A new machine learning tool that can detect whether emerging strains of the bacterium, Salmonella are more likely to cause dangerous bloodstream infections rather than food poisoning has been developed. The tool, created by a scientist at the Wellcome Sanger Institute and her collaborators at the University of Otago, New Zealand and the Helmholtz Institute for RNA-based Infection Research, a site of the Helmholtz Centre for Infection Research, Germany, greatly speeds up the process for identifying the genetic changes underlying new invasive types of Salmonella that are of public health concern. Reported today (8 May) in PLOS Genetics, the machine learning tool could be useful for flagging dangerous bacteria before they cause an outbreak, from hospital wards to a global scale. As the cost of genomic sequencing falls, scientists around the world are using genetics to better understand the bacteria causing infections, how diseases spread, how bacteria gain resistance to drugs, and which strains of bacteria may cause outbreaks. However, current methods to identify the genetic adaptations in emerging strains of bacteria behind an outbreak are time-consuming and often involve manually comparing the new strain to an older reference collection.

In Italy, 120 high school students helped solve a centuries-old problem: how to give researchers access to the Vatican Secret Archives, a massive collection of documents detailing the Vatican's activities as far back as the eighth century. That should look pretty great on their college applications. The shelves of the Vatican Secret Archives are about 85 kilometers (53 miles) long and house 35,000 volumes of catalogues. But the documents that researchers have scanned and uploaded take up less than an inch. That's because the Vatican seems to not have wanted to share the information.

In this seminar, Dr Barry Devlin lays the architectural foundation to enable you to take advantage of AI and IoT data in the context of data warehouses and lakes, operational systems, analytics and business intelligence. With the enormous growth of data from Internet of Things (IoT) devices and social media, as well as the reinvention of business through analytics and artificial intelligence (AI), the time has come to revamp your information architecture, expand your technology, and upskill your staff to support automated and augmented decision making and action taking in a fully digital business. A digital business combines the traditional physical environment and the modern digital world in transformative ways. In the process, it creates innovative opportunities for success as well as insidious threats to old ways of doing business. From finance to fashion, telecommunications to transport, businesses that reinvent their processes to become pervasively digitalized will survive and thrive; those that ignore this major shift will wither and die.

The future is also not all bleak as clearly there are many new jobs being created. The IDC sees 2.1m jobs being created in the US by technology augmenting workers in the world of CRM over the five years to 2021. AI has the potential to make many jobs more productive and remove a lot of the mundane and boring elements of jobs. A key conclusion worth noting is how to protect yourself from unemployment which is listed as education. The World Economic Forum estimates that those with GCSE2 have a 46% risk of losing their jobs whilst those with degrees are only at a 12% risk. What we are going to see are jobs being defragmented and the tasks that can be automated being undertaken by machines and tasks that requiring humans being consolidated into new roles. In fact the World Economic Forum's Future of Jobs 2016 estimates that 65% of primary school children will end up working in jobs that don't exist yet. Part of the reason for the variation in predictions comes form the fact that there are a number of scenarios being considered.

We introduce a new neural network model, together with a tractable and monotone online learning algorithm. Our model describes feed-forward networks for classification, with one output node for each class. The only nonlinear operation is rectification using a ReLU function with a bias. However, there is a rectifier on every edge rather than at the nodes of the network. There are also weights, but these are positive, static, and associated with the nodes. Our "rectified wire" networks are able to represent arbitrary Boolean functions. Only the bias parameters, on the edges of the network, are learned. Another departure in our approach, from standard neural networks, is that the loss function is replaced by a constraint. This constraint is simply that the value of the output node associated with the correct class should be zero. Our model has the property that the exact norm-minimizing parameter update, required to correctly classify a training item, is the solution to a quadratic program that can be computed with a few passes through the network. We demonstrate a training algorithm using this update, called sequential deactivation (SDA), on MNIST and some synthetic datasets. Upon adopting a natural choice for the nodal weights, SDA has no hyperparameters other than those describing the network structure. Our experiments explore behavior with respect to network size and depth in a family of sparse expander networks.

Deep Nets: What have they ever done for Vision? This is an opinion paper about the strengths and weaknesses of Deep Nets. They are at the center of recent progress on Artificial Intelligence and are of growing importance in Cognitive Science and Neuroscience since they enable the development of computational models that can deal with a large range of visually realistic stimuli and visual tasks. They have clear limitations but they also have enormous successes. There is also gradual, though incomplete, understanding of their inner workings. It seems unlikely that Deep Nets in their current form will be the best long-term solution either for building general purpose intelligent machines or for understanding the mind/brain, but it is likely that many aspects of them will remain. At present Deep Nets do very well on specific types of visual tasks and on specific benchmarked datasets. But Deep Nets are much less general purpose, flexible, and adaptive than the human visual system. Moreover, methods like Deep Nets may run into fundamental difficulties when faced with the enormous complexity of natural images. To illustrate our main points, while keeping the references small, this paper is slightly biased towards work from our group. We are in the third wave of neural network approaches.

Natural Language Inference is a challenging task that has received substantial attention, and state-of-the-art models now achieve impressive test set performance in the form of accuracy scores. Here, we go beyond this single evaluation metric to examine robustness to semantically-valid alterations to the input data. We identify three factors - insensitivity, polarity and unseen pairs - and compare their impact on three SNLI models under a variety of conditions. Our results demonstrate a number of strengths and weaknesses in the models' ability to generalise to new in-domain instances. In particular, while strong performance is possible on unseen hypernyms, unseen antonyms are more challenging for all the models. More generally, the models suffer from an insensitivity to certain small but semantically significant alterations, and are also often influenced by simple statistical correlations between words and training labels. Overall, we show that evaluations of NLI models can benefit from studying the influence of factors intrinsic to the models or found in the dataset used.

Microbiology research has access to a very large amount of public information on the habitats of microorganisms. Many areas of microbiology research uses this information, primarily in biodiversity studies. However the habitat information is expressed in unstructured natural language form, which hinders its exploitation at large-scale. It is very common for similar habitats to be described by different terms, which makes them hard to compare automatically, e.g. intestine and gut. The use of a common reference to standardize these habitat descriptions as claimed by (Ivana et al., 2010) is a necessity. We propose the ontology called OntoBiotope that we have been developing since 2010. The OntoBiotope ontology is in a formal machine-readable representation that enables indexing of information as well as conceptualization and reasoning.

Predictive business process monitoring methods exploit historical process execution logs to generate predictions about running instances (called cases) of a business process, such as the prediction of the outcome, next activity or remaining cycle time of a given process case. These insights could be used to support operational managers in taking remedial actions as business processes unfold, e.g. shifting resources from one case onto another to ensure this latter is completed on time. A number of methods to tackle the remaining cycle time prediction problem have been proposed in the literature. However, due to differences in their experimental setup, choice of datasets, evaluation measures and baselines, the relative merits of each method remain unclear. This article presents a systematic literature review and taxonomy of methods for remaining time prediction in the context of business processes, as well as a cross-benchmark comparison of 16 such methods based on 16 real-life datasets originating from different industry domains.

Collaborative machine learning and related techniques such as distributed and federated learning allow multiple participants, each with his own training dataset, to build a joint model. Participants train local models and periodically exchange model parameters or gradient updates computed during the training. We demonstrate that the training data used by participants in collaborative learning is vulnerable to inference attacks. First, we show that an adversarial participant can infer the presence of exact data points in others' training data (i.e., membership inference). Then, we demonstrate that the adversary can infer properties that hold only for a subset of the training data and are independent of the properties that the joint model aims to capture. We evaluate the efficacy of our attacks on a variety of tasks, datasets, and learning configurations, and conclude with a discussion of possible defenses.