Topic Segmentation – a brief summary of recent approaches
Location: D'Arcy Thompson Room
Date: 11.00 20 Dec 2011
Speaker: Prof. Lei Xie
Organiser: Prof Stephen Cox
Institution: Northwestern Polytechnical University (NWPU), Xi'an, China
Abstract: Topic segmentation is to automatically divide single long recordings (audio, video) or transcripts (text) into topically coherent segments.
It is a necessary prerequisite to many downstream tasks, such as topic detection, tracking and identification, retrieval, summarization and question answering, etc. In this talk, I will summarize our recent work on topic segmentation, covering approaches on detecting lexical changes, clustering, classification and generative modeling. I will also point out some future directions in this topic.

Introduction to Xi'an, NPU and Audio, Speech and Language Processing Group
Location: SCI 3.05
Date: 14.00 15 Dec 2011
Speaker: Prof. Lei Xie
Organiser: Prof Stephen Cox
Institution: Northwestern Polytechnical University (NWPU), Xi'an, China
Abstract: In this talk, I will give a brief introduction to the ancient capital city in China – Xi'an, the Northwestern Polytechnical University and the mostly the audio, speech and language processing (ASLP) group.
My introduction to ASLP will involve the members, research areas, facilities, research projects, research outputs, collaborations and demonstrations. In the demonstration part, I will show some of our latest work on speech recognition and synthesis, virtual auditory, visual speech synthesis, emotion recognition, speech enhancement, echo cancellation and human tracking.

Building Active Appearance Models of Dynamics
Location: D'Arcy Thompson Room
Date: 14.00-15.00 23 Sep 2011
Speaker: David Marshall
Organiser: Dr. Barry John Theobald
Institution: Cardiff University
Materials: Details (Word doc 29 KB)

Now we're talking: an overview and demo of the speech technology features built into Mac OS X
Location: D'Arcy Thompson Room, School of Computing Sciences, UEA
Date: 12.00-13.00 9 Sep 2011
Speaker: Dr Kim Silverman, Manager of Apple Spoken Language Technologies programme
Organiser: Prof. Stephen Cox
Institution: Apple Inc.
There will be content appropriate for for both experts and casual users in the talk.

Cepstral normalisation and the SNR-spectrum in ASR
Location: D'Arcy Thompson room, School of Computing Sciences, UEA
Date: 10.00 5 Apr 2011
Speaker: Phil Garner
Organiser: Prof. Stephen Cox
Abstract: I'll talk about some work involving the effect of cepstral normalisation on additive noise (rather than convolutional as is usual) in ASR. This leads naturally to a speech feature based on signal to noise ratio rather than absolute energy (or power). I'll show that explicit calculation of this SNR-cepstrum by means of a noise estimate has theoretical and practical advantages over the usual (energy based) cepstrum. I'll also talk about how the SNR-cepstrum is almost identical to the articulation index known in psychoacoustics.

Methods of time series analysis for climate tipping points
Location: D'Arcy Thomson Room, School of Computing Sciences, UEA
Date: 14.00 11 Mar 2011
Speaker: Dr. Valerie Livina
Organiser: Dr. Hywel Williams
Institution: School of Environmental Sciences, UEA
Abstract: We develop and apply a set of novel techniques of time series analysis to studying transitions and bifurcations in climatic records. The techniques are degenerate fingerprinting (indicators based on lag-1 autocorrelation and detrended fluctuation analysis) and potential analysis (derivation of underlying system potential and its dynamics).
The techniques allow us to distinguish transitions and bifurcations in the climate system and to detect approaching critical behaviour (early warning). We test the toolbox on artificial data with known properties and then apply to case studies of geophysical data.

  1. Lenton et al, PNAS 2008
  2. Livina & Lenton, GRL 2007
  3. Livina et al, Climate of the Past 2010
  4. Livina et al, Climate Dynamics, in press
  5. Lenton et al, Phil Trans Royal Soc A, in press.

Gaia - perhaps not so lucky after all?
Location: D'Arcy Thompson Room
Date: 16.00 23 Feb 2011
Speaker: James Dyke
Institution: Max Planck Institute for Biogeochemistry, Germany
Abstract: Some notions of Gaia would have us believe that life on Earth is a component of a planetary system that is in some sense self-regulating or homeostatic. Moreover, that this homeostasis benefits life in the sense that those conditions that are good for life are in part stabilised by the effects of life. Much debate about Gaia has centred around the parable of Daisyworld. This simple mathematical model demonstrated that planetary homeostasis can arise with a set of arguably plausible physical assumptions. There have been numerous extensions and developments of the original Daisyworld. Some of this work has been motivated by relaxing the set of assumptions required for homeostasis to emerge.
Until relatively recently, one fundamental assumption seemed necessary: that somehow stabilising negative feedback loops rather than destabilising positive feedback loops had to be somehow selected for. In many Daisyworld models this selection was performed by those that designed the model with the assumption that organisms experienced an increase in fitness when altering the environment in ways that were beneficial to life. This led to James Kirchner arguing that proponents of homeostatic Gaia were advocating a form of 'lucky Gaia'. If the Earth is homeostatic Gaia-like then the planet has simply been lucky to produce negative feedback loops. In conjunction with Andy Watson's anthropomorphic arguments, we have a possible explanation for the emergence and persistence of Gaia in the universe.
In my talk I will show why James Kirchner was wrong. Gaia can emerge in the presence of initially equally probable negative and positive feedback loops. In that respect, the Earth need not have been particularly lucky (although it has certainly been fortunate in many other ways). The explanatory mechanism is something called rein control and is potentially very general and widespread on the Earth and perhaps any other planet that harbours life. In showing how negative feedback loops can emerge in Daisyworld, a more nuanced notion of planetary homeostasis is developed that is arguably a form of homeorhesis and more closely correlates with the actual history of the Earth.

Modelling multiple sources of dissemination bias in meta-analysis
Location: D'Arcy Thompson Room
Date: 14.00 4 Feb 2011
Speaker: Dan Jackson
Organiser: Elena Kulinskaya
Institution: MRC Biostatistics Unit, Institute of Public Health, University of Cambridge
Abstract: Dissemination bias may be caused by publication bias through the decisions of journal editors, by selective reporting of research results by authors or by a combination of both. Typically, study results that are statistically significant or have larger estimated effect sizes are more likely to appear in the published literature, hence giving a biased picture of the evidence-base. Previous statistical approaches for addressing dissemination bias have assumed a single selection mechanism. Here we consider a more realistic scenario in which multiple dissemination processes, involving both the publishing authors and journals, are operating. The methods can be used to provide sensitivity analyses for the potential effects of multiple dissemination biases operating in meta-analysis.

Complex networks in food safety and food security studies
Location: D'Arcy Thomson Room, School of Computing Sciences, UEA
Date: 14.00 28 Jan 2011
Speaker: Dr. Joszef Baranyi
Organiser: Dr. Hywel Williams
Institution: Institute of Food Research
Abstract: The science of Complex Networks was heralded as one of the potential hits of the new disciplines in the 21st century (Science, 2009; 325. 405-432).
One of its main findings is that some organizational principles of complex networks show fundamental laws regardless of the area which it is applied to, and this makes Network Science probably the most multidisciplinary modern scientific tool.
Food science is especially suitable for network science applications, since food-related problems (microbiology, risk assessment, transport, epidemiology, food security and food trade, etc) involve many players in various social contexts, entangled in a COMPLEX NETWORK OF INTERACTIONS where emergent properties of the system cannot be described and predicted by studying only its parts.
In this talk we touch some applications to food science and go into more details on the world's "food-fluxes". In total, how many sources does the ingredients of a pizza go through before you eat it? If a terrorist's aim is food contamination, where in the world would he has the most chance to cause a big havoc, still untraced? Some conclusions will be expected and intuitive, some surprising.

Medical Data Mining: discovery of hypothesis through the analysis of large datasets
Location: D'arcy Thompson, School of Computing Sciences, UEA
Date: 12.00-13.00 25 Jan 2011
Speaker: Dr. Beatriz de la Iglesia
Organiser: Dr. Beatriz de la Iglesia
Institution: UEA
Abstract: In this talk, I will introduce briefly the discipline of data mining looking at its history and motivation. We will explore some of the differences between traditional medical analysis and data mining analysis, highlighting the areas where data mining offers some promising solutions. We will then explore two case studies.
The first case study concerns the analysis of a very large primary care database (over 1 million patients from THIN) for the purpose of testing models of Cardio Vascular Disease (CVD) risk. We will look at the challenges in analysis primary care data and how different CVD risk models (ASSING, Framingham and Cox Framingham) performed. This will serve to present some of the main results for the first external validation of ASSIGN (about to appear in Heart). As a conclusion to this case study we highlight solutions that could be offered by data mining for building better CVD risk models.
The second case study concerns the analysis of data from secondary care. In particular, we will look at an example from gastroenterology in which data relating to numerous procedures was analysed to explore safety problems in relation to sedation. We will look at how the initial exploratory analysis led to further analysis with traditional techniques and eventually to changes in some of the protocols in relation to sedation administration.
To conclude we will summarise the lessons learned in these studies and we will discuss the potential for data mining medical data with the audience.