Abstract--Handling the tremendous amount of network data, produced by the explosive growth of mobile traffic volume, is becoming of main priority to achieve desired performance targets efficiently. Opportunistic communication such as Floating Content (FC), can be used to offload part of the cellular traffic volume to vehicular-to-vehicular communication (V2V), leaving to the infrastructure the task of coordinating the communication. Existing FC dimensioning approaches have limitations, mainly due to unrealistic assumptions and on a coarse partitioning of users, which results in over-dimensioning. Shaping the opportunistic communication area is a crucial task to achieve desired application performance efficiently. In this work, we propose a solution for this open challenge. In particular, the broadcasting areas called Anchor Zone (AZ), are selected via a deep learning approach to minimize communication resources achieving desired message availability. No assumption required to fit the classifier in both synthetic and real mobility. A numerical study is made to validate the effectiveness and efficiency of the proposed method. The predicted AZ configuration can achieve an accuracy of 89.7% within 98% of confidence level. By cause of the learning approach, the method performs even better in real scenarios, saving up to 27% of resources compared to previous work analytically modeled. I NTRODUCTION New offloading techniques to cope with the explosive growth in mobile traffic volumes, are a fundamental component of the next generation radio access network (5G). Part of the cellular traffic volume can be offloaded to vehicular-to- vehicular communication (V2V), leaving to the infrastructure the task of managing and coordinating the communication. In this context, of special interest are communication paradigms such as Floating Content (FC), an opportunistic communication scheme for the local dissemination of information [1]. FC as an infrastructure-less communication model, enables probabilistic contents storing in geographically constrained locations - denoted as Anchor Zone (AZ) - and over a limited amount of time based on the application requirements.

Deep generative models are capable of learning probability distributions over large, high-dimensional datasets such as images, video and natural language. Generative models trained on samples from p(x) ought to assign low likelihoods to out-of-distribution (OoD) samples from q(x), making them suitable for anomaly detection applications. We show that in practice, likelihood models are themselves susceptible to OoD errors, and even assign large likelihoods to images from other natural datasets. To mitigate these issues, we propose Generative Ensembles, a model-independent technique for OoD detection that combines density-based anomaly detection with uncertainty estimation. Our method outperforms ODIN and VIB baselines on image datasets, and achieves comparable performance to a classification model on the Kaggle Credit Fraud dataset. Knowing when a machine learning (ML) model is qualified to make predictions on an input is critical to safe deployment of ML technology in the real world.

By injecting adversarial examples into training data, the adversarial training method is promising for improving the robustness of deep learning models. However, most existing adversarial training approaches are based on a specific type of adversarial attack. It may not provide sufficiently representative samples from the adversarial domain, leading to a weak generalization ability on adversarial examples from other attacks. To scale to large datasets, perturbations on inputs to generate adversarial examples are usually crafted using fast single-step attacks. This work is mainly focused on the adversarial training with the single-step yet efficient FGSM adversary. In this scenario, it is difficult to train a model with great generalization due to the lack of representative adversarial samples, aka the samples are unable to accurately reflect the adversarial domain. To alleviate this problem, we propose a novel Adversarial Training with Domain Adaptation (ATDA) method. Our intuition is regarding adversarial training on FGSM adversary as a domain adaption task with limited number of target domain samples. The main idea is to learn a representation that is semantically meaningful and domain invariant on the clean domain as well as the adversarial domain. Empirical evaluations on Fashion-MNIST, SVHN, CIFAR-10 and CIFAR-100 demonstrate that ATDA can greatly improve the generalization of adversarial training and outperforms state-of-the-art methods on standard benchmark datasets.

In silico drug-target interaction (DTI) prediction is an important and challenging problem in biomedical research with a huge potential benefit to the pharmaceutical industry and patients. Most existing methods for DTI prediction including deep learning models generally have binary endpoints, which could be an oversimplification of the problem, and those methods are typically unable to handle cold-target problems, i.e., problems involving target protein that never appeared in the training set. Towards this, we contrived PADME (Protein And Drug Molecule interaction prEdiction), a framework based on Deep Neural Networks, to predict real-valued interaction strength between compounds and proteins. PADME takes both compound and protein information as inputs, so it is capable of solving cold-target (and cold-drug) problems. To our knowledge, we are the first to combine Molecular Graph Convolution (MGC) for compound featurization with protein descriptors for DTI prediction. We used multiple cross-validation split schemes and evaluation metrics to measure the performance of PADME on multiple datasets, including the ToxCast dataset, which we believe should be a standard benchmark for DTI problems, and PADME consistently dominates baseline methods. The results of a case study, which predicts the interactions between various compounds and androgen receptor (AR), suggest PADME's potential in drug development. The scalability of PADME is another advantage in the age of Big Data.

China Abstract: Multistep traffic forecasting on road networks is a crucial task in successful intelligent transportation system applications. To capture the complex non-stationary temporal dynamics and spatial dependency in multistep traffic-condition prediction, we propose a novel deep learning framework named attention graph convolutional sequence-to-sequence model (AGC-Seq2Seq). In the proposed deep learning framework, spatial and temporal dependencies are modeled through the Seq2Seq model and graph convolution network separately, and the attention mechanism along with a newly designed training method based on the Seq2Seq architecture is proposed to overcome the difficulty in multistep prediction and further capture the temporal heterogeneity of traffic pattern. We conduct numerical tests to compare AGC-Seq2Seq with other benchmark models using a real-world dataset. The results indicate that our model yields the best prediction performance in terms of various prediction error measures. Keywords: traffic forecasting; deep learning; attention mechanism; graph convolution; multistep prediction; sequence-to-sequence model 1. INTRODUCTION Automobile use has significantly increased in the past few decades owing to the steady development in both technology and economy. However, the increased automobile use has resulted in a series of social problems such as traffic congestion, traffic accidents, energy overconsumption, and carbon emissions (Gao et al., 2011). The intelligent transportation system (ITS) has been considered as a promising solution to improve transportation management and services (Qureshi and Abdullah, 2013; Lin et al., 2017).

ABSTRACT We consider the problem of localizing and segmenting individual teeth inside 3D Cone-Beam Computed Tomography (CBCT) images. To handle large image sizes we approach this task with a coarse-to-fine framework, where the whole volume is first analyzed as a 33-class semantic segmentation (adults have up to 32 teeth) in coarse resolution, followed by binary semantic segmentation of the cropped region of interest in original resolution. To improve the performance of the challenging 33-class segmentation, we first train the Coarse step model on a large weakly labeled dataset, then fine-tune it on a smaller precisely labeled dataset. The Fine step model is trained with precise labels only. Empirically, this framework yields precise teeth masks with low localization errors sufficient for many real-world applications.

Simulating the dynamic characteristics of a PN junction at the microscopic level requires solving the Poisson's equation at every time step. Solving at every time step is a necessary but time-consuming process when using the traditional finite difference (FDM) approach. Deep learning is a powerful technique to fit complex functions. In this work, deep learning is utilized to accelerate solving Poisson's equation in a PN junction. The role of the boundary condition is emphasized in the loss function to ensure a better fitting. The resulting I-V curve for the PN junction, using the deep learning solver presented in this work, shows a perfect match to the I-V curve obtained using the finite difference method, with the advantage of being 10 times faster at every time step.

The application to search ranking is one of the biggest machine learning success stories at Airbnb. Much of the initial gains were driven by a gradient boosted decision tree model. The gains, however, plateaued over time. This paper discusses the work done in applying neural networks in an attempt to break out of that plateau. We present our perspective not with the intention of pushing the frontier of new modeling techniques. Instead, ours is a story of the elements we found useful in applying neural networks to a real life product. Deep learning was steep learning for us. To other teams embarking on similar journeys, we hope an account of our struggles and triumphs will provide some useful pointers. Bon voyage!

WILSONVILLE, Ore.--(BUSINESS WIRE)--FLIR Systems, Inc. (NASDAQ: FLIR) today announced the FLIR Firefly camera family, the industry's first deep learning inference-enabled machine vision camera. The FLIR Firefly, which integrates the Intel Movidius Myriad 2 Vision Processing Unit (VPU), is designed for image analysis professionals using deep learning for more accurate decisions, and faster, easier system development. Traditional rules-based software is ideal for straightforward tasks such as barcode reading or checking a manufactured part against specifications. The FLIR Firefly combines a new, affordable machine vision platform with the power of deep learning to address complex and subjective problems such as recognizing faces or classifying the quality of a solar panel. The FLIR Firefly leverages the Intel Movidius Myriad 2 VPU's advanced capabilities in a compact and low-power camera, ideal for embedded and handheld systems.

Google is treading carefully when it comes to patient information, particularly as public scrutiny over data-collection rises. Last year, British regulators slapped DeepMind, another Alphabet AI lab, for testing an app that analyzed public medical records without telling patients that their information would be used like this. With the latest study, Google and its hospital partners insist their data is anonymous, secure and used with patient permission. Volchenboum said the company may have a more difficult time maintaining that data rigor if it expands to smaller hospitals and health-care networks.