Simon Dobson

Jiawei Luo, Di Wu, Simon Dobson et al.

Modern edge applications increasingly require multi-DNN inference systems to execute tasks on heterogeneous processors, gaining performance from both concurrent execution and from matching each model to the most suited accelerator. However, existing systems support only a single model (or a few sparse variants) per task, which impedes the efficiency of this matching and results in high Service Level Objective violation rates. We introduce model stitching for multi-DNN inference systems, which creates model variants by recombining subgraphs from sparse models without re-training. We present a demonstrator system, SparseLoom, that shows model stitching can be deployed to SoCs. We show experimentally that SparseLoom reduces SLO violation rates by up to 74%, improves throughput by up to 2.31x, and lowers memory overhead by an average of 28% compared to state-of-the-art multi-DNN inference systems.

Muffy Calder, Simon Dobson, Michael Fisher et al.

Sensor-driven systems are increasingly ubiquitous: they provide both data and information that can facilitate real-time decision-making and autonomous actuation, as well as enabling informed policy choices by service providers and regulators. But can we guarantee these system do what we expect, can their stake-holders ask deep questions and be confident of obtaining reliable answers? This is more than standard software engineering: uncertainty pervades not only sensors themselves, but the physical and digital environments in which these systems operate. While we cannot engineer this uncertainty away, through the use of models we can manage its impact in the design, development and deployment of sensor network software. Our contribution consists of two new concepts that improve the modelling process: frames of reference bringing together the different perspectives being modelled and their context; and the roles of different types of model in sensor-driven systems. In this position paper we develop these new concepts, illustrate their application to two example systems, and describe some of the new research challenges involved in modelling for assurance.

Chris Schneider, Adam Barker, Simon Dobson

Autonomously detecting and recovering from faults is one approach for reducing the operational complexity and costs associated with managing computing environments. We present a novel methodology for autonomously generating investigation leads that help identify systems faults, and extends our previous work in this area by leveraging Restricted Boltzmann Machines (RBMs) and contrastive divergence learning to analyse changes in historical feature data. This allows us to heuristically identify the root cause of a fault, and demonstrate an improvement to the state of the art by showing feature data can be predicted heuristically beyond a single instance to include entire sequences of information.