When generating adversarial examples to attack deep neural networks (DNNs), $\ell_p$ norm of the added perturbation is usually used to measure the similarity between original image and adversarial example. However, such adversarial attacks may fail to capture key infomation hidden in the input. This work develops a more general attack model i.e., the structured attack that explores group sparsity in adversarial perturbations by sliding a mask through images aiming for extracting key structures. An ADMM (alternating direction method of multipliers)-based framework is proposed that can split the original problem into a sequence of analytically solvable subproblems and can be generalized to implement other state-of-the-art attacks. Strong group sparsity is achieved in adversarial perturbations even with the same level of distortion in terms of $\ell_p$ norm as the state-of-the-art attacks. Extensive experimental results on MNIST, CIFAR-10 and ImageNet show that our attack could be much stronger (in terms of smaller $\ell_0$ distortion) than the existing ones, and its better interpretability from group sparse structures aids in uncovering the origins of adversarial examples.

The use of machine learning (ML) is on the rise in many sectors of software development, and automotive software development is no different. In particular, Advanced Driver Assistance Systems (ADAS) and Automated Driving Systems (ADS) are two areas where ML plays a significant role. In automotive development, safety is a critical objective, and the emergence of standards such as ISO 26262 has helped focus industry practices to address safety in a systematic and consistent way. Unfortunately, these standards were not designed to accommodate technologies such as ML or the type of functionality that is provided by an ADS and this has created a conflict between the need to innovate and the need to improve safety. In this report, we take steps to address this conflict by doing a detailed assessment and adaption of ISO 26262 for ML, specifically in the context of supervised learning. First we analyze the key factors that are the source of the conflict. Then we assess each software development process requirement (Part 6 of ISO 26262) for applicability to ML. Where there are gaps, we propose new requirements to address the gaps. Finally we discuss the application of this adapted and extended variant of Part 6 to ML development scenarios.

Deep neural networks have gained tremendous popularity in last few years. They have been applied for the task of classification in almost every domain. Despite the success, deep networks can be incredibly slow to train for even moderate sized models on sufficiently large datasets. Additionally, these networks require large amounts of data to be able to generalize. The importance of speeding up convergence, and generalization in deep networks can not be overstated. In this work, we develop an optimization algorithm based on generalized-optimal updates derived from minibatches that lead to faster convergence. Towards the end, we demonstrate on two benchmark datasets that the proposed method achieves two orders of magnitude speed up over traditional back-propagation, and is more robust to noise/over-fitting.

Recurrent neural networks (RNNs) are temporal networks and cumulative in nature that have shown promising results in various natural language processing tasks. Despite their success, it still remains a challenge to understand their hidden behavior. In this work, we analyze and interpret the cumulative nature of RNN via a proposed technique named as Layer-wIse-Semantic-Accumulation (LISA) for explaining decisions and detecting the most likely (i.e., saliency) patterns that the network relies on while decision making. We demonstrate (1) LISA: "How an RNN accumulates or builds semantics during its sequential processing for a given text example and expected response" (2) Example2pattern: "How the saliency patterns look like for each category in the data according to the network in decision making". We analyse the sensitiveness of RNNs about different inputs to check the increase or decrease in prediction scores and further extract the saliency patterns learned by the network. We employ two relation classification datasets: SemEval 10 Task 8 and TAC KBP Slot Filling to explain RNN predictions via the LISA and example2pattern.

Answer triggering is the task of selecting the best-suited answer for a given question from a set of candidate answers if exists. In this paper, we present a hybrid deep learning model for answer triggering, which combines several dependency graph based alignment features, namely graph edit distance, graph-based similarity and dependency graph coverage, with dense vector embeddings from a Convolutional Neural Network (CNN). Our experiments on the WikiQA dataset show that such a combination can more accurately trigger a candidate answer compared to the previous state-of-the-art models. Comparative study on WikiQA dataset shows 5.86% absolute F-score improvement at the question level.

Image classification is the task of recognizing objects or patterns on an image and assign a category to classify the image based on a pool of given categories/classes. There are a lot of different sub fields like recognizing multiple objects on a single image or even localize the objects within the image. In this post, I want to share with you my experience competing in my first image classification challenge as well as using the Deep Learning Studio(DLS) platform. DLS provides at a high abstraction level, a graphical interface to build and train deep neural networks based on backends like keras / tensorflow / theano and others, for free with a desktop or cloud application. If you have never heard or used DLS go check out this post by Rajat or try it out at DeepCognition.ai.

The next time you sit down to watch a movie, the algorithm behind your streaming service might recommend a blockbuster that was written by AI, performed by robots, and animated and rendered by a deep learning algorithm. An AI algorithm may have even read the script and suggested the studio buy the rights. It's easy to think that technology like algorithms and robots will make the film industry go the way of the factory worker and the customer service rep, and argue that artistic filmmaking is in its death throes. Just like computers made it so animators didn't have to draw every frame by hand, advanced algorithms can automatically render advanced visual effects. In both cases, the animator didn't lose their job.

Tensorflow is a low-level deep learning package which requires users to deal with many complicated elements to construct a successful model. However, tensorflow is also powerful for production that's why most companies choose tensorflow as their major platforms. On the other hand, Keras provides a user-friendly API to help users quickly build complicated deep learning models but it is not appropriate for making products. Can we build our models in Keras and output it to tensorflow compatiable format (Protocol Buffers .pb In this tutorial, I will show to how to make it step-by-step.

Auto-Keras is an open source software library for automated machine learning (AutoML). The ultimate goal of AutoML is to allow domain experts with limited data science or machine learning background easily accessible to deep learning models. Auto-Keras provides functions to automatically search for architecture and hyperparameters of deep learning models. Note: currently, Auto-Keras is only compatible with: Python 3.6. Here is a short example of using the package.

Hospital and radiology specialists will invest some $2 billion every year to deploy artificial intelligence technologies for medical imaging, Signify Research said, and the firm estimated that will happen by 2023. Signify's AI medical imaging category encompasses software for automated detection, quantification, decision support and diagnosis as well as machine learning. Many consultants and analyst houses, in fact, are projecting big growth for AI and related technologies. Consultancy IDC estimated that the worldwide spending on artificial intelligence and cognitive computing overall technologies will grow by 60 percent to $12.5 billion this year. By 2020, it will have climbed to $46 billion.