Hilary Term 2016

HILARY TERM 2016

Week 2:     Thursday 28th January, 2.15 p.m.

Speaker:     Sofia Olhede, Department of Statistical Science, University College London
Title:            Big Data in Time
Abstract: 
  Sensors produce continuous recordings of temporal phenomena associated with spatial specificity. Their analysis will depend on the form of acquisition, and normally with a large set of observations the challenge is to combine information to obtain a cohesive spatial picture. Additionally acquisition is associated with irregular sampling in time and space, requiring careful interpolation. An important characteristic of such observations is whether sensors are stationary (normally referred to as Eulerian observations), or meandering (normally referred to as Lagrangian observations). I will discuss how stochastic models can reproduce the variability seen in practice in the example of oceanographic monitoring, and what outstanding challenges still remain in modelling and making sense of such data.

This is joint work with Jeffrey Early, Jonathan Lilly and Adam Sykulski.

Week 5:     Thursday 18th February, 2.15 p.m.

Speaker:    Jonas Peters, Max Planck Institute for Intelligent Systems, Germany
Title:           Invariance and Causality

Abstract:    Why are we interested in the causal structure of a data-generating process? In a classical regression problem, for example, we include a variable into the model if it improves the prediction; it seems that no causal knowledge is required. In many situations, however, we are interested in the system's behavior under a change of environment. Here, causal models become important because they are usually considered invariant under those changes. A causal prediction (which uses only direct causes of the target variable as predictors) remains valid even if we intervene on predictor variables or change the whole experimental setting. We show how we can use data from different environments in order to estimate the causal structure and apply this methodology to a gene deletion data set. This talk does not require any knowledge about causal concepts.

J. Peters, P. Bühlmann, N. Meinshausen: "Causal inference using invariant prediction: identification and confidence intervals" (to appear in JRSSB, with discussion).

Week 7     Thursday 3rd March

POSTGRADUATE POSTER SESSION (2nd years)

Week 8     Friday 11th March, 3.30 p.m. (Please note new later start time)

Speaker:   Mark Newman, University of Michigan (Distinguished seminar speaker)
Title:          Large-scale structure in complex networks

Abstract:   Many systems of interest in science and engineering can be represented as networks: the internet, the power grid, transport networks, metabolic networks, ecological networks and social networks are just a few of the many well studied examples.  This talk will discuss methods for quantifying and understanding the large-scale structure of these complex objects, including core-periphery structure, hierarchical structure, ranking, latent-space structure, and especially community structure.  In particular I will consider two broad classes of methods, the first based on statistical inference and the second on spectral techniques, along with a set of beautiful recent results that suggest deep connections between these two apparently quite different approaches.

Week 9:     Monday 14th March, 3.30 p.m. (Please note new later start time)

Speaker:    Pradeep Ravikumar, University of Texas
Title:          The Distributional Rank Aggregation Problem, and an Axiomatic Analysis

Abstract:   The rank aggregation problem has been studied in varied communities such as Theoretical Computer Science, Statistics, Information Retrieval and Social Welfare Theory. We introduce a variant of this problem we call distributional rank aggregation, where the ranking data is only available via the induced distribution over the set of all permutations. We provide a novel translation of the usual social welfare theory axioms to our setting, which has two consequences. First, this allows for a more quantitative characterization of these axioms, thus less prone to misinterpretation. Secondly, these quantitative characterizations lead to natural and novel relaxations of these axioms, which as we show, actually allow us to finesse celebrated impossibility results in social choice theory: providing rules that satisfy all the "impossible" axioms simultaneously, but with some slack.

Joint work with Adarsh Prasad and Harsh Pareek.