James Kostas Ray
My areas of interests are in cosmology, galaxy evolution and unsupervised learning.
![jamesray](/data-intensive-science-industry/sites/data_intensive_science_industry/files/research-project/jamesray-feed.jpg)
1 January 2022
Project title:ÌýExploiting Galaxy Images with Deep Learning and Explainable AI LearningÌý
Research Group: Astrophysics
Supervisor(s): Prof Ofer LahavÌý
Introduction:Ìý
After working for several years as a data scientist, boredom had properly set in. Outside of the biggest brightest companies, data science and AI is relatively new to many companies, meaning learning while on the job was rather challenging as there was very strict limitation as to the methods we were allowed to use. While I was able to work in various infrastructure sectors, such as wastewater, rail and flood mitigation, none of these domains were as exciting as astrophysics. These factors were my motivations for returning to academia.
Since coming back to study the PhD with the Data Intensive Science & Industry with the Centre of Doctoral Training at Ïã¸ÛÁùºÏ²Ê, I’ve been able to learn state of the art techniques and practice them in training courses, gain industry experience in data mature environments, implement techniques in new areas in cosmology, gives lectures and talks about AI to students and other departments.
Areas which interest me (outside of astronomy) are computer vision, reinforcement learning, self/semi supervised techniques, generative modelling and system design. I am also an advocate for data literacy and enjoy giving talks about how these methods can be used.
I’m also a trained tennis coach.
Project description:ÌýÌý
The new era of big data in astronomy is here. The total volume of image data provided by the next generation of instruments is set to reach the same size as the internet was back in 2013. This has created a new challenge; how can we effectively use all this data? Existing analytical methods are unable to process such volumes in a sufficient timeframe, and don’t leverage the total data available. Artificial intelligence offers a set of tools capable of this.
With images dominating the data available it’s important to extract as much information out of them as possible. Is it feasible to extract important physical quantities from the images alone? Does leveraging other information better inform the system? Can an entire field be analysed in a one go? What biases can be accounted for? What are the most important features? Is the inference faster than existing methods? These are the types of questions that interest me.
First year group project:Ìý
London Data Company: Anti-Microbial Resistance (AMR) in Wastewater Systems: Apparently, I can’t escape wastewater and sewage. This project sought to predict spikes in AMR levels across the Welsh wastewater network and attempt to correlate these influxes with environmental, agricultural and human factors. By developing a model to use this information to nowcast and forecast AMR in various regions, we could extract important features which may have led to the increase in AMR so that governors, medical professions and farmers can make informed policy changes to reduce these effects.
The report produced by this model was presented to the Welsh government to help their strategy to reduce the presence of AMR in their population.
Placement:Ìý
Ìý
Ìý