BrainChip raises $21.5 million to create tech that thinks like a human
ASX-listed artificial intelligence software and hardware company BrainChip has raised $21.5 million from investors to fund the development of its "general purpose neural processing unit", which it says has been designed to mimic the function of the human brain.
Unlike a central processing unit, which traditionally functions as a computer's brain, BrainChip's new development – which it is calling AKIDA – is human-like because it considers many probabilities and conducts numerous calculations simultaneously.
The development of AKIDA is based on BrainChip's spiking neural network (SNN) technology, which so far has been applied to image recognition and is used to detect patterns. It is due to be released in late 2018
BrainChip chief executive Louis DiNardo told The Australian Financial Review AKIDA would have a broader use and would be able to detect patterns in everything from audio data to cyber security threats and in commodities and equities trading.
"We're going from a $3 billion opportunity in civil and commercial surveillance to opening up a market that I can't even size yet as we enter cyber security and fintech," he said.
"The combination of visual inputs and the ability to recognise patterns also opens up applications in Advanced Driver Assisted Systems [in autonomous vehicles]. They have to be able to recognise things like street signs ... then there's also things that can't be identified in advance. We call this unsupervised learning ... with an autonomous vehicle it runs into things every fraction of a second that aren't familiar and can't be modelled."
BrainChip, which was founded in 2013 and went public via a backdoor listing in 2015, is also preparing to expand its products to Australia.
Last week it locked in its first deal for its one of its hardware products, the BrainChip Accelerator, with a major European car manufacturer.
The BrainChip Accelerator increases the performance of object recognition provided by BrainChip's software, providing a sevenfold improvement in the speed of image recognition compared with traditional neural networks provided by graphics processing units (GPUs).
BrainChip's SNN software is already in use by casinos and the French defence forces to detect terrorist threats and respond to emergencies.
Rather than focusing on facial recognition when used by defence agencies, BrainChip's technology pinpoints patterns, such as someone's cufflinks or a distinctive T-shirt or tattoo.
Mr DiNardo gave the example of a real case in France in which its technology was applied to 7.5 hours f video footage from three cameras, making up thousands of frames.
The investigators focused on the person of interest's T-shirt and even though the cameras were poor quality, detected him at public transport locations in the city. They were then able to correlate the timestamp and location to his subway card and bus pass and by the time the person in question arrived home, the police were waiting.
"There's no need for machine training where you need hundreds of thousands of learnings, we can just take an image and circle in on the T-shirt or cufflink and tell it that's the person," Mr DiNardo said.
"We extract that spike map, which is a small amount of data, and it can be processed quickly."
BrainChip, which has been in a trading halt but last closed at 24.5¢, will also use the capital injection, which was first reported by Street Talk, to fund a sales and marketing push, targeting equipment manufacturers and other potential partners.
Mr DiNardo said it would also free him up to be more client-facing and enable the business to aim for profitability in later 2018.
(20min delay)