LIDA Seminar Series 21st February 2019

The Leeds Institute for Data Analytics is pleased to present the next seminar in our series showcasing data analytics.

The seminar will be held in 8.43Y, Worsley Building, at 15.30 on Thursday 21st February.

This seminar will consist of five ten minute presentations from our current LIDA Data Scientist interns.

Agenda

15.30

Understanding and Quantifying Uncertainty in Agent-Based Models for Smart City Forecasts
Kevin Minors
Agent-based models are an effective way to model the dynamics of smart cities but it is difficult to assimilate real-time data into the models. The aim of this project is to determine whether a particle filter, one form of data assimilation, can be used to assimilate new observations in a non-linear, non-Gaussian agent-based model.

15.40

Understanding customers’ missions from the products they purchase
Ivana Kocanova
"What products do our customers buy together" is a common question that retailers want to answer, because it provides them with insights about customers’ shopping behaviour that underpin strategic investments (e.g., in new stores, store layouts, or product lines). Today, retailers use a variety of statistical modelling approaches to analyse customers’ shopping transactions, but the models are not very accurate, and require a large amount of human effort to update. This project investigates how novel data mining and visualization techniques can speed-up the analysis of shopping behaviour and produce models that are more accurate.

15.50

Procedural environment generation for agent-based models of crime
Jack Lewis
Urban morphologies, such as patterns of land usage and street network configurations, fundamentally shape people’s day-to-day activities in complex and non-linear ways, dictating where, when, and in what contexts individuals interact with one another. As a result of difficulties associated with the empirical investigation of these interactions, scholars in recent years have applied agent-based models to study the relationship between urban morphology and space time behaviour.
To date, these models have taken one of two approaches, exploring the implications of individual-level behaviour in either: (1) multiple abstract manipulable environments; or (2) single real world environments derived from GIS data (e.g. Seattle, Leeds etc.). Both approaches have strengths and weaknesses. This project aims to explore techniques for procedural environment generation for agent based models. Developing algorithms capable of generating realistic synthetic but experimentally manipulable simulation environments, we seek to explore how these techniques might increase our understanding of how street networks influence patterns of crime.

16.00

Using machine learning to understand the dynamics of cities
Benjamin Wilson
In the recent decade, new innovations in sensor technologies have given rise to a surge of data on our modern cities but new methods are needed to make use of these rich sources. A common type of data that is captured from transportation systems contains origin destination information. These rich data structures hold spatial-temporal information of human behaviours spanning urban environments and provide opportunity for exploration and development of novel approaches. This project examines the effectiveness of novel approaches involving clustering algorithms, dimension reduction and visualisation methods.

16.10

Probabilistic Programming and Data Assimilation for Next Generation City Simulation
Luke Archer
Agent-Based Models (ABMs) are ideally suited to modelling the dynamics of social systems, but their predictive power is limited in that they can only be trained once on historical data, meaning simulations quickly diverge from reality. Dynamic Data Assimilation (DDA) is a group of well-studied techniques pioneered in meteorology, which allow a model to be updated with data in real-time. Here, we attempt to integrate a probabilistic programming library keanu (in development by Improbable Worlds Ltd.) with a test model created using the MASON framework. This will allow us to perform state estimation and data assimilation on an ABM using keanu's built-in algorithms and data structures.

16.20

Q&A with all speakers

17.00

Networking reception with drinks and nibbles hosted in the LIDA staff kitchen

18.00

Close

To book your free place at this seminar please email Hayley Irving with your name, occupation and faculty/organisation.