We describe two architectures that generalize convolutional neural networks (CNNs) for the processing of signals supported on graphs. The selection graph neural network (GNN) replaces linear time invariant filters with linear shift invariant graph filters to generate convolutional features and reinterprets pooling as a possibly nonlinear subsampling stage where nearby nodes pool their information in a set of preselected sample nodes. A key component of the architecture is to remember the position of sampled nodes to permit computation of convolutional features at deeper layers. The aggregation GNN diffuses the signal through the graph and stores the sequence of diffused components observed by a designated node. This procedure effectively aggregates all components into a stream of information having temporal structure to which the convolution and pooling stages of regular CNNs can be applied. A multinode version of aggregation GNNs is further introduced for operation in large scale graphs. An important property of selection and aggregation GNNs is that they reduce to conventional CNNs when particularized to time signals reinterpreted as graph signals in a circulant graph. Comparative numerical analyses are performed in a synthetic source localization application. Performance is evaluated for a text category classification problem using word proximity networks. Multinode aggregation GNNs are consistently the best performing GNN architecture.

Computer vision performances have been significantly improved in recent years by Convolutional Neural Networks(CNN). Currently, applications using CNN algorithms are deployed mainly on general purpose hardwares, such as CPUs, GPUs or FPGAs. However, power consumption, speed, accuracy, memory footprint, and die size should all be taken into consideration for mobile and embedded applications. Domain Specific Architecture (DSA) for CNN is the efficient and practical solution for CNN deployment and implementation. We designed and produced a 28nm Two-Dimensional CNN-DSA accelerator with an ultra power-efficient performance of 9.3TOPS/Watt and with all processing done in the internal memory instead of outside DRAM. It classifies 224x224 RGB image inputs at more than 140fps with peak power consumption at less than 300mW and an accuracy comparable to the VGG benchmark. The CNN-DSA accelerator is reconfigurable to support CNN model coefficients of various layer sizes and layer types, including convolution, depth-wise convolution, short-cut connections, max pooling, and ReLU. Furthermore, in order to better support real-world deployment for various application scenarios, especially with low-end mobile and embedded platforms and MCUs (Microcontroller Units), we also designed algorithms to fully utilize the CNN-DSA accelerator efficiently by reducing the dependency on external accelerator computation resources, including implementation of Fully-Connected (FC) layers within the accelerator and compression of extracted features from the CNN-DSA accelerator. Live demos with our CNN-DSA accelerator on mobile and embedded systems show its capabilities to be widely and practically applied in the real world.

Whilst deep neural networks have shown great empirical success, there is still much work to be done to understand their theoretical properties. In this paper, we study the relationship between Gaussian processes with a recursive kernel definition and random wide fully connected feedforward networks with more than one hidden layer. We show that, under broad conditions, as we make the architecture increasingly wide, the implied random function converges in distribution to a Gaussian process, formalising and extending existing results by Neal (1996) to deep networks. To evaluate convergence rates empirically, we use maximum mean discrepancy. We then exhibit situations where existing Bayesian deep networks are close to Gaussian processes in terms of the key quantities of interest. Any Gaussian process has a flat representation. Since this behaviour may be undesirable in certain situations we discuss ways in which it might be prevented.

Dialogue management (DM) decides the next action of a dialogue system according to the current dialogue state, and thus plays a central role in task-oriented dialogue systems. Since dialogue management requires to have access to not only local utterances, but also the global semantics of the entire dialogue session, modeling the long-range history information is a critical issue. To this end, we propose a novel Memory-Augmented Dialogue management model (MAD) which employs a memory controller and two additional memory structures, i.e., a slot-value memory and an external memory. The slot-value memory tracks the dialogue state by memorizing and updating the values of semantic slots (for instance, cuisine, price, and location), and the external memory augments the representation of hidden states of traditional recurrent neural networks through storing more context information. To update the dialogue state efficiently, we also propose slot-level attention on user utterances to extract specific semantic information for each slot. Experiments show that our model can obtain state-of-the-art performance and outperforms existing baselines.

With the development of deep learning, Deep Metric Learning (DML) has achieved great improvements in face recognition. Specifically, the widely used softmax loss in the training process often bring large intra-class variations, and feature normalization is only exploited in the testing process to compute the pair similarities. To bridge the gap, we impose the intra-class cosine similarity between the features and weight vectors in softmax loss larger than a margin in the training step, and extend it from four aspects. First, we explore the effect of a hard sample mining strategy. To alleviate the human labor of adjusting the margin hyper-parameter, a self-adaptive margin updating strategy is proposed. Then, a normalized version is given to take full advantage of the cosine similarity constraint. Furthermore, we enhance the former constraint to force the intra-class cosine similarity larger than the mean inter-class cosine similarity with a margin in the exponential feature projection space. Extensive experiments on Labeled Face in the Wild (LFW), Youtube Faces (YTF) and IARPA Janus Benchmark A (IJB-A) datasets demonstrate that the proposed methods outperform the mainstream DML methods and approach the state-of-the-art performance.

In this paper we propose a new conditional GAN for image captioning that enforces semantic alignment between images and captions through a co-attentive discriminator and a context-aware LSTM sequence generator. In order to train these sequence GANs, we empirically study two algorithms: Self-critical Sequence Training (SCST) and Gumbel Straight-Through. Both techniques are confirmed to be viable for training sequence GANs. However, SCST displays better gradient behavior despite not directly leveraging gradients from the discriminator. This ensures a stronger stability of sequence GANs training and ultimately produces models with improved results under human evaluation. Automatic evaluation of GAN trained captioning models is an open question. To remedy this, we introduce a new semantic score with strong correlation to human judgement. As a paradigm for evaluation, we suggest that the generalization ability of the captioner to Out of Context (OOC) scenes is an important criterion to assess generalization and composition. To this end, we propose an OOC dataset which, combined with our automatic metric of semantic score, is a new benchmark for the captioning community to measure the generalization ability of automatic image captioning. Under this new OOC benchmark, and on the traditional MSCOCO dataset, our models trained with SCST have strong performance in both semantic score and human evaluation.

Text generation is a crucial task in NLP. Recently, several adversarial generative models have been proposed to improve the exposure bias problem in text generation. Though these models gain great success, they still suffer from the problems of reward sparsity and mode collapse. In order to address these two problems, in this paper, we employ inverse reinforcement learning (IRL) for text generation. Specifically, the IRL framework learns a reward function on training data, and then an optimal policy to maximum the expected total reward. Similar to the adversarial models, the reward and policy function in IRL are optimized alternately. Our method has two advantages: (1) the reward function can produce more dense reward signals. (2) the generation policy, trained by "entropy regularized" policy gradient, encourages to generate more diversified texts. Experiment results demonstrate that our proposed method can generate higher quality texts than the previous methods.

Language recognition system is typically trained directly to optimize classification error on the target language labels, without using the external, or meta-information in the estimation of the model parameters. However labels are not independent of each other, there is a dependency enforced by, for example, the language family, which affects negatively on classification. The other external information sources (e.g. audio encoding, telephony or video speech) can also decrease classification accuracy. In this paper, we attempt to solve these issues by constructing a deep hierarchical neural network, where different levels of meta-information are encapsulated by attentive prediction units and also embedded into the training progress. The proposed method learns auxiliary tasks to obtain robust internal representation and to construct a variant of attentive units within the hierarchical model. The final result is the structural prediction of the target language and a closely related language family. The algorithm reflects a "staircase" way of learning in both its architecture and training, advancing from the fundamental audio encoding to the language family level and finally to the target language level. This process not only improves generalization but also tackles the issues of imbalanced class priors and channel variability in the deep neural network model. Our experimental findings show that the proposed architecture outperforms the state-of-the-art i-vector approaches on both small and big language corpora by a significant margin.

Abstract--This short paper describes our solution to the 2018 IEEE World Congress on Computational Intelligence One-Minute Gradual-Emotional Behavior Challenge, whose goal was to estimate continuous arousal and valence values from short videos. We designed four base regression models using visual and audio features, and then used a spectral approach to fuse them to obtain improved performance. (IEEE WCCI 2018). The dataset was composed of 420 relatively long emotion videos with an average length of 1 minute, collected from a variety of Youtube channels. Videos were separated into clips based on utterances, and each utterance's valence and arousal levels were annotated by at least five independent subjects using the Amazon Mechanical Turk tool.

Latent factor models with implicit feedback typically treat unobserved user-item interactions (i.e. missing information) as negative feedback. This is frequently done either through negative sampling (point-wise loss) or with a ranking loss function (pair- or list-wise estimation). Since a zero preference recommendation is a valid solution for most common objective functions, regarding unknown values as actual zeros results in users having a zero preference recommendation for most of the available items. In this paper we propose a novel objective function, the Missing Information Loss (MIL) function, that explicitly forbids treating unobserved user-item interactions as positive or negative feedback. We apply this loss to a user--based Denoising Autoencoder and compare it with other known objective functions such as cross-entropy (both point-- and pair--wise) or the recently proposed multinomial log-likelihood. The MIL function achieves best results in ranking-aware metrics when applied to the Movielens-20M and Netflix datasets, slightly above those obtained with cross-entropy in point-wise estimation. Furthermore, such a competitive performance is obtained while recommending popular items less frequently, a valuable feature for Recommender Systems with a large catalogue of products.