The Beatles were a huge cultural phenomenon. Their timeless music still resonates with people today, both young and old. In my humble opinion, they are the greatest band to have ever lived¹. Their songs are full of interesting lyrics and deep ideas. When you've seen beyond yourself Then you may find peace of mind is waiting there² However, the thing that made the Beatles great was their versatility.

Clinicians who order common diagnostic chest x-rays for patients have been sitting on a goldmine of unused prognostic information. The radiographs, used since the 19th century to detect specific abnormalities, could soon be repurposed to identify long-term mortality risk -- with a little help from machine learning. Using data from two large randomized trials, researchers have developed a convolutional neural network, called CXR-risk, that stratifies participants by all-cause mortality risk. They trained the artificial intelligence (AI) system with 85,000 x-rays and follow-up data from more than 40,000 individuals. Extracting information from single chest radiographs, the system found a graded association between risk score and mortality.

In the previous article in this series, "Diving into Machine Learning" we looked at some common approaches to machine learning, which is a subset of AI that provides systems with the ability to learn from data and improve over time without being explicitly programmed. In this latest article in our Enterprise AI series, we provide an overview of deep learning, which is a specific approach to the more general category of machine learning. As with other machine learning techniques, deep learning is an important building block for artificial intelligence in the enterprise. First, let's quickly review what machine learning is. Machine learning refers to the process of training a model, which is nothing more than a function that maps inputs (e.g., house size, customer preferences) to outputs (e.g., house value, new product recommendations).

AI refers to a category comprised of a whole host of technologies that mimic cognitive functions, traditionally ascribed to the human mind, from neutral networks, natural language processing (NLP), robotics, expert systems, to intelligent systems. In the commercial context today, the term is most often used to refer to speech and vision recognition systems, machine learning, and deep learning. Machine learning is a branch of AI where systems can "learn" from data, identify patterns, and make decisions with minimal human assistance. As Adeel Najmi, Senior Vice President, Products at One Network Enterprises puts it, "learning occurs when a machine takes the output, observes the accuracy of the output, and updates its own model so that better outputs will occur." Deep learning, a specialized form of machine learning, uses many layers of neural networks to classify images without extracting features from images.

Deep learning is not a magic bullet, but the techniques have shown to be highly effective in a large number of very challenging problem domains. This means that there is a ton of demand by businesses for effective deep learning practitioners. The problem is, how can the average business differentiate between good and bad practitioners? As a deep learning practitioner, how can you best demonstrate that you can deliver skillful deep learning models? In this post, you will discover the three levels of deep learning competence, and as a practitioner, what you must demonstrate at each level.

Artificial intelligence has passed a classic theory of mind test used with chimpanzees. The test probes the ability to perceive the world from the view of another individual and so AIs with this skill could be better at cooperating and communicating with humans and each other. AIs with theory of mind are key to building machines that can understand the world around them. In recent years, the skill has emerged in a robot whose memories are modelled on human brains and in DeepMind's ToM-net, which understands that others can have false beliefs.

We provide recovery guarantees for compressible signals that have been corrupted with noise and extend the framework introduced in [1] to defend neural networks against $\ell_0$-norm and $\ell_2$-norm attacks. Concretely, for a signal that is approximately sparse in some transform domain and has been perturbed with noise, we provide guarantees for accurately recovering the signal in the transform domain. We can then use the recovered signal to reconstruct the signal in its original domain while largely removing the noise. Our results are general as they can be directly applied to most unitary transforms used in practice and hold for both $\ell_0$-norm bounded noise and $\ell_2$-norm bounded noise. In the case of $\ell_0$-norm bounded noise, we prove recovery guarantees for Iterative Hard Thresholding (IHT) and Basis Pursuit (BP). For the case of $\ell_2$-norm bounded noise, we provide recovery guarantees for BP. These guarantees theoretically bolster the defense framework introduced in [1] for defending neural networks against adversarial inputs. Finally, we experimentally demonstrate this defense framework using both IHT and BP against the One Pixel Attack [21], Carlini-Wagner $\ell_0$ and $\ell_2$ attacks [3], Jacobian Saliency Based attack [18], and the DeepFool attack [17] on CIFAR-10 [12], MNIST [13], and Fashion-MNIST [27] datasets. This expands beyond the experimental demonstrations of [1].

There is some theoretical evidence that deep neural networks with multiple hidden layers have a potential for more efficient representation of multidimensional mappings than shallow networks with a single hidden layer. The question is whether it is possible to exploit this theoretical advantage for finding such representations with help of numerical training methods. Tests using prototypical problems with a known mean square minimum did not confirm this hypothesis. Minima found with the help of deep networks have always been worse than those found using shallow networks. This does not directly contradict the theoretical findings---it is possible that the superior representational capacity of deep networks is genuine while finding the mean square minimum of such deep networks is a substantially harder problem than with shallow ones.

Predicting the remaining useful life of machinery, infrastructure, or other equipment can facilitate preemptive maintenance decisions, whereby a failure is prevented through timely repair or replacement. This allows for a better decision support by considering the anticipated time-to-failure and thus promises to reduce costs. Here a common baseline may be derived by fitting a probability density function to past lifetimes and then utilizing the (conditional) expected remaining useful life as a prognostic. This approach finds widespread use in practice because of its high explanatory power. A more accurate alternative is promised by machine learning, where forecasts incorporate deterioration processes and environmental variables through sensor data. However, machine learning largely functions as a black-box method and its forecasts thus forfeit most of the desired interpretability. As our primary contribution, we propose a structured-effect neural network for predicting the remaining useful life which combines the favorable properties of both approaches: its key innovation is that it offers both a high accountability and the flexibility of deep learning. The parameters are estimated via variational Bayesian inferences. The different approaches are compared based on the actual time-to-failure for aircraft engines. This demonstrates the performance and superior interpretability of our method, while we finally discuss implications for decision support.

Regional rainfall-runoff modeling is an old but still mostly outstanding problem in Hydrological Sciences. The problem currently is that traditional hydrological models degrade significantly in performance when calibrated for multiple basins together instead of for a single basin alone. In this paper, we propose a novel, data-driven approach using Long Short-Term Memory networks (LSTMs), and demonstrate that under a'big data' paradigm, this is not necessarily the case. By training a single LSTM model on 531 basins from the CAMELS data set using meteorological time series data and static catchment attributes, we were able to significantly improve performance compared to a set of several different hydrological benchmark models. Our proposed approach not only significantly outperforms hydrological models that were calibrated regionally but also achieves better performance than hydrological models that were calibrated for each basin individually. Furthermore, we propose an adaption to the standard LSTM architecture, which we call an Entity-A ware-LSTM (EA-LSTM), that allows for learning, and embedding as a feature layer in a deep learning model, catchment similarities. We show that this learned catchment similarity corresponds well with what we would expect from prior hydrological understanding. 1 Introduction A longstanding problem in the Hydrological Sciences is about how to use one model, or one set of models, to provide spatially continuous hydrological simulations across large areas (e.g., regional, continental, global). This is the so-called regional modeling problem, and the central challenge is about how to extrapolate hydrologic information from one area to another - e.g., from gauged to ungauged watersheds, from instrumented to non-instrumented hillslopes, from areas with flux towers to areas without, etc. (Blöschl and Sivapalan, 1995). Often this is done using ancillary data (e.g.