Asthma is the most common chronic disease of childhood and causes many hospital admissions. The number of children suffering from asthma has increased dramatically over the past 30 years. It is unclear why some people get asthma and others do not. Asthma is largely heritable, but despite lots of effort we have had limited success identifying which genes are important, and genetic studies of asthma have not yet had a positive impact on patient care. Many factors in the environment may contribute to the development of asthma (for example diet, immunizations, antibiotics, pets and tobacco smoke) but we don't know how to modify the environment to reduce the risks. One reason for the difficulty in understanding causes of asthma is that asthma may be a collection of different diseases which cause similar symptoms. As asthma generally starts early in life, the best way to study it is to recruit new born babies and follow them as they grow (so-called birth cohort). During early life it is possible to measure many things, such as exposure to allergens, (generally cats, dogs and dust mite) constituents of the diet and antibiotic usage. Questionnaires are used to collect information from parents about symptoms, and as the children get older they can take part in measurements of lung function and allergy testing. Most are willing to give sample for genetic testing. In the UK there are 5 such birth cohorts that have been designed to facilitate the study of risk factors for asthma and allergies. The Manchester Study has more than 1,000 children under active follow up; clinical follow up is complete for age 1, 3, 5, 8 and 11 years. The study from Aberdeen included 1,924 children who were followed up 6 months, 1, 2, 5 and 10 years. The Ashford birth cohort followed 642 children until age 12 years. The Isle of Wight study recruited 1,456 children, completing follow up at ages 1, 2, 4, 10 and 18 years. ALSPAC recruited through antenatal clinics in the former County of Avon, and enrolled 14,062 infants. Follow up is complete to age 16 years. All have collected data using questionnaire and performed measures of lung function and skin tests at intervals throughout early life. The researchers who lead these studies have worked together as a network over the last 7 years - the Study Team for Early Life Asthma Research (STELAR consortium). Over recent years we have adopted identical research methodologies, recognising that although each cohort is unique, there are many aspects in which we can work together, to increase our chances of detecting the main causes of asthma. We now propose to create a major new alliance which will combine our world-leading expertise in birth cohorts (STELAR consortium) with expertise in health informatics research (NW Institute for Bio-Health Informatics) and cutting-edge computational statistical methods (so-called statistical machine learning, with experts at our Industrial partner at Microsoft Research Cambridge). Our alliance includes clinicians, public health and epidemiology researchers, statisticians, informaticians and software engineers. We will construct a web based resource (Asthma e-Lab) in which to securely store all the data collected on the cohorts and recipes for analysing the data so that a larger group of scientists can repeat the work. This will enable consistent recording, description and sharing of data and emerging findings across all partners. We will also complete clinical follow-up of cohort participants where necessary. We will work together to apply newly developed state-of-the-art data analysis techniques to build complex models to describe different types of 'asthma' and investigate risk factors for each asthma subtype. In doing so we hope to understand the basic biological mechanisms that underlie the different forms of asthma, Our findings may underpin new trials of asthma prevention and may help identify targets for the discovery of novel therapies which are matched to specific patients.

Complex software systems involve many components and make use of many external libraries. Programmers who work on such software must remember the protocols for using all of those components correctly, and the process of learning to use a new component can be time consuming and a source of bugs. We believe that there is a major untapped resource that can help address this problem. Billions of lines of code are readily available on the Internet, much of which are of professional quality. Hidden within this code is a large amount of knowledge about good coding practices, for example, about avoiding error-prone constructs or about the best protocol for using a particular library. We envision a new type of programming tool, which could be called data-driven development tools, that aggregate knowledge about programming from a large corpus of mature software projects, for presentation within the development environment. Just as the current generation of IDEs helps developers to manage their code, the next generation of IDEs will help developers to learn how to write better code. Fortunately, there is a research field that has already developed a large body of sophisticated tools for analyzing large amounts of text: namely, statistical natural language processing. The long-term strategic goal of this project is to develop new natural language processing techniques aimed at analyzing computer program source code, in order to help programmers learn coding techniques from the code of others. There is a large area for research here that has been almost completely unexplored. As a first step in this research area, in this project we will focus on automatically identifying short code fragments, which we call idioms, that occur repeatedly across different software projects. An example of an idiom is the typical construct for iterating over an array in Java. Although they are ubiquitous in source code, idioms of this form have not to our knowledge been systematically studied, and we are unaware of any techniques for automatically identifying idioms. The main objective of this project is to develop new statistical NLP methods with the goal of automatically identifying idioms from a corpus of source code text. We call this research problem idiom mining, and it is to our knowledge a new research problem. This is an interdisciplinary project that draws from statistical NLP, machine learning, and software engineering. The research work of this project is primarily in statistical NLP and machine learning, and will involve developing new statistical methods for finding idioms in programming language text.

Separation logic is a formalism designed to symbolically reason about computer programs that dynamically allocate data structures on memory. It allows to one mathematically _prove_ the correctness of such programs, for example showing the absence of memory leaks or violations. Using this proof system theoreticians could provide much cleaner and succinct proofs on the whiteboard, but their efforts to automate such reasoning process faced a number of limitations. Standard automated theorem proving techniques, based e.g. on resolution, did not seem to naturally apply to the new logic; thus researchers opted either to implement reasoning tools from scratch---missing the opportunity to exploit advances in modern theorem proving---or seek alternatives that avoid the perceived nuances of separation logic altogether---missing the conceptual advantages that separation logic _does_ provide. This research proposal seeks to demonstrate that modern theorem proving techniques are, in fact, compatible with separation logic reasoning. An approach that leads not only to much more efficient separation logic reasoners, but also broadens the scope of applications in which automated theorem proving can be practically applied.

FLoC (Federated Logic Conference) is a quadrennial two-week event that brings together eight top conferencesapplying methods of logic in computer science, and about 60 workshops. The area is of great importance to computing research: logic provides computer science with both a unifying foundational framework and a tool for modelling computing systems; it has been called ``the calculus of computer science,'' and played a crucial role in diverse areas such as artificial intelligence, computational complexity, distributed computing, database systems, hardware design, programming languages, and software engineering. The research of just over 20% of Turing award winners has been on the application of logic to computer science; in the UK, five of the current eight UKCRC grand challenges are in FLoC areas.The proposal is to enable us to invite top-level researchers to deliver plenaries and tutorials at FLoC, to makeFLoC particularly attractive to students and to make it easier for researchers from the UK and elsewhereto attend FLoC.

The goal of computer vision is to impart machines with the ability to see, that is, to understand an image similar to a human. This consists of identifying which object categories are present in an image and where (car, person, trees), their actions (running, driving, sitting), and their relative locations (person inside the car, car above the road). The challenge of computer vision lies in obtaining a powerful discriminative representation of an image that allows us to infer the scene encoded within it. Consider, for instance, the representation of an image as captured by a camera, which consists of color values of each of its pixels. Two images that differ greatly in their color values may still depict the same scene (for example, images of the same location at day and night). At the same time, small changes in the color values may result in a completely different scene where objects have moved considerably. This makes raw color values unsuitable for understanding the scene depicted in the image. Traditional computer vision approaches have relied on hand-designed representations of an image that are more amenable to scene interpretation. Given such a representation, there exist several principled formulations for learning to interpret scenes, which take advantage of the powerful mathematical programming framework of convex optimization. Convex optimization offers many computational advantages: it scales elegantly with the size of the problem, it provides global optimality, it offers convergence guarantees and it can be parallelised over multiple machines without affecting the accuracy. Recent years have seen the rise of deep learning, and specifically convolutional neural networks (CNN), which aim to automatically obtain the representation of visual data from a large training set. While an automated approach is highly desirable due to its scalability, it comes with the challenge of solving highly complex non-convex mathematical programs. The aim of our research is to overcome these challenges by finding connections between convex optimization and deep learning. The key observation is that the non-convex programs encountered in deep learning for computer vision have a special structure that is closely related to convexity. Specifically, while the mathematical programs are not convex, they are of a difference-of-convex (DC) form. A DC program can be optimized efficiently by an iterative algorithm, which, at each iteration, solves a convex optimization problem. Our aim is to exploit the structure of DC-CNNs to design the next generation of algorithms for computer vision. Specifically, we will build customized algorithms that will scale up the dimensionality of the CNN by orders of magnitude while keeping the computational cost low. Our algorithms will retain many of the highly desirable benefits of convex programming (convergence, quality guarantees, elegant scaling, distributed computing) while still allow the automatic estimation of image representations. The impact of such principled and efficient algorithms is potentially huge. The new CNN architectures that this enables will allow researchers to address significantly more complex visual tasks. For example, a generative network that can provide a set of diverse future frames of a given video sequence, or a intelligent agent that can crawl the web for images and videos and complete the captions in order to bridge the gap between visual data and searchable content. Our research results will be made publicly available via open source software. The project is also likely to have a large academic impact, consolidating the leadership of the UK in machine learning and computer vision.