PupilNet: Principled Face Alignment with Recurrent Attention – In this paper, we propose an attention-based model for visual attention. Previous work explicitly uses the attention mechanism to learn attention maps instead of a feature. However, previous studies focused on the visual attention mechanism which was not explored. Here, we explore the visual attention mechanism using a feature. A key assumption in previous attention-based approaches is that visual attention consists of learning two representations of visual features, and each of these representations may be used in different tasks. We propose a novel visual attention mechanism that learns attention maps by visualizing the task at hand and using a deep learning algorithm to adaptively update the representations of visual features. Experimental results using a new state-of-the-art visual attention system, the CNN-D+R-DI, demonstrate that the proposed method achieves competitive recognition rate of 90.9 per cent (95%) on the MNIST dataset.

This paper proposes a new method for extracting feature representations using probabilistic model representations. It assumes that the model is parametrically parametrized, and that the input data is modeled as a probabilistic data structure. We show that with a strong inference structure, we obtain a probabilistic representation of the model and that one can use this representation to provide representations with natural visualizations, such as semantic annotations and informative representations. The method is efficient and can be used for image classification and image captioning applications. Experimental results show that our method outperforms the state-of-the-art classification methods by over 70% accuracy while being much more accurate.

Learning LSTM Compressible Models with the K-Lipschitz Transform

Neural Networks for Activity Recognition in Mobile Social Media

PupilNet: Principled Face Alignment with Recurrent Attention

Deep Learning with Risk-Aware Adaptation for Driver Test Count Prediction

Hierarchical Constraint Programming with Constraint ReasoningsThis paper proposes a new method for extracting feature representations using probabilistic model representations. It assumes that the model is parametrically parametrized, and that the input data is modeled as a probabilistic data structure. We show that with a strong inference structure, we obtain a probabilistic representation of the model and that one can use this representation to provide representations with natural visualizations, such as semantic annotations and informative representations. The method is efficient and can be used for image classification and image captioning applications. Experimental results show that our method outperforms the state-of-the-art classification methods by over 70% accuracy while being much more accurate.