Welcome to CRiSM seminar series!

Seminars take place biweekly during term time between 3 and 4pm, and will be hosted on Microsoft Teams.

We encourage all postgraduate students (MSc and PhD) to attend this series: it is a great opportunity to know more about current research within the department and outside.

CRiSM seminars 2020/21 are organised by Tom Berrett.

The following talks are scheduled for the academic year 2020/2021:

Term 1

• Wednesday October 28 - François Caron - University of Oxford

• Thursday November 12 - Stefan Wager - Stanford University

• Thursday November 26 (10am) - Wai Kin Wong - Hong Kong Observatory

• Thursday 10 December - Sofia Olhede - EPFL Lausanne

Term 2 (on Wednesdays)

• January 20 - Pragya Sur - Harvard University

• February 3 - Rina Foygel Barber - University of Chicago

• February 17 - Alexandra Carpentier - Otto-von-Guericke-Universität Magdeburg

• March 3/4 - TBA

• March 17 - Peter Bühlmann - ETH Zürich

Term 3

• May 5/6 - TBA

• May 19/20 - TBA

• June 2/3 - TBA

• June 16/17 - TBA

• June 30/July 1 - TBA

Titles and abstracts will appear below as soon as they are available.

-------------------------------------------------------------------------------------------------------------------------------

October 28: François Caron (University of Oxford) Meeting Link

Title: Non-exchangeable random partition models for microclustering

Abstract: Many popular random partition models, such as the Chinese restaurant process and its two-parameter extension, fall in the class of exchangeable random partitions, and have found wide applicability in model-based clustering, population genetics, ecology or network analysis. While the exchangeability assumption is sensible in many cases, it has some strong implications. In particular, Kingman’s representation theorem implies that the size of the clusters necessarily grows linearly with the sample size; this feature may be undesirable for some applications, as recently pointed out by Miller et al. (2015). We present here a flexible class of non-exchangeable random partition models which are able to generate partitions whose cluster sizes grow sublinearly with the sample size, and where the growth rate is controlled by one parameter. Along with this result, we provide the asymptotic behaviour of the number of clusters of a given size, and show that the model can exhibit a power-law behaviour, controlled by another parameter. The construction is based on completely random measures and a Poisson embedding of the random partition, and inference is performed using a Sequential Monte Carlo algorithm. Additionally, we show how the model can also be directly used, by relaxing the exchangeability assumption in edge-exchangeable models, to obtain a class of sparse multigraphs with power-law degree distribution and sublinear growth of the node degrees. Finally, experiments on real datasets emphasize the usefulness of the approach compared to a two-parameter Chinese restaurant process.

Joint work with Giuseppe di Benedetto and Yee Whye Teh

https://arxiv.org/pdf/1711.07287.pdf

November 12: Stefan Wager (Stanford University) Meeting Link

Title: Noise-induced randomization in regression discontinuity designs

Abstract: Regression discontinuity designs are used to estimate causal effects in settings where treatment is determined by whether an observed running variable crosses a pre-specified threshold. While the resulting sampling design is sometimes described as akin to a locally randomized experiment in a neighborhood of the threshold, standard formal analyses do not make reference to probabilistic treatment assignment and instead identify treatment effects via continuity arguments. Here we propose a new approach to identification, estimation, and inference in regression discontinuity designs that exploits measurement error in the running variable. Under an assumption that the measurement error is exogenous, we show how to consistently estimate causal effects using a class of linear estimators that weight treated and control units so as to balance a latent variable of which the running variable is a noisy measure. We find this approach to facilitate identification of both familiar estimands from the literature, as well as policy-relevant estimands that correspond to the effects of realistic changes to the existing treatment assignment rule. We demonstrate the method with a study of retention of HIV patients and evaluate its performance using simulated data and a regression discontinuity design artificially constructed from test scores in early childhood.

https://arxiv.org/abs/2004.09458 

-------------------------------------------------------------------------------------------------------------------------------

November 26 at 10am: Wai Kin Wong (Hong Kong Observatory) Meeting Link

Title: Machine learning in rainfall nowcasting.

Abstract: Rainfall nowcasting refers to the prediction of precipitation in very high spatial and temporal resolutions for the next 1-6 hours. Timely and quality rainfall nowcast provides indispensable source of information in support of rainstorm monitoring, alerting or warning systems that are invaluable to weather services, and disaster risk reduction of high-impact weather or rainstorms for protecting people's lives. In Hong Kong Observatory (HKO), artificial intelligence technique based on image processing algorithms have been utilized in the in-house nowcasting system, namely SWIRLS (Short-range Warning of Intense Rainstorms in Localised Systems) to track the motion of precipitation systems detected by weather radars, followed by predicting their future location and rainfall using the motion field. However, the intensity is assumed to remain unchanged in computation that results in decreasing skill of precipitation forecast beyond one or two hours. In recent years, novel deep learning (DL) based methods have been developed for precipitation nowcasting that have shown improved performance compared to the above operational algorithm. In this talk, the current progress of DL based methods for precipitation nowcasting will be introduced, including mathematical formulation of precipitation nowcasting as a spatiotemporal sequence forecasting problem, and a couple of general learning strategies. Performance of DL based nowcasting model and a systematic benchmark for performance evaluation will be presented. Finally, future research directions on development of DL in precipitation nowcasting and meteorological forecasting applications are discussed.

Past Seminars

July 2007 - Present

2005 - July 2007

Slides