In just a few short years, breakthroughs from the field of deep learning have transformed how computers perform a wide-variety of tasks such as recognizing a face, tracking emotions or monitoring physical activities. Unfortunately, the models and algorithms used by deep learning typically exert severe energy, memory and compute demands on local device resources and this conventionally limits their adoption within mobile and embedded devices. Data perception and understanding tasks powered by deep learning are so fundamental to platforms like phones, wearables and home/industrial sensors, that we must reach a point where current -- and future -- innovations in this area can be simply and efficiently integrated within even such resource constrained systems. This research vector will lead directly to outcomes like: brand new types of sensor-based products in the home/workplace, as well as enabling increasing the intelligence within not only consumer devices, but also in fields like medicine (smart stethoscopes) and anonymous systems (robotics/drones). The MOA fellowship aims to fund basic research, development and eventual commercialization (through collaborations with a series of industry partners) algorithms that aims to enable general support for deep learning techniques on resource-constrained mobile and embedded devices. Primarily, this requires a radical reduction in the resources (viz. energy, memory and computation) consumed by these computational models -- especially at inference (i.e., execution) time. The proposal seeks will have two main thrusts. First, build upon the existing work of the PI in this area towards achieving this goal which includes: sparse intra-model layer representations (resulting in small models), dynamic forms of compression (models that can be squeezed smaller or bigger as needed), and scheduling partitioned model architectures (splitting models and running parts of them on the processor that suits that model fraction best on certain processors found inside a mobile/embedded device). This thrust will re-examine these methods towards solving key remaining issues that would prevent such techniques from being used within products and as part of common practices. Second, investigate a new set of ambitious directions that seek to increase the utilization of emerging purpose-built small-form-factor hardware processor accelerators designed for deep learning algorithms (these accelerators are suitable for use within phones, wearables and drones). However, like any piece of hardware, it is still limited by how it is programmed - and software toolchains that map deep learning models to the accelerator hardware remain infancy. Our preliminary results show that existing approaches to optimizing deep models, conceived first for conventional processors (e.g., DSPs, GPUs, CPUs), poorly use the new capabilities of these hardware accelerators. We will examine the development of important new approaches that modify the representation and inference algorithms used within deep learning so that they can fully utilize the new hardware capabilities. Directions include: mixed precision models and algorithms, low-data movement representations (that can trade memory operations for compute), and enhanced parallelization.

In just a few short years, breakthroughs from the field of deep learning have transformed how computers perform a wide-variety of tasks such as recognizing a face, tracking emotions or monitoring physical activities. Unfortunately, the models and algorithms used by deep learning typically exert severe energy, memory and compute demands on local device resources and this conventionally limits their adoption within mobile and embedded devices. Data perception and understanding tasks powered by deep learning are so fundamental to platforms like phones, wearables and home/industrial sensors, that we must reach a point where current -- and future -- innovations in this area can be simply and efficiently integrated within even such resource constrained systems. This research vector will lead directly to outcomes like: brand new types of sensor-based products in the home/workplace, as well as enabling increasing the intelligence within not only consumer devices, but also in fields like medicine (smart stethoscopes) and anonymous systems (robotics/drones). The MOA fellowship aims to fund basic research, development and eventual commercialization (through collaborations with a series of industry partners) algorithms that aims to enable general support for deep learning techniques on resource-constrained mobile and embedded devices. Primarily, this requires a radical reduction in the resources (viz. energy, memory and computation) consumed by these computational models -- especially at inference (i.e., execution) time. The proposal seeks will have two main thrusts. First, build upon the existing work of the PI in this area towards achieving this goal which includes: sparse intra-model layer representations (resulting in small models), dynamic forms of compression (models that can be squeezed smaller or bigger as needed), and scheduling partitioned model architectures (splitting models and running parts of them on the processor that suits that model fraction best on certain processors found inside a mobile/embedded device). This thrust will re-examine these methods towards solving key remaining issues that would prevent such techniques from being used within products and as part of common practices. Second, investigate a new set of ambitious directions that seek to increase the utilization of emerging purpose-built small-form-factor hardware processor accelerators designed for deep learning algorithms (these accelerators are suitable for use within phones, wearables and drones). However, like any piece of hardware, it is still limited by how it is programmed - and software toolchains that map deep learning models to the accelerator hardware remain infancy. Our preliminary results show that existing approaches to optimizing deep models, conceived first for conventional processors (e.g., DSPs, GPUs, CPUs), poorly use the new capabilities of these hardware accelerators. We will examine the development of important new approaches that modify the representation and inference algorithms used within deep learning so that they can fully utilize the new hardware capabilities. Directions include: mixed precision models and algorithms, low-data movement representations (that can trade memory operations for compute), and enhanced parallelization.

Female academics, particularly in STEM subjects, score consistently lower than male academics in metrics measuring international [1] and industrial collaborations [2]. These two related assessment criteria are key at all stages in academic careers and particularly important at senior levels to secure the highest value research grants and promotions. While several barriers have been identified to academic career advancement for women and have led to strategic interventions at national and institutional levels, there remains a lack of data and action specifically targeting networking and collaboration - the focus of this VisNET programme. Our vision is 1) To identify key barriers to international collaboration for female engineering academics 2) To design and demonstrate interventions and new best practices in networking and collaborations to define a new and more effective normal. The emergence and rapid development of technologies that support geographically remote working relationships presents a timely opportunity. Effective use of such tools could help to correct the disadvantages experienced by women in international collaboration. We propose an intervention to determine and remodel the implicit 'rules' of networking and collaboration. This pilot project is aimed at a cohort of female post-doctoral researchers (PDRAs). Transition from post-doc to academic is a key attrition point for women in engineering. Success is reliant on demonstrating the means to develop academic independence. Possession of a strong network can be crucial. At the same time this group has relative freedom to trial new approaches of working and represents a critical mass to demonstrate and embed novel methods, including a route to involve more established academics. Thus, the interdisciplinary academic and industrial consortium we have brought together will lead the way in developing, integrating and advocating a new approach where networking and collaboration is conducted predominantly in situ (i.e. from home institutions). We believe that at this critical postdoctoral stage implementation of strategic networking and collaboration can be career defining, providing crucial routes to build confidence, establish future academic independence and funding success. Furthermore, it has the potential to mitigate the impact of future career breaks and parenthood. By demonstrating that networks can be built without frequent travel, it will also address the perception that an academic career is incompatible with work-life balance or family responsibilities, factors identified by junior researchers when consulted about their choice to leave academia [3]. While we see here an opportunity to have a rapid tangible impact on the academic career of a finite group of women, VisNET will also act as an effective route to embed our approaches into the working practices of our universities. Effective in situ networking has the potential to directly tackle negative perceptions of work-life balance in academia, contribute to the promotion of flexible working patterns and advance inclusivity for other minority academic communities such as academics with disabilities or remotely located. The coordinated outcome of this programme fits directly into EPSRC's and our Universities' strategic plans to build leadership, accelerate impact and balance capabilities ensuring the continued progression of UK emerging research leaders by enhancing their experiences and embedding career robustness. [1] Larivière et al., "Bibliometrics: Global gender disparities in science," Nat. News, vol. 504, no. 7479, p. 211, 2013 [2] Tartari & A. Salter, "The engagement gap: Exploring gender differences in University - Industry collaboration activities," Res. Policy, vol. 44, no. 6, pp. 1176-1191, 2015 [3] Shaw & Stanton, "Leaks in the pipeline: separating demographic inertia from ongoing gender differences in academia," Proc. R. Soc. B Biol. Sci., vol. 279, no. 1743, p. 3736, 2012

Topic of Centre: This i4Nano CDT will accelerate the discovery cycle of functional nanotechnologies and materials, effectively bridging from ground-breaking fundamental science toward industrial device integration, and to drive technological innovation via an interdisciplinary approach. A key overarching theme is understanding and control of the nano-interfaces connecting complex architectures, which is essential for going beyond simple model systems and key to major advances in emerging scientific grand challenges across vital areas of Energy, Health, Manufacturing (particularly considering sustainability), ICT/Internet of things, and Quantum. We focus on the science of nano-interfaces across multiple time scales and material systems (organic-inorganic, bio-nonbio interfaces, gas-liquid-solid, crystalline-amorphous), to control nano-interfaces in a scalable manner across different size scales, and to integrate them into functional systems using engineering approaches, combining interfaces, integration, innovation, and interdisciplinarity (hence 'i4Nano'). The vast range of knowledge, tools and techniques necessary for this underpins the requirement for high-quality broad-based PhD training that effectively links scientific depth and application breadth. National Need: Most breakthrough nanoscience as well as successful translation to innovative technology relies on scientists bridging boundaries between disciplines, but this is hindered by the constrained subject focus of undergraduate courses across the UK. Our recent industry-academia nano-roadmapping event attended by numerous industrial partners strongly emphasised the need for broadly-trained interdisciplinary nanoscience acolytes who are highly valuable across their businesses, acting as transformers and integrators of new knowledge, crucial for the UK. They consistently emphasise there is a clear national need to produce this cadre of interdisciplinary nanoscientists to maintain the UK's international academic leadership, to feed entrepreneurial activity, and to capitalise industrially in the UK by driving innovations in health, energy, ICT and Quantum Technologies. Training Approach: The vision of this i4Nano CDT is to deliver bespoke training in key areas of nano to translate exploratory nanoscience into impactful technologies, and stimulate new interactions that support this vision. We have already demonstrated an ability to attract world-class postgraduates and build high-calibre cohorts of independent young Nano scientists through a distinctive PhD nursery in our current CDT, with cohorts co-housed and jointly mentored in the initial year of intense interdisciplinary training through formal courses, practicals and project work. This programme encourages young researchers to move outside their core disciplines, and is crucial for them to go beyond fragmented graduate training normally experienced. Interactions between cohorts from different years and different CDTs, as well as interactions with >200 other PhD researchers across Cambridge, widens their horizons, making them suited to breaking disciplinary barriers and building an integrated approach to research. The 1st year of this CDT course provides high-quality advanced-level training prior to final selection of preferred PhD research projects. Student progression will depend on passing examinable components assessed both by exams and coursework, providing a formal MRes qualification. Components of the first year training include lectures and practicals on key scientific topics, mini/midi projects, science communication and innovation/scale-up training, and also training for understanding societal and ethical dimensions of Nanoscience. Activities in the later years include conferences, pilot projects, further innovation and scale up training, leadership and team-building weekends, and ED&I and Responsible Innovation workshops