Deep Neural Networks (DNNs) have recently been shown vulnerable to adversarial attacks in which the input examples are perturbed to fool these DNNs towards confidence reduction and (targeted or random) misclassification. In this paper, we demonstrate that how an efficient quantization technique can be leveraged to increase the robustness of a given DNN against adversarial attacks. We present two quantization-based defense mechanisms, namely Constant Quantization (CQ) and Variable Quantization (VQ), applied at the input to increase the robustness of DNNs. In CQ, the intensity of the input pixel is quantized according to the number of quantization levels. While in VQ, the quantization levels are recursively updated during the training phase, thereby providing a stronger defense mechanism. We apply our techniques on the Convolutional Neural Networks (CNNs, a particular type of DNN which is heavily used in vision-based applications) against adversarial attacks from the open-source Cleverhans library. Our experimental results show 1%-5% increase in the adversarial accuracy for MNIST and 0%-2.4% increase in the adversarial accuracy for CIFAR10.

Conditional random fields (CRFs) have been shown to be one of the most successful approaches to sequence labeling. Various linear-chain neural CRFs (NCRFs) are developed to implement the non-linear node potentials in CRFs, but still keeping the linear-chain hidden structure. In this paper, we propose NCRF transducers, which consists of two RNNs, one extracting features from observations and the other capturing (theoretically infinite) long-range dependencies between labels. Different sequence labeling methods are evaluated over POS tagging, chunking and NER (English, Dutch). Experiment results show that NCRF transducers achieve consistent improvements over linear-chain NCRFs and RNN transducers across all the four tasks, and can improve state-of-the-art results.

Recent studies have introduced end-to-end TTS, which integrates the production of context and acoustic features in statistical parametric speech synthesis. As a result, a single neural network replaced laborious feature engineering with automated feature learning. However, little is known about what types of context information end-to-end TTS extracts from text input before synthesizing speech, and the previous knowledge about context features is barely utilized. In this work, we first point out the model similarity between end-to-end TTS and parametric TTS. Based on the similarity, we evaluate the quality of encoder outputs from an end-to-end TTS system against eight criteria that are derived from a standard set of context information used in parametric TTS. We conduct experiments using an evaluation procedure that has been newly developed in the machine learning literature for quantitative analysis of neural representations, while adapting it to the TTS domain. Experimental results show that the encoder outputs reflect both linguistic and phonetic contexts, such as vowel reduction at phoneme level, lexical stress at syllable level, and part-of-speech at word level, possibly due to the joint optimization of context and acoustic features.

We propose a new optimization method for training feed-forward neural networks. By rewriting the activation function as an equivalent proximal operator, we approximate a feed-forward neural network by adding the proximal operators to the objective function as penalties, hence we call the lifted proximal operator machine (LPOM). LPOM is block multi-convex in all layer-wise weights and activations. This allows us to use block coordinate descent to update the layer-wise weights and activations in parallel. Most notably, we only use the mapping of the activation function itself, rather than its derivatives, thus avoiding the gradient vanishing or blow-up issues in gradient based training methods. So our method is applicable to various non-decreasing Lipschitz continuous activation functions, which can be saturating and non-differentiable. LPOM does not require more auxiliary variables than the layer-wise activations, thus using roughly the same amount of memory as stochastic gradient descent (SGD) does. We further prove the convergence of updating the layer-wise weights and activations. Experiments on MNIST and CIFAR-10 datasets testify to the advantages of LPOM.

This paper investigates whether learning contingency-awareness and controllable aspects of an environment can lead to better exploration in reinforcement learning. To investigate this question, we consider an instantiation of this hypothesis evaluated on the Arcade Learning Element (ALE). In this study, we develop an attentive dynamics model (ADM) that discovers controllable elements of the observations, which are often associated with the location of the character in Atari games. The ADM is trained in a self-supervised fashion to predict the actions taken by the agent. The learned contingency information is used as a part of the state representation for exploration purposes. We demonstrate that combining A2C with count-based exploration using our representation achieves impressive results on a set of notoriously challenging Atari games due to sparse rewards. For example, we report a state-of-the-art score of >6600 points on Montezuma's Revenge without using expert demonstrations, explicit high-level information (e.g., RAM states), or supervised data. Our experiments confirm that indeed contingency-awareness is an extremely powerful concept for tackling exploration problems in reinforcement learning and opens up interesting research questions for further investigations.

A ubiquitous task in processing electronic medical data is the assignment of standardized codes representing diagnoses and/or procedures to free-text documents such as medical reports. This is a difficult natural language processing task that requires parsing long, heterogeneous documents and selecting a set of appropriate codes from tens of thousands of possibilities---many of which have very few positive training samples. We present a deep learning system that advances the state of the art for the MIMIC-III dataset, achieving a new best micro F1-measure of 55.85\%, significantly outperforming the previous best result (Mullenbach et al. 2018). We achieve this through a number of enhancements, including two major novel contributions: multi-view convolutional channels, which effectively learn to adjust kernel sizes throughout the input; and attention regularization, mediated by natural-language code descriptions, which helps overcome sparsity for thousands of uncommon codes. These and other modifications are selected to address difficulties inherent to both automated coding specifically and deep learning generally. Finally, we investigate our accuracy results in detail to individually measure the impact of these contributions and point the way towards future algorithmic improvements.

When observing the actions of others, humans carry out inferences about why the others acted as they did, and what this implies about their view of the world. Humans also use the fact that their actions will be interpreted in this manner when observed by others, allowing them to act informatively and thereby communicate efficiently with others. Although learning algorithms have recently achieved superhuman performance in a number of two-player, zero-sum games, scalable multi-agent reinforcement learning algorithms that can discover effective strategies and conventions in complex, partially observable settings have proven elusive. We present the Bayesian action decoder (BAD), a new multi-agent learning method that uses an approximate Bayesian update to obtain a public belief that conditions on the actions taken by all agents in the environment. Together with the public belief, this Bayesian update effectively defines a new Markov decision process, the public belief MDP, in which the action space consists of deterministic partial policies, parameterised by deep neural networks, that can be sampled for a given public state. It exploits the fact that an agent acting only on this public belief state can still learn to use its private information if the action space is augmented to be over partial policies mapping private information into environment actions. The Bayesian update is also closely related to the theory of mind reasoning that humans carry out when observing others' actions. We first validate BAD on a proof-of-principle two-step matrix game, where it outperforms traditional policy gradient methods. We then evaluate BAD on the challenging, cooperative partial-information card game Hanabi, where in the two-player setting the method surpasses all previously published learning and hand-coded approaches.

Knowledge graph embedding aims at modeling entities and relations with low-dimensional vectors. Most previous methods require that all entities should be seen during training, which is unpractical for real-world knowledge graphs with new entities emerging on a daily basis. Recent efforts on this issue suggest training a neighborhood aggregator in conjunction with the conventional entity and relation embeddings, which may help embed new entities inductively via their existing neighbors. However, their neighborhood aggregators neglect the unordered and unequal natures of an entity's neighbors. To this end, we summarize the desired properties that may lead to effective neighborhood aggregators. We also introduce a novel aggregator, namely, Logic Attention Network (LAN), which addresses the properties by aggregating neighbors with both rules- and network-based attention weights. By comparing with conventional aggregators on two knowledge graph completion tasks, we experimentally validate LAN's superiority in terms of the desired properties.

This study introduces a stochastic predictive-coding RNN model that can learn to extract probabilistic structures hidden in fluctuating temporal patterns by dynamically changing the uncertainty of latent variables. The learning process of the model involves maximizing the lower bound on the marginal likelihood of the sequential data, which consists of two terms. The first one is the expectation of prediction errors and the second one is the divergence of the prior and the approximated posterior. The main focus in the current study is to examine how weighting of the second term during learning affects the way of internally representing the uncertainty hidden in the sequence data. The simulation experiment on learning a simple probabilistic finite state machine demonstrates that the estimation of uncertainty in the latent variable approaches zero at each time step and that the network imitates the probabilistic structure of the target sequences by developing deterministic chaos in the case of the high weighting. On the contrary, in the case of the low weighting, the estimate of uncertainty increases significantly because of developing a random process in the network. The analysis shows that generalization in learning is most successful between these two extremes. Qualitatively, the same property has been observed in a trail of learning more complex sequence data consisting of probabilistic transitions between a set of hand-drawn primitive patterns using the model extended with hierarchy.

Abstract: Attention mechanisms have been around since at least 2014, and they have been driving the state of the art in NLP (natural language processing) ever since. Now we are at the end of 2018, and yet it is still very uncommon to hear about their use in enterprise environments. This talk will explain attention mechanisms and how they fit into the deep learning landscape. It will show you the different types of attention mechanisms and examples of them being applied to several different problems. Finally, you will see how you might apply them yourself in models build with Deeplearning4J. Speaker: Paul Dubs (Paul Dubs IT Consulting) Paul holds a Master of Science in Computer Science from TU Darmstadt and has over a decade of professional software engineering experience.