Joshua Talks Email & Phone Number

Heidelberg, Baden-Württemberg, Germany

PhD student in the Kreshuk group investigating machine learning for bio-image analysis.

Cambridge, England, United Kingdom

I worked in the R&D department at Oxford Gene technology, investigating and implementing AI algorithms for analysing images produced by a MI FCM (Molecular Imaging Flow cytometry) machine. This combines FISH (fluorescent In-Situ Hybridisation) and FCM to produce an automated system that is hoped will increase the testing efficiency and precision for genetic diseases such as Leukaemia.I worked on image analysis, using Machine Learning to analyse subcellular fluorescent images to predict fluorescent spots. One of the major challenges with the dataset was it was unlabelled and labelling it would have required significant time and labour cost. Firstly, I developed an automated annotation pipeline that combined thresholding and filtering stages along with segmentation algorithms to produce semantically segmented images. I then implemented a Convolutional Neural Network (CNN), using the labelled training data that combined a U-Net base along with star-convex polygon approximations to accurately predict fluorescent spot shapes. As an alternative approach, I implemented the previous CNN in a semi-supervised loop that combined clustering analysis to improve semi-supervised classification. This framework combined Fuzzy C-Means (FCM) clustering on unlabelled data to reveal the underlying data space structure, along with a self-training classifier pipeline (CNN previously discussed). The framework aimed to iteratively convert unlabelled data into labelled data, training and improving a classifier at each iteration. The advantage being it only requires a small subset of initially labelled data. I also started looking into active learning as an additional method for training a classifier.Over the project I conducted a literature review, reading and implementing papers on many areas of image analysis from Clustering and Deep Neural Nets to thresholding and image processing. I coded throughout the project in Python, extensively using Keras, Scikit-learn, Open-CV and Pandas.

Cambridgeshire, England, United Kingdom

While at PragmatIC I worked within the Data Test Team on a full stack project to produce a custom web-based data analysis dashboard to allow more flexible and accessible analysis of their RFID chips production and development data.I used SQL to dynamically querying the company database, containing millions of entries, depending on interactive inputs from the user. Allowing the user to easily select any rational combination of data that is was of interest for further inspection. I then used Pandas to sort/format the data into appropriate data-frames and created interactive visualisations using Altair plots. The Dashboard had several layers of analysis and visualisation allowing the user to combine any stage of the production process and click through interactively to different levels with built in analysis of the data or all the way to raw data traces and tables.For the Dashboard I used a Python backend with a Django web framework with a simple modular design to allow for further development and debugging. I developed the user interface and structure of the dashboard using a combination of HTML, CSS styling and dynamic JavaScript to create interactive responses to user inputs. The code also had to be fully and extensively documented to fit into the companies current software standard, throughout the project I maintained best practices for workflow such as using GitHub for version control and working across multiple branches, with scheduled code reviews before merging to a master branch. The project was a great way for me to further strengthen my Python coding and learn and develop skills in new languages such as HTML, CSS and JavaScript. I was constantly faced with new challenges and had to research and learn new skills to overcome them.

Cambridge, England, United Kingdom

While at Huxley Bertram I worked in a combination of the software department and the design office on many different projects, I was part of the whole process from the conceptual design to the implementation of ideas and testing and manufacturing of finished products. I frequently encountered problems outside my expertise and solved this by studying case studies and by learning from the professional engineers around me.As part of the software department I strengthened my Python abilities and worked on the setup of the machine-user interface and also on my own independent project to design and produce an automated microscope inspection rig to detect microscopic defects in products. This involved designing and manufacturing a suitable inspection system that could be integrated into existing systems and then coding a test sequence to automatically inspect products for microscopic failures.As part of the design team I worked with SolidWorks on several design projects such as automated silicon chip production unit, a Vectura inhaler test unit, a high precision tablet press for research into news medicines and a hydrogen aircraft panel permeability test unit. Each project provided its own difficulties and constraints and challenged me to work with the team to create bespoke solutions for each task.The internship not only developed my technical skills, but also developed my ability to work effectively and efficiently in a team within the limitations of time and cost on an industrial scale project, maintaining the highest standard of work while sticking to deadlines in a real-world setting.

Master Of Engineering - Meng, Information And Computer Engineering, Bioengineering, Honours With Merit (1St Project, Ii.1 Modules (32 Percentile))

Investigation of a Novel DNA Alignment Algorithm (Masters Project): As technology advances it will become feasible to store data.

Bachelor'S Degree, Information And Computer Engineering, Bioengineering

Maths, Further Maths, Physics, Chemistry, A*, A*, A*, A*