We held our first fully-digital community meetup on 1 May 2020, with over 100 colleagues coming together to share how they are using Machine Learning (ML) in the UK government and the public sector.
Machine Learning is a method for creating algorithms that enable computers to learn from data to make predictions. Examples of machine learning techniques are; reinforcement learning, deep learning and supervised learning. Machine Learning is also a sub-field of Artificial Intelligence (AI).
We were able to reach a wider proportion of the community with this being a virtual meet up. This also allowed for new areas and subject matter experts; below you will find information related to the subject areas that were covered.
Predictive Analysis in Criminal Justice
Davin Parrott, Principal Data Scientist at West Midlands Police, presented on the integrated offender management model. The model aims to reduce re-offending by enabling Integrated Offender Managers to make early interventions towards identified high-harm offenders. High-harm crimes include modern day slavery, gun and knife crime.
To create the model, data such as criminal history, network information and intelligence was user. Afterwards the model is tested using training data to predict the probability of becoming a high-harm offender. The final state of the model used a random forest with the XGboost algorithm, for multi-label classification on the next type of crime for the group predicted to be high-harm offenders.
Davin covered data ethics challenges that arise with using predictive analytics within the criminal justice system, and explained ways in which his team are addressing the challenges raised.
Data Science in Defence
A Data Scientist at the Defence Science and Technology Laboratory (DSTL) shared how they are using hierarchical reinforcement learning to explore its value for strategy planning in a simple 4-room environment, using the game engine Unity and the reinforcement learning toolkit ML Agents.
Within the 4-room environment, an agent is set a series of goals or waypoints by the commander, to reach their target in the quickest way possible. Through reward experience and exploration both the agent and commander continuously update their policies on how best to achieve the objective.
Piloting data analysis for early health interventions
Nik Haliasos, Digital Transformation Lead and Consultant Neurosurgeon at Queen's Hospital and visiting researcher at Alan Turing Institute, presented a pilot app using machine learning to assist triaging in accident and emergency.
Queen’s Hospital sees over 290,000 emergency department admissions per year. The app aims to support NHS analysts, nurses, and clinicians in decision-making and knowledge transfer. To create the app, the team used admissions and diagnosis patient data.
Logistic regression and random forest approaches were used for structured clinical data. Deep learning methods, such as long short-term memory (LSTM)-based models, were applied to analyse free text data. App users can input symptoms and patient characteristics to find out if the patient has a high probability of being admitted to hospital, therefore allowing for early interventions to take place.
Learning about the past
Luke Shaw, a Statistician from the Department for Education, took us back in time with the presentation of his co-authored RSS paper, “The Flying Bomb and the Actuary”. Following in the footsteps of a government actuary, the paper journeys through the probability and distribution of bomb landings during World War II. The paper uses analyses to verify the findings of R. D Clarke. The analysis is available in the RSS online library.
Data ethics play an important role in the creation of algorithms, because ‘overfitting’ models (where the model is trained too well and can't capture the ‘dominant trend’ in the data and therefore can’t make predictions) or using machine learning for malicious purposes can have significant consequences.
We were joined by Benedict Dellot, Head of AI Monitoring at the Centre for Data Ethics and Innovation. Ben shared with us the centre’s approach to tracking the opportunities and risks posed by AI and data-driven technology. The centre has produced public briefing papers on topics such as AI and personal insurance. The briefing papers are designed to build understanding of ethical and governance issues in deploying and developing AI.
Data Science Campus approach to coronavirus (COVID-19) pandemic
Dr Li Chen, from the Office for National Statistics Data Science Campus talked about a COVID-19 analysis project. The project techniques include computer vision and machine learning to extract vehicles and pedestrians from open traffic camera data; and using Google Cloud to create a pipeline and output statistics to help assess the impact of social distancing measures.
Li talked about large volumes of image/video data from real-time CCTV cameras across different areas at different illuminations, weathers, camera settings and similar; and the challenges of computer vision, machine learning and scalability to process such large databases.
Focusing on machine learning, Li evaluated and compared faster region-based convolutional neural networks against Google Vision API, and looked ahead at potential proposals to improve.
Thank you to speakers and co-organisers in helping us deliver the meet-up and communications.
Our next Community of Interest meeting will focus on data science skills and careers on the 8th July. If you are a data science practitioner, data analyst or data science enthusiast in the public sector, please feel free to join the data science #community_of_interests channel on the government data science Slack. If you have any questions onboarding or slack related questions please email email@example.com