Achieving the optimal delay/energy/accuracy requires the co-optimization of Neural Network (NN) workload together with the system hardware it is deployed in. This is particularly important on mobile systems running on batteries, where to deploy more complex workloads, retraining-adaptation might be limited due to energy or computation constraints. Neural Architecture Search (NAS) solutions enable the exploration of the NN graph characteristics in an efficient way, adapting the design specifics of the neural architecture to the machine learning task at hand. The aim of this PhD tackles the design of an open-source, hardware aware NAS methodology to optimize not only inference but also on-device training. Furthermore, hardware considerations will be expanded to lower-level processor specifications (for instance, NVM parameters, DVFS design) alongside system-level characteristics. Key Research Questions The objective of the research is to investigate the open problems in the domain of hardware-aware network architecture search (NAS). In particular, it will focus on a proof of concept system model that will be aware of established opportunities for efficiency. Examples of these include different frequency/voltage operation points (DVFS) and hybrid memories (off-chip/on-chip, non-volatile/volatile) with multiple power operation modes. A key challenge will be integrating this hardware-defined search-space with that of search-space of the neural architecture to perform joint-optimization of both neural architecture and hardware design spaces in a tractable manner. The design of this NAS method will be a useful tool, but just as important will be the results this tool is able to produce - specifically novel neural architectures and novel hardware designs that especially when run jointly produce high efficiency systems. It is expected that by analysing the workload-aware optimized systems, design and micro-architectural innovations will result in the exceptional power, performance and area metrics for such heterogeneous hardware designs. Work expected of the ICASE student During the Ph.D, the student will research and address the computing and memory requirements of a system model that includes the NAS scheme and optimises an on-device training workload. While addressing the key research questions, the student will work in a top-class environment, with colleagues from the CaMLSys group at the University of Cambridge. There will be a minimum of a 3 month placement with the industrial partner IMEC where the student will work in a diverse environment with scientists devoted to world leading R&D and innovation in nanoelectronics and digital technologies.

Simulation tools are only as useful as the accuracy of the data and results that they produce. Particularly in robotics, when using machine learning techniques, there exists a real-world simulation gap. This means that data or learning schema produced in simulation, won't necessarily account for real-world nuances. In terra-mechanics (and therefore machine-soil interaction), this problem is only exacerbated by almost random terrain compounds (soils are a culmination of all the materials laden within it), and the terrains complex self-interactions. Even miniscule changes in mixture percentages (not to mention other environmental effects such as humidity) could have a large effect on the outcome of any particular attempt to manipulate the earth. However, within the field of terra-mechanics there is a lack of rigorous datasets that are able to provide well needed foundations to help guide the simulation tools. Utilising our industry standard 6-axis robotic manipulator and a custom-built earth-box, we will produce high quality datasets describing precise force measurements and the resulting terrain deformation. These tests will be repeated for numerous different shapes/tools/form factors. For example, most quadrupedal robots have simple ball feet. With this shape as the end effector, following a fully-fleshed out experimental process which considers all possible movements for a wide range of forces. During/following the gathering of these datasets, the project plans to build an autonomous zen-garden robot system (ZenBot) that is able to interact with the material and draw provided patterns. This would be a closed-loop system, where the current garden is compared against the desired garden. This is a challenge that encompasses not only the prior-mentioned problem but also requires skills ranging from terrain mapping, to control algorithm design. The data gathered would have many potential use cases. One of particular interest to me is utilising the knowledge of the terrain to inform locomotion policies for quadrupedal robotics. Knowing the environment could be used to understand how particular movements not only deform the terrain, but propel the robot in a given direction. This could enable a new wave of intelligent robotics, that not only understands where to move, but how to move most efficiently.

My project intends to queer the history of masculinity and masculine dress in early modern history. There is evidence for a significant number of female sex workers wearing masculine attire in early modern Italy. However, there is yet to be a study that investigates the forms that this masculine expression took and women's motivations for gender nonconformity. My aim is to fill this gap in the scholarship, and to explore the following questions. Did sex workers choose masculine attire for only professional reasons (to attract homosexual male clients, as authorities worried), or as an expression of identity? By extension, is there a queer and/or trans history to be brought to light here? What can sartorial expression tell us about the permeability of gender boundaries in early modern, pre-Enlightenment Europe? I intend to use a combination of syphilis hospital records, court records, popular pamphlets and travel writing as source material. Syphilis hospitals recorded detailed notes on patients' attire. Since sex workers were particularly afflicted, an exploration of syphilis hospital records from across Italy could provide insight into how common it was for female sex workers to wear masculine attire, what kinds of garments were worn, and their materiality. Court record accounts will be useful for uncovering sex workers' motivations for cross-dressing. Travel writing, popular pamphlets and publications will provide evidence of sartorial expression embedded in urban space.

TBC

Speech constitutes action (Austin 1962) and, sometimes, it constitutes the action of subordination (Langton 1993). One's utterance can constitutively (rather than causally) unfairly rank someone as inferior and make them count as inferior in a particular domain, unfairly deprive them of rights or powers, and/or legitimate discriminatory behaviour against them (Langton 1993). Such an utterance is the illocution of subordination (henceforth subordinating illocution). The philosophical literature on subordinating illocutions focuses primarily on in-person illocutions. Little systematic attention has been paid to how speakers perform subordinating illocutions in online communicative environments (social media platforms like Twitter and Facebook). This is a gap that needs to be filled given that online environments are a primary outlet for hate speech, they occupy an important role in our communicative practices, and they are importantly different to in-person conversational exchanges, therefore requiring special philosophical treatment. I aim to offer a detailed analysis of online subordinating illocutions, adding both to the literature on subordinating illocutions and to the nascent literature on online illocutions.