Search site

School of Computing

Events

School of Computing Research Colloquia - 2014/2015        

Colloquia are 11.30-12.30pm Fridays during term time (40 mins talk + 20 mins questions).
Venue: School of Computing Staff Room (Level 9 EC Stoner)

Fri 24th Oct James Charles, Artificial Intelligence Research Theme
Title: Upper body pose estimation for sign language recognition


Abstract: There is growing evidence that large-scale weakly supervised learning can contribute to sign language recognition. Using the correlations between subtitles and signers in signed TV broadcasts it is possible to train a sign language classifier. However, current research is held back by the difficulty in obtaining a sufficient amount of training video with the position of upper body parts, such as arm and hand, of the signer annotated. This is a great pity because there is a practically limitless supply of such signed TV broadcasts. In this talk I will present a very fast computer vision algorithm which we have developed to automatically annotate the position of the upper body parts. This method can be applied to hundreds of hours of signed video footage, and frees up large amounts of video data for use in improving sign language classifier performance.

Fri 31st Oct Tim Yang, Computational Science and Engineering Research Theme
Title: Multigrid solution methods for nonlinear time-dependent systems


Abstract: An efficient, accurate and reliable numerical solver is essential for solving complex mathematical models and obtaining their computational approximations. The solver presented in this talk is built upon nonlinear multigrid with the full approximation scheme (FAS). Its implementation is generally discussed, which includes using a complex, open source software library PARAMESH, and the resulting numerical solver, Campfire, also combines with adaptive mesh refinement, adaptive time stepping and parallelisation through domain decomposition.

There are three mathematical models considered in this talk, these models consist of nonlinear, time-dependent and coupled partial differential equations (PDEs). The first model is a droplet spreading model with precursor film and moving contact lines. The second model is for fully-developed flows. The final model is a multi-phase-field model which is used to study the growth of tumour.

Fri 7th Nov Richard Emes, Advanced Data Analysis Centre (ADAC), University of Nottingham
Title: Bioinformatics, bugs and big data.


Abstract: Generation and analysis of large, complex data underpins modern biological research, often driving discovery. The successful interpretation of these data is essential to their appropriate and comprehensive use. Recent advances and reducing costs mean that “next generation” sequencing is now the method of choice to investigate biological systems. I will describe recent research using sequencing technologies to understand pathogen biology. Richard Emes is Associate Professor and Reader in Bioinformatics and Director of the Advanced Data Analysis Centre University of Nottingham.

Fri 14th Nov No colloquium

Fri 21st Nov Jie Xu, Distributed Systems and Services Research Theme
Title: Computing at Scale


Abstract: My talk will address some of the unique challenges of systems with hundreds of thousands of server nodes, including 1) performance interference, 2) bottleneck of resource allocation, 3) “long tails”, and 4) efficient error recovery. We have analysed such large-scale systems using real-world tracelogs, from Google, AliCloud and Adapt, demonstrating both analytic and experimental rigor in investigating systems’ behavioural characteristics, actual resource consumption, and overall system dependability due to scale and virtualization. I'll discuss the solutions we have developed so far and outline the way forward.

Fri 28th Nov Margaritis Voliotis, School of Mathematics, University of Bristol
Title: Information transfer by leaky, heterogeneous, protein kinase signaling systems


Abstract: Cells must sense extracellular signals and transfer the information contained about their environment reliably to make appropriate decisions. To perform these tasks, cells use signal transduction networks that are subject to various sources of noise, such as cell-to-cell variability in the componentry of the network and basal activity (the propensity for activation of the network output in the absence of the signal of interest). Basal activity is widespread in signaling systems under physiological conditions, has phenotypic consequences, and is often raised in disease. We combine theory and experiment to study how basal activity and heterogeneity of network componentry affect information transfer by cell signaling systems. First, using theoretical models of protein kinase signaling, we show that negative feedback makes information transfer by the system robust against the combined effect of the two types of noise. We verify our theoretical predictions in an experimental study of ERK signaling by single cells with heterogeneous ERK expression levels. In particular, we find that negative feedback substantially increases information transfer to the nucleus by (i) preventing a near-flat average response curve due to basal network activity and (ii) reducing sensitivity to variation in substrate expression levels. Our results reveal an important role for negative feedback mechanisms in protecting the information transfer function of saturable, heterogeneous cell signaling systems from basal activity.

Fri 5th Dec Dave Westhead, School of Molecular and Cellular Biology
Title: Computational methods in biology and medicine


Fri 12th Dec Andy Bulpitt, Artificial Intelligence Research Theme
Title: Towards Automatic Assessment of the Standard of Orthodontic Treatment


Previous events

Fri 17th Oct Peter Garraghan, Distributed Systems and Services Research Theme
Title: Big Data, Machine Learning and Energy-efficiency in Cloud Computing


Abstract: Big Data, Machine Learning and Energy-efficiency: Three research domains which have gained huge momentum within the past couple of years. Researchers, providers, and pundits alike have claimed that these will be instrumental in solving the research and technical challenges of the 21st century. This is especially true for Cloud datacenters, a multi-billion dollar industry which continues to grow rapidly in terms of consumer uptake, technical innovation and system size. This presentation discusses the experiences and findings from working with commercial Cloud datacenters such as Google (USA), Alibaba (China), and Adapt (UK) to outline common research challenges, current industry practise, and bleeding-edge research innovations developed by Leeds University. These innovations include advanced system analytics for decision support, fault-tolerance for massive scale systems, and practical energy-efficient resource management. We also present a vision for the future of distributed systems research, and discuss a number of problems which are currently unsolved within industry and academia alike.


Fri 10th Oct Peter Bollada, Computational Science and Engineering Research Theme
Title: Phase field modelling


Abstract: I will attempt to explain to a general audience the essence of this subject as applied to the solidification of metals and metal alloys. See the following link for pictures of the sort of solidification I have in mind: http://math.nist.gov/mcsd/savg/vis/dendrite/


Research Colloquia: Agile Development at Songkick
Dr Dan Crow, Chief Technology Officer at Songkick
Active Learning Lab, School of Computing, (9.30a)
Time: 11.30am - 12.30pm, Friday 3 October 2014

Abstract: Songkick is one of London’s original tech startups. Over the past seven years, we’ve used a number of Agile development techniques. We’ve evolved a software development process that works for us and lets us move fast.

We use Kanban and continuous delivery to push 10 releases a day. We use feature flippers to control and aggregate changes and to run extensive multivariate tests on new features. Most importantly, we have a product development culture that is inclusive, devolved and very effective.

I’ll talk about how Songkick’s software development works, the results we’ve achieved, and the lessons we learned - the hard way. This talk focuses on the practicalities of building software and the solutions we use every day.

There will be a reception later on in the day at 4pm in the School of Computing’s Longroom.


Research Colloquia: Twitter: metrics, models and data
Jonathan Ward, School of Mathematics
School of Computing Staff Room (E.C. Stoner level 9)
11.30am, Friday 13 June 2014

Abstract: The widespread adoption of the social media platform Twitter has created an unprecedented volume of data concerning peoples' communication patterns, their social networks and also their attitudes and opinions towards current events. This has proved valuable to media agencies, who use data analytics and social network analysis to provide insights to clients by, for example, identifying influential consumers that can be engaged in real-time. Evolving graph centrality metrics are particularly suitable for this task since they account for the effects resulting from the rapid pace of communication in Twitter. In this talk I will give an overview of one such metric, known as `communicability', with examples of applications to Twitter data. I will also discuss a simple model of Twitter user behaviour that illustrates how the content of some tweets can become extremely widespread.


Research Colloquia: Mobility diversity for robots
Daniel Bonilla Licea, School of Electronic & Electrical Engineering
School of Computing Staff Room (E.C. Stoner level 9)
11.30am, Friday 6 June 2014

Abstract: Communication systems are a very important part of mobile robots. By using communication systems robot can explore areas which are dangerous or inaccessible to humans and send to a base station (BS) sensors measurements and other relevant information. Another popular application is the surveillance of a place by a robot it can patrol some area and send periodically photographs and videos to a surveillance center.

The transmission of the data from the robot to the BS will suffer from the impairments of the wireless channel. Among these impairments we have the small scale fading. Since the robot may operate indoors or in places with rich scattering the small scale fading will affect the robot's communications. It is therefore very important that the robot's communications systems include some form of diversity techniques to combat fading. To solve the problem we could just apply all the classical theory for diversity, select one of the many techniques presented in the literature (e.g., selection combining, maximum ration combining, etc.). Nevertheless, robotic communications are not a classical problem, the key differences of robotic communications with all other wireless communications areas are that the transceiver knows its physical position and can control it, i.e., it can move to wherever it wants. These particular abilities and the fact that the small scale fading differ from position to position enables the use of a new diversity technique.

The mobility diversity is a new kind of diversity still in an early stage of development. This class of diversity consists in moving the robot (and in consequence its transceiver) to different locations (which experience different fading) in order to find a position with high channel gain. This kind of techniques presents a compromise: the more the robot moves the more it improves the channel gain but it also expends more mechanical energy. Therefore, smart algorithms are required.


Research Colloquia: User interfaces for navigating gigapixel images
Roy Ruddle, Applied Computing in Biology, Medicine and Health research theme, School of Computing
School of Computing Staff Room (E.C. Stoner level 9)
11.30am, Friday 30 May 2014

Abstract: In many disciplines people need to navigate large visualizations to analyze data, and our work focuses on a specific aspect of this – visualizing gigapixel medical images. In this talk, I will describe the design and evaluation of user interfaces for navigating such images. The interfaces are novel because they exploit a gigantic ‘thumbnail’ overview that breaks existing design guidelines to let users navigate gigapixel images quickly and accurately. Our evaluations involved images that are orders of magnitude larger than those used in any previous study.


Fri 23rd May No Colloquium


Research Colloquia: Cartographic Information Visualization

Jason Dykes, School of Informatics, City University
School of Computing Staff Room (E.C. Stoner level 9)
11.30am, Friday 16 May 2014

Abstract: Work at the giCentre at City University London creatively explores interactions between Cartography - the development and study of symbolised depictions of geographical settings, and Information Visualization - the use of physical space in graphics to represent non-spatial relationships.

This seminar will focus on recent work in which we develop novel maps and graphics by adding structure to geographic representations to help with comparison and geography to non-spatial representations to reveal geographic relationships.

These partial geographies are used amongst other things to track 21st century bicycles and 19th century migrants, present the UK census on a single page, provide an exploratory public-facing interface to data on local government service provision and reveal bias in London's local elections as giCentre ideas and applications for making sense of masses of data are showcased.


Research Colloquia: Characterizing and Exploiting Heterogeneity for Enhancing Energy-Efficiency of Cloud Datacenters

Ismael Solis Moreno, Distributed Systems and Services research theme, School of Computing
School of Computing Staff Room (E.C. Stoner level 9)
11.30am, Friday 9 May 2014

Abstract: Cloud Computing environments are composed of large and power-consuming datacenters designed to support the elasticity required by their customers. The adoption of Cloud Computing is rapidly growing since it promises cost reductions for customers in comparison with permanent investments in traditional datacenters. However, for Cloud providers, energy consumption represents a serious problem since they have to deal with the increasing demand and diverse Quality of Service requirements. Contemporary energy-efficient Cloud approaches exploit the advantages of virtualization to maximize the use of physical resources and minimize the number of active servers.

A major problem not considered by current Cloud resource management schemes is that of the inherent heterogeneity of customer, workload and server types in multi-tenant environments. This is an issue when improving energy-efficiency, as co-location of specific workload types may result in strong contention for the physical resources. This then affects the resource consumption patterns and therefore the energy-efficiency of virtualized servers. In addition, because of the on-demand self-service characteristic of the Cloud model, different types of customers tend to highly overestimate the amount of required resources. This creates a non-negligible amount of underutilized servers that affects the energy-efficiency of the datacenter.

This thesis analyzes a production Cloud environment to determine the characteristics of the heterogeneous customer, workload and server types, and proposes a novel way to exploit such heterogeneity in order to improve energy-efficiency through two mechanisms. The first improves energy-efficiency by co-locating diverse workload types according to the minimum level of produced interference in a heterogeneous pool of servers. The second mitigates the waste generated by customer overestimation by dynamically overallocating resources based on heterogeneous customer profiles and the levels of produced interference. The evaluation of the proposed mechanisms demonstrates that considering the heterogeneity of elements in a Cloud environment supports the effective improvement of the datacenter energy-efficiency and the performance of individual workloads.