Dr Jane Bromley

Lecturer In Cyber Security

jane.m.bromley@open.ac.uk

Biography

Professional biography

I've done something fun recently - I was interviewed about my views on Neural Networks, Machine Learning and whether we should fear AI (the answer is "no" but do fear how it is used and by whom!) This was linked to the OU's support of the Ri's 2023 Xmas Lectures on AI,

I grew up and went to school in High Wycombe, a leafy, furniture making town in the Chilterns. This was followed by a BSc in Physics and PhD in Biophysics (1987) at Imperial College (IC). I then held a joint postdoc funded by the Wellcome Trust between IC and the Royal London Hospital during which I constructed a visual psychophysics lab in a small room off the Neurology Ward. I was instrumental in measuring the effectiveness of neurological treatments. I also occasionally visited the museum to see the skeleton of the "Elephant Man" (no longer on public show).

My research at this time was on Visual Dysfunction and my most interesting subject was a person with Visual Agnosia. As a result of this research I became interested in modelling the human visual system and with a travel grant from the Royal Society I went to Bell Labs in 1990 to learn about artificial Neural Networks. [By the by I think I was the first person ever to negotiate a job via email - this was back in 1988!] Here I became involved in the work of the Adaptive Systems Research Department on recognizing characters - handwritten or machine printed, on paper or a tablet. Initially this was applied to automating mail sorting for the US Post Office, but later the technology was (and is) used by banks around the world to read and process checks. I also pioneered a neural network architecture called a Siamese Neural Network which I used for Signature Verification. You can see a demo of it from 1993 here. Since then it has found useful applications in things like facial recognition to open your phone or at passport control. Less good uses have been in its uncritical application for example in policing systems which has led to miscarriages of justice.

In case you think I'm just about Computing and AI, I have wide ranging interests. I've tutored school children in Robotics, including representing the UK in RoboCup Junior at the World Finals in Suzhou, China in 2008; grow bananas; used to drive a 1970 Triumph Spitfre Mark III, but now have a much too powerful 1990 ZR1 Corvette; and can run 5k in under 30 minutes.

Research interests

Machine Learning in Astronomy

This collaboration started 7 years ago with Prof Stephen Serjeant and just one student. Now, along with Hugh Dickinson and Helen Fraser we share a large and lively group of PhD students who are using Machine Learning to, amongst other things, find Lensed Galaxies, improving angular resolution on Wide-field (but blurry) Extragalactic Surveys and to untangle the vast amounts of data for Ice Astrochemistry.

Complexity and Design (at the OU)

A highly interdisciplinary area involving fundamental research into the methods of complex systems science supported by research into the design of its applications.

Information extraction and data mining (at the OU)

This work was done under the EU FP7 Project: agINFRA: a data infrastructure to support agricultural scientific communities. I worked on information extraction from the legacy biodiversity literature.

Convolutional Neural Networks (at Bell Labs, NJ)
Initially for modelling the visual system, but I never got very far on this, instead becoming engrossed in projects that applied variants of the famous LeNet architecture created by Yann LeCun to a broad variety of problems in image recognition and signal processing

handwritten and machine printed address readers (for the US post office)
faxed form reader (for AT&T business forms)
use of a special purpose chip (called ANNA) that implemented a handwriting recogniser in silicon (with Bernard Boser and Eduard Sackinger)
automatic teller machine that can read checks and bills deposited by bank customers
signature verification with a Siamese time delay neural network

The last 2 were collaborative projects with NCR.

My unique contribution, apart from training software that ran like the Disney version of the Sorcerer's Apprentice and couldn't be stopped, was introducing training on “rubbish” to improve rejection performance.

Human Vision (at Imperial College and the Royal London Hospital with Keith Ruddock and Chris Kennard)
The application of psychophysics (the investigation of the relations between physical stimuli and sensation) to the understanding of the normal mechanisms of visual processing, but also to the study of visual abnormalities in order to offer practical assistance to clinical patients. This was experimental work using computer monitors, Maxwellian View Optical Systems , essentially an optical bench with mirrors, lenses, filters and light sources and the wonderful W. D. Wright Trichromatic Colorimeter built in the 1920s.

the role of blue-sensitive cones on spatial processing
parallel and sequential visual processing (particularly in dyslexia and amblyopia)
visual perseveration
visual agnosia
hemianopia (in particular blindsight)
various central visual defects

Teaching interests

Lots!

Physics labs and classworks at Imperial College
Maths from Primary School students through to High School

At the Open University since 2006:

ME627 Developing geometric thinking for which I was critical reader (2006) for the course text book
TM359 Systems penetration testing
TM358 Machine learning & artificial intelligence
T855 Team engineering
TM256 Cyber security
M269 Algorithms, data structures and computability
TM129 Technologies in practice
T212 Electronics: sensing, logic and actuation
T122/TXY122 Career development and employability
T227/TXY227 Change, strategy and projects at work
TMXY125 and 225 Professional practice
TT284 Web technologies
T176 Engineering: professions, practice and skills - a week of face to face Teaching at the Engineering Residential School
T192 Engineering: origins, methods, context.

Impact and engagement

Projects

Forensic 'Big Code' Analytics in Secure Software Engineering

The aim of this project is to collect forensic evidence from the "Big Code" that violate security requirements in the past and predict the risks of security incidents in the future. "Big code" consists of a variety of artefacts including security goals and requirements, software licenses with terms and conditions, bug reports and code patches at development time, and microservice logs at runtime. Apart from being large in quantity and high in diversity, the big code is also evolving continuously over time. Substantial human efforts have been spent on identifying forensic evidence from the big code, in order to identify computer-related frauds or other security-related incidents. Therefore, the project will focus on three objectives: 1) To identify forensic evidence from unstructured big code, mostly in natural languages, by selecting or extracting the relevant features in software artefacts; 2) To predict changes in structured big code, mostly in programming languages, by evaluating machine learning models against the precision/recall metrics about security-related incidents; 3) To maintain the predictability of forensic analytics continuously, during evolving software development, by updating the learning models incrementally according to the newly arrived big code.