Abstract Task‐oriented dialogue system (TOD) is one kind of application of artificial intelligence (AI). The response generation module is a key component of TOD for replying to user's questions and concerns in sequential natural words. In the past few years, the works on response generation have attracted increasing research attention and have seen much progress. However, existing works ignore the fact that not each turn of dialogue history contributes to the dialogue response generation and give little consideration to the different weights of utterances in a dialogue history. In this article, we propose a hierarchical memory network mechanism with two steps to filter out unnecessary information of dialogue history. First, an utterance‐level memory network distributes various weights to each utterance (coarse‐grained). Second, a token‐level memory network assigns higher weights to keywords based on the former's output (fine‐grained). Furthermore, the output of the token‐level memory network will be employed to query the knowledge base (KB) to capture the dialogue‐related information. In the decoding stage, we take a gated‐mechanism to generate response word by word from dialogue history, vocabulary, or KB. Experiments show that the proposed model achieves superior results compared with state‐of‐the‐art models on several public datasets. Further analysis demonstrates the effectiveness of the proposed method and the robustness of the model in the case of an incomplete training set.

Developing efficient motion policies for multiagents is a challenge in a decentralized dynamic situation, where each agent plans its own paths without knowing the policies of the other agents involved. This paper presents an efficient learning-based motion planning method for multiagent systems. It adopts the framework of multiagent deep deterministic policy gradient (MADDPG) to directly map partially observed information to motion commands for multiple agents. To improve the efficiency of MADDPG in sample utilization, so as to train more brilliant agents that can adapt to more complex environments, a strategy named mixed experience (ME) is introduced to MADDPG, and this has led to our proposed ME-MADDPG algorithm. The novel ME strategy can be embodied into three specific mechanisms: (1) an artificial potential field-based sample generator to produce high-quality samples in the early training stage; (2) a dynamic mixed sampling strategy to mix the training data from different sources with a variable proportion; (3) a delayed learning skill to stabilize the training of the multiple agents. A series of experiments have been conducted to verify the performance of the proposed ME-MADDPG algorithm, and it has been demonstrated that, compared with MADDPG, the proposed algorithm can significantly improve the convergence speed and convergence effect in the training process, and it has also shown better efficiency and better adaptability in complex dynamic environments while it is used for multiagent motion planning applications.

Parameter learning is an important aspect of learning in Bayesian networks. Although the maximum likelihood algorithm is often effective, it suffers from overfitting when there is insufficient data. To address this, prior distributions of model parameters are often imposed. When training a Bayesian network, the parameters of the network are optimized to fit the data. However, imposing prior distributions can reduce the fitness between parameters and data. Therefore, a trade-off is needed between fitting and overfitting. In this study, a new algorithm, named MiniMax Fitness (MMF) is developed to address this problem. The method includes three main steps. First, the maximum a posterior estimation that combines data and prior distribution is derived. Then, the hyper-parameters of the prior distribution are optimized to minimize the fitness between posterior estimation and data. Finally, the order of posterior estimation is checked and adjusted to match the order of the statistical counts from the data. In addition, we introduce an improved constrained maximum entropy method, named Prior Free Constrained Maximum Entropy (PF-CME), to facilitate parameter learning when domain knowledge is provided. Experiments show that the proposed methods outperforms most of existing parameter learning methods.

In this paper, we demonstrate how an existing deep learning architecture for automatically lip reading individuals can be adapted it so that it can be made speaker independent, and by doing so, improved accuracies can be achieved on a variety of different speakers. The architecture itself is multi-layered consisting of a convolutional neural network, but if we are to apply an initial edge detection-based stage to pre-process the image inputs so that only the contours are required, the architecture can be made to be less speaker favourable. The neural network architecture achieves good accuracy rates when trained and tested on some of the same speakers in the ”overlapped speakers” phase of simulations, where word error rates of just 1.3% and 0.4% are achieved when applied to two individual speakers respectively, as well as character error rates of 0.6% and 0.3%. The ”unseen speakers” phase fails to achieve as good an accuracy, with greater recorded word error rates of 20.6% and 17.0% when tested on the two speakers with character error rates of 11.5% and 8.3%. The variation in size and colour of different people’s lips will result in different outputs at the convolution layer of a convolutional neural network as the output depends on the pixel intensity of the red, green and blue channels of an input image so a convolutional neural network will naturally favour the observations of the individual whom the network was tested on. This paper proposes an initial ”contour mapping stage” which makes all inputs uniform so that the system can be speaker independent.

One of the essential problems on Bayesian networks (BNs) is parameter learning. When purely data-driven methods fail to work, incorporating supplemental information, like expert judgments, can improve the learning of BN parameters. In practice, expert judgments are provided and transformed into qualitative parameter constraints. Moreover, prior distributions of BN parameters are also useful information. In this paper we propose a Bayesian approach to learn parameters from small datasets by integrating both parameter constraints and prior distributions. First, the feasible parameter region is derived from constraints. Then, using the prior distribution, a posterior distribution over the feasible region is developed based on the Bayes theorem. Finally, the parameter estimations are taken as the mean values of the posterior distribution. Learning experiments on standard BNs reveal that the proposed method outperforms most of the existing methods.

In this survey paper, the-state-of-the-art of the optimal structure design of Multilayer Feedforward Neural Network (MFNN) for pattern recognition is reviewed. Special emphasis is laid on the scale-limited MFNN and the internal representation and decision boundary-based design methodologies. A comprehensively comparative study of the main characteristics of each method is presented. Also, future research directions are outlined.