Eli Sheppard CO09

Dr Eli Sheppard studied at Truro School in the 2000’s where his fondest memories were camping trips to Minions with Mr Golds (TS 1968 – 2007).   

Eli started playing guitar aged nine, and took lessons with Dave Stevens at Truro School, a skill he continues today. “Dave Stevens was a big influence on my taste in music – we used to jam the blues a lot which developed into my love of jazz and blues.” 

After completing his A-Levels, Eli attended Heriot-Watt University in Edinburgh to read Electrical and Electronic Engineering  graduating with a master’s degree with distinction. Eli then joined the Edinburgh Centre for Robotics Centre for Doctoral Training (CDT). In 2016 he was awarded an MSc in Robotics and Autonomous Systems with distinction from the CDT, which led to his PhD studies.

The topic of Eli’s thesis was “Multimodal Representation Learning: an unsupervised approach to symbol grounding*”, in which he focused on the use of machine learning to jointly learn language and computer vision without human intervention. The result of his PhD was a Machine Learning (ML) System which allows robots to interactively learn to recognise and name objects as well as their properties (e.g., colour, shape, size). Eli deployed this system on the iCub robot. The iCub is a state-of-the-art humanoid robot developed by the IIT (Istituto Italiano di Tecnologia), for exploring developmental robotics. Eli explains: 

“I was attempting to develop a system which could learn in a similar way to how babies learn to speak and recognise objects (think babies’ first words sort of thing).  The robot is child-like, designed for researching how children learn and how we can use this new knowledge as well as usual teaching methods to teach robots.” 

Upon completion of his PhD, Eli wrote to Mr Mark Vanstone (TS Director of Studies), to update him on his progress.  Mr Vanstone was Eli’s form tutor when he started at Truro School and taught him physics and chemistry at various points during his education.  

“Mr Vanstone was always encouraging my curiosity, giving me more in-depth explanations of physical phenomena, and I attribute part of my academic success to his excellent teaching. I believe that teachers and good schooling environments are vital to the success of individuals and society alike, so I am very grateful for the excellent education I received both at primary school at Marlborough and secondary at Truro School.”  

Eli’s working life began at the University of Lincoln (UoL) developing the vision and control system for a mushroom picking robot, before he moved to work for an Oxford-based start-up called Living Optics.  

“I love being able to freely explore ideas and develop my own understanding or to satisfy my own curiosity. Most recently, a colleague and I developed a framework in ‘Python’** which allows for the automatic definition of neural network architectures – this allows us to rapidly experiment with different ideas. 

“We are developing a ‘Hyperspectral Camera’ (a camera which sees in at least 20 colours, as opposed to the three (red, green, and blue) which are used in a normal camera). My role as the company Computer Vision Engineer mostly entails working as a software engineer, developing Auto-ML (automated machine learning) techniques to enhance the capabilities of our camera. In this case, Auto-ML is using algorithms which automatically optimise the architecture of the neural networks we are using to produce our hyperspectral images i.e., we are having a machine learning system design a machine learning system to produce hyperspectral images from our sensor data.”

Having just turned 30, we asked Eli about what his plans are for this new decade in his life? 

“In the short term my plans remain with Living Optics. It’s a company in its infancy at just over a year old but has the potential to be a major player in the computer vision and remote sensing industry. Long-term I have plans to buy some land and build a self-sufficient home using the principals of permaculture. I’m passionate about reducing my environmental impact and would like to help others reduce theirs too – to that end I would like to write a book or perhaps make a video series detailing how to do everything from generating your own electricity to growing your own food.”   

And advice for anyone considering a career in machine learning? 

“Whilst understanding the mathematical underpinnings is important, developing strong software development skills are equally vital. I was hired not only because of my experience with machine learning but also because I have experience developing software systems from the ground up.  Careers in software development require solid problem-solving skills; many people can program, not so many are good at breaking problems into pieces and finding scalable ways of solving them. In that regard, time spent in the design and technology labs with Mr Tall (1980 – 2014) learning how to build furniture was probably more beneficial to my software development skills than learning how to solve quadratics in maths class. Once you understand how to break a piece of furniture into components, you can adapt that understanding to making more complex things. First you build a box, then a table, then a chair… eventually you can build a house. Software development is the same; first you solve simple problems, then more and more complex ones by breaking the hard problems into components.

In general, my advice would be to focus on finding what you enjoy doing and making that into your career – it’s all well and good chasing a pay cheque, but you’ll be much happier working at something you find valuable for its own sake.”

_________

*Symbol grounding is the bidirectional action of learning the meaning of a symbol e.g., learning that the word apple refers to the object apple and vice versa. Babies (and robots) have to learn to link the auditory perception of the word apple with the visual perception of an apple. 

**Python is an interpreted high-level general-purpose programming language.