|
|
Scientists from the National Institute of Standards and Technology (NIST) and the University of Maryland in the USA have presented an energy-saving artificial intelligence (AI) hardware in the journal Physical Review Applied that has passed a virtual wine tasting. The neural network is supposed to imitate the processes of the human brain when tasting wine. When wine lovers taste a new wine, the neural networks in the brain process countless data with each sip. The synapses in the neurons weigh the importance of each piece of data - acidity, fruitiness, bitterness, for example - before passing it on to the next layer of neurons in the network. As the information flows, the brain uses it to analyse the type of wine. For this, the human brain needs an estimated average power consumption of 20 watts. AI systems need a thousand times that for the same calculations. One branch of AI research is therefore looking for less energy-intensive alternatives - with success.
The research team trained the network with a dataset of around 150 wines made from three grape varieties. Each virtual wine had 13 features to be considered, such as alcohol content, colour, flavonoids, and impressions of ash, alkalinity and magnesium. Each feature was assigned a value between 0 and 1 that the network could take into account when distinguishing one wine from the others. The AI system was then subjected to a virtual wine tasting with the entire data set, which contained 30 wines that it had not processed before. The system passed the sample with a recognition rate of 95.3 per cent. For the 30 wines it had not been trained on, it made only two errors. The researchers consider this a great success "It is a virtual wine tasting. The tasting is done by analysers, which are more efficient but less fun than doing your own tasting," said NIST physicist Brian Hoskins.
(ru / source: Nist; photo:123rf.com)
More on the topic:
AI system from 'Tastry' aims to optimally match cuvées