Diabetic eye disease is one of the fastest growing causes of preventable blindness. With the advent of anti-VEGF (vascular endothelial growth factor) therapies, it has become increasingly important to detect center-involved diabetic macular edema. However, center-involved diabetic macular edema is diagnosed using optical coherence tomography (OCT), which is not generally available at screening sites because of cost and workflow constraints. Instead, screening programs rely on the detection of hard exudates as a proxy for DME on color fundus photographs, often resulting in high false positive or false negative calls. To improve the accuracy of DME screening, we trained a deep learning model to use color fundus photographs to predict DME grades derived from OCT exams. Our "OCT-DME" model had an AUC of 0.89 (95% CI: 0.87-0.91), which corresponds to a sensitivity of 85% at a specificity of 80%. In comparison, three retinal specialists had similar sensitivities (82-85%), but only half the specificity (45-50%, p<0.001 for each comparison with model). The positive predictive value (PPV) of the OCT-DME model was 61% (95% CI: 56-66%), approximately double the 36-38% by the retina specialists. In addition, we used saliency and other techniques to examine how the model is making its prediction. The ability of deep learning algorithms to make clinically relevant predictions that generally require sophisticated 3D-imaging equipment from simple 2D images has broad relevance to many other applications in medical imaging.

Pyro is a probabilistic programming language built on Python as a platform for developing advanced probabilistic models in AI research. To scale to large datasets and high-dimensional models, Pyro uses stochastic variational inference algorithms and probability distributions built on top of PyTorch, a modern GPU-accelerated deep learning framework. To accommodate complex or model-specific algorithmic behavior, Pyro leverages Poutine, a library of composable building blocks for modifying the behavior of probabilistic programs.

Convolutional neural networks have been successfully applied to various NLP tasks. However, it is not obvious whether they model different linguistic patterns such as negation, intensification, and clause compositionality to help the decision-making process. In this paper, we apply visualization techniques to observe how the model can capture different linguistic features and how these features can affect the performance of the model. Later on, we try to identify the model errors and their sources. We believe that interpreting CNNs is the first step to understand the underlying semantic features which can raise awareness to further improve the performance and explainability of CNN models.

We propose sequenced-replacement sampling (SRS) for training deep neural networks. The basic idea is to assign a fixed sequence index to each sample in the dataset. Once a mini-batch is randomly drawn in each training iteration, we refill the original dataset by successively adding samples according to their sequence index. Thus we carry out replacement sampling but in a batched and sequenced way. In a sense, SRS could be viewed as a way of performing "mini-batch augmentation". It is particularly useful for a task where we have a relatively small images-per-class such as CIFAR-100. Together with a longer period of initial large learning rate, it significantly improves the classification accuracy in CIFAR-100 over the current state-of-the-art results. Our experiments indicate that training deeper networks with SRS is less prone to over-fitting. In the best case, we achieve an error rate as low as 10.10%.

Often models that operate on source code consume ASTs by linearizing them (usually with a depth-first traversal) (Amodio et al., 2017; Liu et al., 2017; Li et al., 2017), but they can also be processed by deep learning models that take graphs as input, as in White et al. (2016) and Chen et al. (2018) who use Recursive Neural Networks (RveNNs) (Goller & Kuchler, 1996) on ASTs. RveNNs are models that operate on tree-topology graphs, and have been used extensively for language modeling (Socher et al., 2013) and on domains similar to source code, like mathematical expressions (Zaremba et al., 2014; Arabshahi et al., 2018). They can be considered a special case of Message Passing Neural Networks (MPNNs) in the framework of Gilmer et al. (2017): in this analogy RveNNs are to Belief Propagation as MPNNs are to Loopy Belief Propagation. They can also be considered a special case of Graph Networks in the framework of Battaglia et al. (2018). ASTs also serve as a natural basis for models that generate code as output, as in Maddison & Tarlow (2014), Yin & Neubig (2017), Rabinovich et al. (2017), Chen et al. (2018), and Brockschmidt et al. (2018).

The Convolutional Neural Network (CNN) architecture is increasingly being applied to new domains, such as malware detection, where it is able to learn malicious behavior from raw bytes extracted from executables. These architectures reach impressive performance with no feature engineering effort involved, but their robustness against active attackers is yet to be understood. Such malware detectors could face a new attack vector in the form of adversarial interference with the classification model. Existing evasion attacks intended to cause misclassification on test-time instances, which have been extensively studied for image classifiers, are not applicable because of the input semantics that prevents arbitrary changes to the binaries. This paper explores the area of adversarial examples for malware detection. By training an existing model on a production-scale dataset, we show that some previous attacks are less effective than initially reported, while simultaneously highlighting architectural weaknesses that facilitate new attack strategies for malware classification. Finally, we explore more generalizable attack strategies that increase the potential effectiveness of evasion attacks.

Stochastic variational inference is an established way to carry out approximate Bayesian inference for deep models. While there have been effective proposals for good initializations for loss minimization in deep learning, far less attention has been devoted to the issue of initialization of stochastic variational inference. We address this by proposing a novel layer-wise initialization strategy based on Bayesian linear models. The proposed method is extensively validated on regression and classification tasks, including Bayesian DeepNets and ConvNets, showing faster convergence compared to alternatives inspired by the literature on initializations for loss minimization.

Deep learning has shown high performances in various types of tasks from visual recognition to natural language processing, which indicates superior flexibility and adaptivity of deep learning. To understand this phenomenon theoretically, we develop a new approximation and estimation error analysis of deep learning with the ReLU activation for functions in a Besov space and its variant with mixed smoothness. The Besov space is a considerably general function space including the Holder space and Sobolev space, and especially can capture spatial inhomogeneity of smoothness. Through the analysis in the Besov space, it is shown that deep learning can achieve the minimax optimal rate and outperform any non-adaptive (linear) estimator such as kernel ridge regression, which shows that deep learning has higher adaptivity to the spatial inhomogeneity of the target function than other estimators such as linear ones. In addition to this, it is shown that deep learning can avoid the curse of dimensionality if the target function is in a mixed smooth Besov space. We also show that the dependency of the convergence rate on the dimensionality is tight due to its minimax optimality. These results support high adaptivity of deep learning and its superior ability as a feature extractor.

At present, the state-of-the-art computational models across a range of sequential data processing tasks, including language modeling, are based on recurrent neural network architectures. This paper begins with the observation that most research on developing computational models capable of processing sequential data fails to explicitly analyze the long distance dependencies (LDDs) within the datasets the models process. In this context, in this paper, we make five research contributions. First, we argue that a key step in modeling sequential data is to understand the characteristics of the LDDs within the data. Second, we present a method to compute and analyze the LDD characteristics of any sequential dataset, and demonstrate this method on a number of sequential datasets that are frequently used for model benchmarking. Third, based on the analysis of the LDD characteristics within the benchmarking datasets, we observe that LDDs are far more complex than previously assumed, and depend on at least four factors: (i) the number of unique symbols in a dataset, (ii) size of the dataset, (iii) the number of interacting symbols within an LDD, and (iv) the distance between the interacting symbols. Fourth, we verify these factors by using synthetic datasets generated using Strictly k-Piecewise (SPk) languages. We then demonstrate how SPk languages can be used to generate benchmarking datasets with varying degrees of LDDs. The advantage of these synthesized datasets being that they enable the targeted testing of recurrent neural architectures. Finally, we demonstrate how understanding the characteristics of the LDDs in a dataset can inform better hyper-parameter selection for current state-of-the-art recurrent neural architectures and also aid in understanding them...

As autonomy becomes prevalent in many applications, ranging from recommendation systems to fully autonomous vehicles, there is an increased need to provide safety guarantees for such systems. The problem is difficult, as these are large, complex systems which operate in uncertain environments, requiring data-driven machine-learning components. However, learning techniques such as Deep Neural Networks, widely used today, are inherently unpredictable and lack the theoretical foundations to provide strong assurance guarantees. We present a compositional approach for the scalable, formal verification of autonomous systems that contain Deep Neural Network components. The approach uses assume-guarantee reasoning whereby {\em contracts}, encoding the input-output behavior of individual components, allow the designer to model and incorporate the behavior of the learning-enabled components working side-by-side with the other components. We illustrate the approach on an example taken from the autonomous vehicles domain.