Intelligent pain's intensity detection and action-taking systems using generative adversarial network and reinforcement learning techniques on video processing

Pain can be expressed behaviourally in several ways; through the facial expression of a person, the way he/she pitches their voice or in their body language. This issue will be tackled in an integrated PhD project that aims at developing an integrated pain’s intensity detection system and action-taking system that are both trained to match a person pain type and manifestation.

The project will develop the pain detection system which will utilise cutting edge deep machine learning techniques, specifically adversarial neural networks, to learn facial and body expressions that are normally used to express levels of pain. The methodology is based on a multi-modal approach that integrates facial, vocal and body signals in order to model the pain patterns. Multi-sensory fusion techniques will be applied on the multi-modal data to feed the system with an appropriate input, while a state-of-the-art deep neural network will be trained to come up with best representation to detect the level and frequency of pain for a person.

A footprint of users’ behaviour associated with pain will be maintained and developed gradually through ongoing interactions with the users. Therefore, the system will get better over time as to how the resident is expressing their pain. The project will build on latest development of self-reflective reinforcement learning agent (Altahhan 2018) where a system can reflect upon own actions with respect to desired outcome and can correct itself online.

For further information please contact Dr Abdulrahman Altahhan, [Email Address Removed]

Objectives for project 1:

- To capture and analyse a person interaction with a virtual assistant such as Siri or Alexa in order to model closely their behaviour through their facial, vocal and gestures’ data coming from a video stream
- To create a pain detection model using deep learning and neural networks that will be trained using facial expression, vocal intensity and gestures
- To measure the accuracy of the new system and investigate potential usage in different scenarios

About the Research Centres

Deep and Reinforced Machine Learning Cluster is a recently formulated group in the School of Built Environment, Engineering and Computing, that is aimed at becoming a leading group in the UK and worldwide. Its mission is to develop state of the art machine learning algorithms and models that are game changers for people’s life to help in areas such as aging population, children with learning and physical difficulties and safety and protection applications. Its ethos is developing better solutions for better life. Currently the group has five academics, a good set of resources and state of the art multi-GPU server.

Candidate specification

• A degree in a relevant discipline
• Good modelling skills
• Experience or some knowledge of machine learning is a plus
• Proficiency in MATLAB or Python is a plus

Research proposal

Candidates should complete the online application form including a clear research proposal, applicant is asked to submit
• A CV
• A research degree application form
• A proposal of the research to be undertaken using the headings below as a guide.

o The proposal of the research can be up to four A4 pages in length (with references as an addition to the proposal) using type Arial 12 point.
o Please include at the start of the proposal the research project title.
o Please use the research project reference (DRL1 or DRL2) in the email subject line when submitting your application.

Interviews criteria

The criteria listed below will be used in both selecting those applicants who will be called for a short interview
• Qualifications, expertise and experience relevant to undertaking study for a PhD;
• Knowledge of the subject area that will allow the development of a focussed line of enquiry;
• Knowledge and understanding of research methods appropriate to undertaking a PhD in your chosen area of research
• Clarity on the original contribution that the completed PhD will make to the body of knowledge in the relevant research literature
• Scale and scope of the proposed research in terms of delivery within a three-year time-frame

Related Subjects

Machine learning, Deep Learning, Reinforcement Learning, Data Analysis, Computer Science, IT and Software Engineering.

Timeline

Closing Date: Midnight 9th of December 2019
Interviews Date: By or during w/c 16th of December 2019
Staring Date: 1st of Feb 2020