9. IOT IN PAEDRIATRIC
RESEARCH
• Qualitative research is cumbersome for both the participant and the researcher
• People do not like doing questionaires
• Most of the time, participants must be physically present to complete research psychometrics,
scales, and questions so that researchers recieve their responses
• Online questionnaires exist, but they are equally boring
9
10. IOT IN PAEDRIATRIC
RESEARCH
• Worked with the a lab at the UCL Great Ormond Street Institute of Child Health to develop 2
smartphone apps
• Collects responses from patients remotely and securely on my server
• Saves responses to a cloud database (mySQL) through a 256-bit SSL connection
• User data is anonymised and encrypted on the server
• Researchers can collect responses in one safe, secure, and GDPR-compliant location
10
17. DENOISING MEDICAL
IMAGES
• Model for experimenting awake rodents animals in fear conditioning paradigm was proposed
for preclinical research (Brydges et al., 2013).
• Animals were scanned through an MRI whilst conscious during fear expression
• Why conscious? Conducted to better understand memory and emotion by accurately
observing the network level haemodynamic changes
• However, this model is highly susceptible to motion-related artifacts
17
22. DENOISING MEDICAL
IMAGES
• Data was too noisy to rely on intensity-analysis alone (combine with observed data)
• Subjective + Objective => Meaninful Training Data
22
Visual
Inspection
Voxel Data
subjective objective
training
inference
OK
BAD
24. DENOISING MEDICAL
IMAGES
• Narrowing the focus on specific datapoint range (brain extraction, remove first slice, remove
first 3 volumes)
• Motion Correction
• Slice Time Correction
• Sinebell filtering
• Timepoint repair: Deweighting/Despiking
• Signal Denoising
24
25. DENOISING MEDICAL
IMAGES
• Framework
25
Identify signal
dropout at
specific data
point to train
DL model
Feed results
into BOLD
analysis
pipeline
Improved
statistical
robustness
Improved
translational
validity
Minimise
impact to
statistical
power
31. CORTICAL CRYPTOGRAPHY
• The rubber hose cryptanalysis problem describes most computer security protocols as
inherently susceptible to coercion (via explicit knowledge).
• Although, implicit sensorimotor learning may provide a solution as the consolidation process
can be non-salient to the user.
• This has also been hypothesised to proceed in a bottom-up fashion, such that sequence
knowledge can be identified by enhanced performance on statistically predictable bigrams and
trigrams at test (Bojinov et al, 2012).
31
32. CORTICAL CRYPTOGRAPHY
• Embed a payload (target sequence) into a user’s implicit memory
• Tacitly express the sequence (through enhanced motor performance) on cue
• Prevent rubber hose attacks
32
33. CORTICAL CRYPTOGRAPHY
• An array of 32 keyboard keys discriminated by left/right hand
• Uniformly distributed and paired through an Euler circuit for optimal motor performance and
enhanced security.
• 3 keys per hand are selected on a pseudorandom basis using a software number generator
• 313,600 possible combinations ((16C3)2), which greatly improves upon Bojinov et al (2012)’s
setup, (4C3)2 = 16.
33
40. CORTICAL CRYPTOGRAPHY
40
Target Sequence occurs more frequently than the Foil Sequence
So, users should perform better (typing reaction time) on target sequence when tested
48. CORTICAL CRYPTOGRAPHY
48
Even if someone was aware of the password,
they would not be able to authenticate due to motor performance
49. CORTICAL CRYPTOGRAPHY
• Introduces the idea of “psycho-biometric authentication”
• High-level banking security
• Military research
• Something for the future
49
53. 53
ABOUT ME
• MScR Integrative Neuroscience - Edinburgh
• BSc Cognitive Neuroscience - London
• Research presented in Barcelona, Edinburgh, Oxford,
London, (and at 35,000ft!)
• Recieved scholarship from Facebook Artificial Intelligence
and IBM to study Data Science (Deep Learning with
PyTorch + IoT)
• Started IT consultancy freelance to fund my studies (Clients
include Institute of Child Health at UCL)
• Stayed in Japan (1 month), Taiwan (1 month for research)
www.alexanderlaurence.co.uk
Editor's Notes
SECURITY:
- RSA 2048 bits and AES 256 bits keys automatically generated on the fly! (Different keys every time for every client session)
- Password salt added automatically
- Password hash: SHA256
- Every information encrypted both sides automatically
- Full data encryption RSA + AES protocols
- Database SHA256 data encryption
- Restricted server access
- IP surveillance
- Anti-injection system
- Protection against brute force attacks
- Protection against hijacking
- Protection against network eavesdropping