In this paper we present a novel approach for extracting a Bag-of-Words (BoW) representation based on a Neural Network codebook. The conventional BoW model is based on a dictionary (codebook) built from elementary representations which are selected randomly or by using a clustering algorithm on a training dataset. A metric is then used to assign unseen elementary representations to the closest dictionary entries in order to produce a histogram. In the proposed approach, an autoencoder (AE) encompasses the role of both the dictionary creation and the assignment metric. The dimension of the encoded layer of the AE corresponds to the size of the dictionary and the output of its neurons represents the assignment metric. Experimental results for the continuous emotion prediction task on the AVEC 2017 audio dataset have shown an improvement of the Concordance Correlation Coefficient (CCC) from 0.225 to 0.322 for arousal dimension and from 0.244 to 0.368 for valence dimension relative to the conventional BoW version implemented in a baseline system.

In this work, we propose an introspection technique for deep neural networks that relies on a generative model to instigate salient editing of the input image for model interpretation. Such modification provides the fundamental interventional operation that allows us to obtain answers to counterfactual inquiries, i.e., what meaningful change can be made to the input image in order to alter the prediction. We demonstrate how to reveal interesting properties of the given classifiers by utilizing the proposed introspection approach on both the MNIST and the CelebA dataset.

Task 2 is a task on semantic similarity between clinical cases and discussions. For this task, we propose an approach based on language models and evaluate the impact on the results of different preprocessings and matching techniques. For task 3, we have developed an information extraction system yielding very encouraging results accuracy-wise. We have experimented two different approaches, one based on the exclusive use of neural networks, the other based on a linguistic analysis.

We describe a new technique which minimizes the amount of neurons in the hidden layer of a random recurrent neural network (rRNN) for time series prediction. Merging Takens-based attractor reconstruction methods with machine learning, we identify a mechanism for feature extraction that can be leveraged to lower the network size. We obtain criteria specific to the particular prediction task and derive the scaling law of the prediction error. The consequences of our theory are demonstrated by designing a Takens-inspired hybrid processor, which extends a rRNN with a priori designed delay external memory. Our hybrid architecture is therefore designed including both, real and virtual nodes. Via this symbiosis, we show performance of the hybrid processor by stabilizing an arrhythmic neural model. Thanks to our obtained design rules, we can reduce the stabilizing neural network's size by a factor of 15 with respect to a standard system.

We consider the problem of learning recurring convolutional patterns from data that are not necessarily real valued, such as binary or count-valued data. We cast the problem as one of learning a convolutional dictionary, subject to sparsity constraints, given observations drawn from a distribution that belongs to the canonical exponential family. We propose two general approaches towards its solution. The first approach uses the $\ell_0$ pseudo-norm to enforce sparsity and is reminiscent of the alternating-minimization algorithm for classical convolutional dictionary learning (CDL). The second approach, which uses the $\ell_1$ norm to enforce sparsity, generalizes to the exponential family the recently-shown connection between CDL and a class of ReLU auto-encoders for Gaussian observations. The two approaches can each be interpreted as an auto-encoder, the weights of which are in one-to-one correspondence with the parameters of the convolutional dictionary. Our key insight is that, unless the observations are Gaussian valued, the input fed into the encoder ought to be modified iteratively, and in a specific manner, using the parameters of the dictionary. Compared to the $\ell_0$ approach, once trained, the forward pass through the $\ell_1$ encoder computes sparse codes orders of magnitude more efficiently. We apply the two approaches to the unsupervised learning of the stimulus effect from neural spiking data acquired in the barrel cortex of mice in response to periodic whisker deflections. We demonstrate that they are both superior to generalized linear models, which rely on hand-crafted features.

Deep networks realize complex mappings that are often understood by their locally linear behavior at or around points of interest. For example, we use the derivative of the mapping with respect to its inputs for sensitivity analysis, or to explain (obtain coordinate relevance for) a prediction. One key challenge is that such derivatives are themselves inherently unstable. In this paper, we propose a new learning problem to encourage deep networks to have stable derivatives over larger regions. While the problem is challenging in general, we focus on networks with piecewise linear activation functions. Our algorithm consists of an inference step that identifies a region around a point where linear approximation is provably stable, and an optimization step to expand such regions. We propose a novel relaxation to scale the algorithm to realistic models. We illustrate our method with residual and recurrent networks on image and sequence datasets.

In this study, we proposed a convolutional neural network model for gender prediction using English Twitter text as input. Ensemble of proposed model achieved an accuracy at 0.8237 on gender prediction and compared favorably with the state-of-the-art performance in a recent author profiling task. We further leveraged the trained models to predict the gender labels from an HPV vaccine related corpus and identified gender difference in public perceptions regarding HPV vaccine. The findings are largely consistent with previous survey-based studies.

Deep neural networks (DNN) are able to successfully process and classify speech utterances. However, understanding the reason behind a classification by DNN is difficult. One such debugging method used with image classification DNNs is activation maximization, which generates example-images that are classified as one of the classes. In this work, we evaluate applicability of this method to speech utterance classifiers as the means to understanding what DNN "listens to". We trained a classifier using the speech command corpus and then use activation maximization to pull samples from the trained model. Then we synthesize audio from features using WaveNet vocoder for subjective analysis. We measure the quality of generated samples by objective measurements and crowd-sourced human evaluations. Results show that when combined with the prior of natural speech, activation maximization can be used to generate examples of different classes. Based on these results, activation maximization can be used to start opening up the DNN black-box in speech tasks.

Deep convolutional neural networks trained on large datsets have emerged as an intriguing alternative for compressing images and solving inverse problems such as denoising and compressive sensing. However, it has only recently been realized that even without training, convolutional networks can function as concise image models, and thus regularize inverse problems. In this paper, we provide further evidence for this finding by studying variations of convolutional neural networks that map few weight parameters to an image. The networks we consider only consist of convolutional operations, with either fixed or parameterized filters followed by ReLU non-linearities. We demonstrate that with both fixed and parameterized convolutional filters those networks enable representing images with few coefficients. What is more, the underparameterization enables regularization of inverse problems, in particular recovering an image from few observations. We show that, similar to standard compressive sensing guarantees, on the order of the number of model parameters many measurements suffice for recovering an image from compressive measurements. Finally, we demonstrate that signal recovery with a un-trained convolutional network outperforms standard l1 and total variation minimization for magnetic resonance imaging (MRI).

Knowledge graphs (KGs) typically contain temporal facts indicating relationships among entities at different times. Due to their incompleteness, several approaches have been proposed to infer new facts for a KG based on the existing ones-a problem known as KG completion. KG embedding approaches have proved effective for KG completion, however, they have been developed mostly for static KGs. Developing temporal KG embedding models is an increasingly important problem. In this paper, we build novel models for temporal KG completion through equipping static models with a diachronic entity embedding function which provides the characteristics of entities at any point in time. This is in contrast to the existing temporal KG embedding approaches where only static entity features are provided. The proposed embedding function is model-agnostic and can be potentially combined with any static model. We prove that combining it with SimplE, a recent model for static KG embedding, results in a fully expressive model for temporal KG completion. Our experiments indicate the superiority of our proposal compared to existing baselines.