Complexity Science » Warwick Annual Retreat Projects

Modelling the influence of human behaviour on the effectiveness of vaccination strategies

A guest user — Thu, 28 Apr 2016 11:37:16 GMT

Vaccination is one method of control used against infectious diseases. However, at times vaccination efforts have been met with some controversy on medical safety and religious grounds, while recent years have seen the rise of an anti-vaccination movement in the USA.
This project would investigate the impact of human behaviour on the effectiveness of vaccination strategies in preventing epidemic outbreaks.

A starting point of the project would be to discuss the modelling assumptions to be made. For example, for the behaviour dynamics the most simple choice would be to have a pro-vaccination class and anti-vaccination class, but should there be a third class where the decision to vaccinate or not is influenced by the current level of infection?

The aim would then be to construct a simple deterministic and/or stochastic mathematical model that incorporates status with respect to the disease, while also allowing for individuals to adjust their opinion on vaccination.

Possible scenarios these models could be used to analyse (based on interest) are the impact of:

Vaccine scares
Vaccine shortages
Comparing different vaccination implementation strategies.

An extension to this work would be to introduce distinct communities that may favour a particular stance towards vaccination via the use of a metapopulation model.
Comments/ ideas are welcome.

Soft matter problems

A guest user — Wed, 27 Apr 2016 14:00:25 GMT

As a result of a quick brain-storming session in the Alexander group, here is a list of recent papers / topics which could serve as a basis for a project related to elasticity and/or fluid dynamics: - critical comment on a paper about elastic membrane deformations in bi-lipid membranes: https://arxiv.org/pdf/1604.03865
- breakdown of axisymmetric buckling of spherical caps: http://arxiv.org/pdf/1604.03686.pdf
- no coffee ring effect in evaporating whiskey drops: http://journals.aps.org/prl/abstract/10.1103/PhysRevLett.116.124501
- any of these papers from Mahadevan's group (origami tesselations, shape of human brains, etc) http://www.seas.harvard.edu/softmat/
- paper folding, origami, hexaflexangons

Re: Project: Raindrops on the window of a moving train

Arthur King — Tue, 26 Apr 2016 14:43:25 GMT

At the BAMC conference Eugene Benilov gave a talk 'A thin drop sliding down an inclined plate' in which he cited this paper; might be useful.

Ho-Young Kim, Heon Ju Lee, Byung Ha Kang, Sliding of Liquid Drops Down an Inclined Solid Surface, Journal of Colloid and Interface Science, Volume 247, Issue 2, 15 March 2002, Pages 372-380, ISSN 0021-9797

http://dx.doi.org/10.1006/jcis.2001.8156

Crowd dynamics of molecular motors

Neil Jenkins — Tue, 26 Apr 2016 09:16:47 GMT

A project to investigate a new experimental set-up for investigating the behaviour of molecular motors.

A simulation code is available and the geometry of the set-up needs adjusting.

I wrote the simulation in C++ and it uses some python but knowldge of these languages wouldn't be required for people to help out.

Hopefully a good oppertunity to have a play with some stochastic modelling and OOP methods. Will also involve some data analysis of the output to help predict what might be seen when the real experiments are performed.

Attachments (follow link to download)

miniproject_proposalr.docx (53 KB)

Investigating the behaviour and demographic of mobile phone users

A guest user — Mon, 25 Apr 2016 07:55:18 GMT

I have several datasets on mobile phone activities that can be analysed. One of them is a very comprehensive dataset referring to seven different main cities in Italy and provides information on:

number of SMSs sent and received
number of phone calls made and received
volume of access to the Internet through smart phones
aggregated demographic information: age range, gender, and post code where the SIM was registered/purchased

All of this information is provided at a time granularity of fifteen minutes and divided on geographical cells provided by the dataset. For instance, the dataset gives you the overall number of phone calls in a fifteen minutes interval for a specific cell. There is also an estimate of how many people were present in any cell at any time interval.

A starting point of the project could investigate the demographic of users and how it changes over time. Is there any interesting information that can be found by using not only the mobile phone activity but also the information on gender, age range and how many people there are in a cell? What are the different behaviours over a day, week or month, of mobile phone users depending on their age and gender?

Part of this would also involve discussing further interesting questions that may be answered through this dataset.

This project could be relevant for everyone with an interest in computational social science and willing to learn/share experience on how to deal with timestamped geographical data, shapefiles and other data analaysis tools.

Re: Optimality and resilience of fire service staff rotas

Ayman Boustati — Fri, 22 Apr 2016 10:21:33 GMT

Attached is the project briefing.

Attachments (follow link to download)

wmfs.doc (34 KB)

Optimality and resilience of fire service staff rotas

A guest user — Thu, 21 Apr 2016 17:21:41 GMT

West Midlands Fire Service (WMFS) has committed to a clear 3 year rolling strategy focused on providing an evidence based service to our community. Through academic research we have evidenced that a 5 minute response time is critical in relation to survivability resulting in the need for 41 fire engine and 19 brigade response vehicles being available. The organisation has historically done this through having a ridership factor of 1.42 (This is 1.42 people per position on a response vehicle to enable leave, training and absence). Due to large budget reductions a different approach to staffing is required.

Historically we have had 1322 fire fighters to put these vehicles on the run over two shift patterns:
• 2 days (0800-1800), 2 nights (1800-0800), 4 days off (4 teams)
• 4 days (1000-2000), 4 days off (2 teams)

The current situation is that we have approximately 1300 fire fighters available and this figure is reducing by 7 per month due to retirements over the next 2 years (going down to approximately 1168).
Consideration needs to be given to resilience (all vehicles need to be on the run at all times) and the need to keep fire fighter movements from one station to another at a minimum.
The brief is to find the most efficient and resilient staffing arrangement to keep all vehicles available at all times in the following scenarios:
1. With no restriction on shift systems other than the European Working Directive
2. Whilst maintaining our 2/2/4 and 10/10 shift systems

Project: Decoding Brain Signals (Microsoft ML Competition)

Ayman Boustati — Sat, 16 Apr 2016 21:00:00 GMT

Microsoft has recently launched the "Cortana Intelligence Competitions" to promote their machine learning cloud service, Azure ML. The intiative is akin to Kaggle in the sense that they will be hosting online machine learning and data science competitions for enthusiasts to parcitipate in. The first competition went live a couple of weeks ago and is titled Decoding Brain Signals. Below is the description:

"The link between object perception and brain activity in visual cortical areas is a problem of fundamental importance in neuroscience. This competition asks you to build machine learning models in Microsoft Cortana Intelligence Suite to decode perceptions of human subjects from brain, specifically Electrocorticographic (ECoG) signals. The learning model needs to predict whether the human subject is seeing a house image (stimulus class 1) or a face image (stimulus class 2) from the ECoG signals collected from the subtemporal cortical surfaces of four seizure patients.

300 gray-scale images of houses (labeled as image class 1) or faces (labeled as image class 2), were displayed in a random order on a screen to the patients. ECoG signals were collected from the cortical surfaces of these patients during the experiments. The competition is to decode visual perceptions of these subjects from the ECoG signals, to predict whether the patient is seeing a house image (class 1) or a face image (class 2).

Each image stimulus is displayed to a patient for exactly 400 milliseconds, followed by a 400-millisecond inter-stimulus interval (ISI) where a blank image is displayed. A stimulus presentation cycle consists of the 400-millisecond ISI, followed by the 400-ms image stimulus. The ECoG signals were collected at the frequency of 1000 per second, i.e., every 1 millisecond there was a signal sample. Each patient has exactly 300 stimulus presentation cycles. In this competition, we share the ECoG signals of the first 200 stimulus presentation cycles and their stimulus types (1-50 are different house stimulus (stimulus class 1 in this binary classification task), and 51-100 are different face stimulus (stimulus class 2 in this binary classification task)).

Similar work on another set of 7 patients has been published at PLOS Computational Biology. The data we used in this competition was provided by the author of this paper, Dr. Kai J. Miller. This data was collected from 4 patients in the same experiment as the 7 patients in his paper. These 4 patients do not overlap with the 7 patients. However, reading through Dr. Miller’s paper may be helpful for you to understand the experimental settings, the ECoG signals, and features that were created in the machine learning experiment."

I think this will be fun for those of us interested in machine learning and data science. Plus, we get some experience with the Azure ML Suite which can look good on a CV. If you want more information you can check out the website.

Re: Project: Raindrops on the window of a moving train

Jonathan Skipp — Sat, 16 Apr 2016 14:34:16 GMT

Colm recently presented some work (Phys. Rev. Lett. 109, 168304 (2012)) on the kinetics of the Smoluchowski equation, one application of which is the aggregation of raindrops in a cloud. As far as I understand it (i.e. very little), the Smoluchowski equation assumes the raindrops are all well mixed, so the probablity of two drops of masses m1 and m2 doesn't depend on where they are in a cloud.

This problem seems like a very nice extension of that because it takes into account the spatial distribution of the raindrops on the window, the history of past mergers (tracks) and external factors (gravity and train speed, i.e. an effective gravity vector).

It would be nice to see whether/how a spatial Smoluchowski approach and an ant colony algorithm approach converge (although I have no idea how to approach either).

Google Hashcode Problems

Giovanni Mizzi — Fri, 15 Apr 2016 10:48:17 GMT

Google Hashcode is a Competition that takes place around February - March every year.
Teams from all over the world can participate, and they are composed by up to 4 people.

The problems that Google presents are definitely Real-World and non-trivial, but the biggest challenge is that they should be solved in less than 4 hours.

Here are the problems proposed in past editions. They seem to be mostly optimization problems, and I think that could be suitable for the retreat since they should be solvable in the amount of time we have.
https://hashcode.withgoogle.com/past_editions.html

The link contains 5 problems, with a small description of each one. The full problem statements and associated datasets are also downloadable.

Also I was thinking to participate to the next Google Hashcode, so it could be interesting to see if those problems are actually solvable that quickly.

Project: Raindrops on the window of a moving train

Yihe Lu — Tue, 12 Apr 2016 12:07:34 GMT

The pattern of raindrops has been bugging me for years and perhaps a model could be construct for it. One can easily observe the phenomenon when travelling on a train/bus and let us be grateful that we are based in UK that we do not need a huge budget for performing raining experiments. It impressed me the most when a raindrop runs down and leaves a track of tiny drops, which may affect some later raindrop's trajectory. Due to friction, different sizes of raindrops run at different speeds and some may stand still but eventually another runs, literally, into it; they merge and fall as one. The pattern would be dramatically changed, when the vehicle accelarates or slows down, or when the rain suddenly becomes more heavy or the opposite. I believe there are various ways to model the phenomenon. The one I am keeping in my mind is to apply ant colony algorithm, which is an agent-based model, and I know for such swarm intelligence things, there are many experts in the centre -- any advice is appreciated.

Project: Text mining and analytics

Rob Eyre — Tue, 12 Apr 2016 09:39:57 GMT

We could do some informal exploratory data analysis of, and create some generative models from, online text sources like project gutenberg (https://www.gutenberg.org/) or from song lyrics (http://www.azlyrics.com/). This would give us the opportunity to gain some experience with a type of data most of us have not used. We could try performing quantitative comparisons of different great works of literature, or create random song lyric generators for artists we like using something simple like a markov chain model or perhaps even something more complicated like a recurrent neural network. It's unlikely anything we do would be novel, but it would be fun. Any ideas welcome.

Project: Exploratory data analysis of UK MOT data

Alex Bishop — Thu, 17 Mar 2016 11:30:33 GMT

The government provides a data set of MOT outcomes since 2005 at:
https://data.gov.uk/dataset/anonymised_mot_test

The dataset includes many fields including:
Test date, test outcome (inc. failure reason), postcode area, car make, car model, car colour etc.

The aim of this project would be to perform data analysis on this dataset.

Possible directions this project could take based on interest:
- Applying simple statistical models/machine learning techniques to gain some
insight into factors contributing to failure reasons
- How to best visualise large datasets
- How to use a SQL database to improve your data-analysis pipe-line

If there was interest in some of the group continuing to work on this after
the retreat then geostatistical modelling could be an extension of the work
done at the retreat.

Suitable for: anyone with an interest in data science