09/10/2024

'Interdisciplinarity is important', says João Pereira

Reprodução

“The discoveries of John Hopfield and Geoffrey Hinton show how interdisciplinarity in science is important for scientific and technological development,” highlights IMPA researcher João Pereira, in reference to the Nobel Prize in Physics awarded this Tuesday morning (8) to the two pioneers in the field of Artificial Intelligence.

"These researchers drew inspiration from physical models of how neurons work to create innovative machine learning methods with immense applications, including compression, classification, and generation of images, video, text, and other types of data," explains the researcher.

Similar work is carried out at the Pi Center (IMPA's Center for Design and Innovation), of which Pereira is a part. Neural networks, machine learning, and algorithm development are part of the daily routine of the researchers in the "laboratory." However, the main distinction between Hopfield networks and other neural networks is the presence of connections between all neurons in the network. "In this type of network, training is also done differently: each neuron communicates with its neighbors in order to try to minimize the total energy of the system. Hinton extended Hopfield's idea to include stochastic connections, and in this way created Boltzmann machines . Stochastic connections have allowed us to learn probabilistic models about our datasets. These probabilistic models allow for better data analysis and enable its compression and even the generation of similar data," added Pereira.

Similar to these models, the most successful artificial intelligence systems are based on cognitive and memory structures that occur in nature. Among these models, convolutional neural networks stand out, used in image and video processing, and which are based on physical and biological models of animal optical systems; and Transformers networks, which use an attention model based on electrical and chemical processes that occur in our neurons.

The Nobel Prize in Chemistry, announced this Wednesday (9), also awarded work that used AI. Americans David Baker and John M. Jumper and Briton Demis Hassabis received this year's award for deciphering the secrets of proteins (molecules fundamental to our cells and, consequently, to life) through computation. Hassabis and Jumper used AI to map almost all protein structures already known to science. Baker, meanwhile, managed to create unprecedented proteins.