networks (HetNets) with wireless backhaul link, which includes four phases: two multiple access (MAC) phases and two broadcast (BC) phases. Firstly, a new transmission and coding-decoding scheme combining generalized signal alignment (GSA), zero forcing (ZF) and physical layer network coding (PLNC) is proposed, which jointly takes account of users, small base-stations (SBSs) and macro BS (MBS). Furthermore, we analyze the sum rate for users relayed by SBS or also helped by MBS, the influence of the number of antenna configurations on the DoF, and the constraints that antenna should meet. Besides, compared with the traditional SSA scheme, the superiority of this scheme is highlighted. Meanwhile, simulation results illustrate that the proposed scheme can achieve much higher system sum rate than ZF scheme.

In recent years, considerable studies have been devoted to physical‐layer network coding (PLNC) in a two‐way relay channel (TWRC) network with the symmetric channel. However, it may not be appropriate for the asymmetric channel in which the source‐to‐relay channels are not equally strong and the two sources may utilise different modulation. In this study, the authors studied the heterogeneous PLNC design in the asymmetric TWRC network with high‐order modulation and proposed a relay denoising mapping approach to reduce the decoding error probability. Firstly, they put forward a cluster maximisation algorithm to enlarge one cluster by sucking the neighbour points, which satisfy the exclusive law. Then, based on this algorithm, they designed the heuristic nearest‐neighbour denoising mapping (HNDM) approach. The HNDM algorithm divides the nearest neighbour superimposed constellation points into the same cluster as far as possible. Points in the same cluster are coded into a new symbol by a specific mapping that achieves a many‐to‐one mapping. Performance results are provided in terms of symbol error rate. Under diversity channel conditions, the performances of 16‐quadrature amplitude modulator (16‐QAM)‐8‐phase‐shift keying and 16‐QAM‐quadrature phase‐shift keying are evaluated.

Attention-based recurrent neural network models for joint intent detection and slot filling have achieved a state-of-the-art performance. Most previous works exploited semantic level information to calculate the attention weights. However, few works have taken the importance of word level information into consideration. In this paper, we propose WAIS, word attention for joint intent detection and slot filling. Considering that intent detection and slot filling have a strong relationship, we further propose a fusion gate that integrates the word level information and semantic level information together for jointly training the two tasks. Extensive experiments show that the proposed model has robust superiority over its competitors and sets the state-of-the-art.