Bio¶

Hello!, my name is Tarek

I am a Machine Learning Research Engineer at the Alan Turing Institute. The Alan Turing Institute is the United Kingdom's national institute for data science and artificial intelligence, founded in 2015 and largely funded by the UK government. It is named after Alan Turing, the British mathematician and computing pioneer.

I did my PhD in Applied Machine Learning at the Centre for Data Intensive Science at University College London (UCL). My PhD research focused on the application of Deep Learning and Model Compression algorithms for real-time classification of Astronomical transient events.

My thesis, for which I won the Perren prize for best CDT thesis 2022, can be freely downloaded here.

Towards the end of my PhD I was one of the 7 research students that were nominated by UCL to apply for the Schmidt Science Fellowship. While I was ultimately unsuccessful, my research proposal on "Efficient Learned Image Reconstruction and Compression Algorithms for Real-time Medical Image Analysis" is a good reflection of some of the areas of research I find most interesting.

Through my role at the ATI, I get to continue my applied research by investigating how the latest techniques in statistical signal processing and probabilistic machine learning can be used for Neural Compression to better enable low-latency energy efficient inference of machine learning models in resource constrained Embedded Systems -- a.k.a #tinyML

I am also a Research Software Engineering Fellow at the Software Sustainability Institute, where I help academics leverage tools and software engineering best practises for better research software : #bettersoftwarebetterscience

Find me online @tallamjr \(\in \{\) , , \(\}\)

Curriculum Vitae¶

Below is an overview of some of my profession and technical experience. You can download a more extensive version of my C.V below.

Education¶

	Ph.D Applied Machine Learning [Astroinformatics] University College London (UCL)	09.2017 - 09.2022
	Centre for Doctoral Training in Data Intensive Science Supervised by: Prof. Jason McEwen (Primary Advisor), Prof. Denise Gorse Thesis: Efficient Deep Learning for Real-time Classification of Astronomical Transients **Awarded Perren Prize for best thesis 2022**
	MSc Computer Science University College London (UCL)	09.2014 - 09.2016
	Department of Computer Science, Engineering Supervised by: Prof. Jason McEwen, Prof. Denise Gorse Project: Radio Interferometric Image Reconstruction for the SKA: A Deep Learning Approach
	MSci Astrophysics Royal Holloway, University of London (RHUL)	09.2007 - 07.2011
	Department of Physics Supervised by: Prof. Stuart Boogert Project: Analytical Methods of Stellar Spectra: Stellar Spectroscopy

Selected Recent Technical Experience¶

	Machine Learning Research Engineer The Alan Turing Institute, London.	10.2022 - present
	Investigating efficient machine learning and on-device deep learning (tinyML) for deployment of models in resource constrained settings. Researching neural data compression methods applied to deep networks. Building energy efficient machine learning prototypes and products through development of custom firmware/software.
	Machine Learning Researcher and Engineer Centre for Doctoral Training in Data Intensive Science & Industry, UCL, London.	09.2017 - 09.2022
	Contributed to numerous collaborative projects including preparatory work for a large-scale community machine learning kaggle competition. Led the design and development of astronet, an open-source scientific research software package introduces novel efficient deep learning architectures for low-latency high-throughput multivariate time-series classification. It also contains a machine learning pipeline which uses pyspark and polars for big-data processing, as well as tensorflow.datasets and tensorflow.distributed for efficient distributed model training. The lightweight architectures currently implemented in astronet have been deployed into live production machine learning systems by way modern model compression techniques for real-time classification of astronomical alerts.
	Machine Learning Researcher [TIN Internship] The Alan Turing Institute, London.	05.2021 - 11.2021
	Conduct research into unsupervised probabilistic machine learning and scalable non-parametric inference techniques for sequential latent factor modelling. Work collaboratively to investigate data and model compression techniques for deep neural networks.
	Research Software Engineer [Ph.D Internship] The Alan Turing Institute, London.	08.2019 - 03.2020
	Working in collaboration with the National Air Traffic Service (NATS), reinforcement learning (RL) was used to investigate machine learning methods to support air traffic controllers. Development of a RESTful API using flask was completed to allow for integration of both open-source and proprietary simulators. This has been followed by development of RL agents using openai-gym.
	Machine Learning Research Engineer [DSG Participant] The Alan Turing Institute, London.	12.2019 - 12.2019
	Invited to explore point cloud segmentation techniques as part of the SenSat - Semantic and Instance Segmentation of 3D Point Clouds Project. Investigating both semantic and instance segmentation in order to recognise objects such as roads, buildings, cars, etc. in a large 3D urban environment to enable safer autonomous vehicles on the road, automated asset management in urban planning, and accurate digital twin simulations. Benchmark deep learning methods implemented using pytorch.
	Graduate Teaching Assistant University College London (UCL)	09.2020 - 09.2021
	Assisting with grading and lesson planning for SPCE038: Machine Learning with Big Data. Lead the migration of tensorFlow 1.x to tensorFlow 2.x. Coordinated infrastructure setup for delivery of course through Jupyter Book
	Graduate Teaching Assistant London Business School (LBS)	03.2020 - 06.2020
	Assisting with grading and support for: E517: Python for Finance, QDE-APP: Applied Python Program- ming and CA22: Basic Python.
	Data Scientist [Ph.D Internship] Transport for London (TfL).	01.2018 - 04.2018
	Working in a team of 4 fellow PhD students, we investigated a variety of machine learning methods for train failure predication, that would be robust to highly imbalanced time-series data. With high cost implications for false-positives, we looked at algorithmic trade-offs that optimised accuracy at a low false positive rate. This work was co-supervised by academics at UCL and data scientists at TfL.