Page 14 - Spring 2024
P. 14
AI applied to create new rugged and physical testing results underscored the
Microstructures methodology’s validity.
Researchers at the Massachusetts Institute of A ‘Neural-Network Accelerated Multi-
Technology Computer Science and Artificial Objective Optimisation algorithm’ resulted,
Intelligence Laboratory have developed an which navigates the intricate landscape of
innovative AI system which utilises simulations microstructure design, uncovering configurations
alongside testing to forge new microstructures. that exhibit near-optimal mechanical properties.
By combining computational design with The algorithm thus serves as a self-optimising
physical experiments and advanced neural mechanism that constantly refines its
networks a new approach to material creation predictions so as to more closely reflect actual
has been devised that has led to the creation of
results.
microstructured composites that redefine the
benchmarks for toughness and durability in Ultimately it is aimed to design fully automated
engineering materials. laboratories that minimise the need for human
intervention and maximise efficiency.
Neural networks are deployed as ‘surrogates’ for
simulations, streamlining the material design Reference: Fowle, H., Electronic Specifier, 19th. February
process. An evolutionary algorithm enhanced by
neural networks efficiently pinpoints the highest
performing samples combining stiffness with AI challenges uniqueness of Fingerprints
toughness.
Engineers at the University of Columbia have
The methodology commenced with the used AI to demonstrate that fingerprints from
fabrication of 3D-printed photopolymers, each different fingers of the same person are not as
no larger than a smartphone, but considerably unique as previously thought.
thinner. These were then subjected to a unique
The team used a public US Government
ultraviolet light treatment followed by tensile
database of around 60,000 fingerprints to feed
testing using an Instron 5984 machine. This
pairs into a deep contrastive network AI system.
physical testing, combined with advanced
Sometimes the pairs belonged to the same
simulations within a high-performance
person (but different fingers) whilst in other
computing framework, enabled the prediction
and refinement of material characteristics with cases they belonged to completely different
unprecedented accuracy. people.
A key aspect of the research was the The AI system, adapted from an advanced
development of a method to coalesce different framework, improved its ability to discern when
materials at a microscopic scale. This process, seemingly unique fingerprints were from the
characterised by a complex pattern of miniscule same person with the accuracy for a single pair
droplets that amalgamated rigid and subtle reaching 77 per cent, whilst with multiple pairs
substances, achieved a harmonious balance it significantly increased to the point where
between strength and flexibility. The close forensic efficiency was potentially enhanced
correlation between simulation outcomes more than ten fold.
A notable achievement was the identification
of a new type of forensic marker used by the
Visit our AI, which was quite different from the standard
‘minutiae’, that is branchings and endings
website! in fingerprint ridges. Instead, the AI focused
on angles and curvatures in the centre of
fingerprints.
Why not head down to our
website and read more
The findings suggest that AI may have
news in our blog section? ‘transformative potential’ in some otherwise
established fields. In particular, AI can provide
Updated daily with
new insights from existing data and it would
hand-picked articles
appear that new AI-led discoveries are waiting to
from around the web...
be made.
institutionofelectronics.ac.uk Reference: West, P., ‘Not every Fingerprint is unique, AI
discovers’, Electronic Specifier, 11th. January
14
Subscribe to Electron here https://membermojo.co.uk/ie
