Carding 4 Carders
Professional
Machines learn to understand the world like humans.
The brain develops an intuitive understanding of the world around it to interpret incoming sensory information. How exactly does this happen? Many scientists believe that the brain can use a process similar to "self-observed learning" in machine learning. This method allows models to learn from the similarities and differences between images without labels.
Two new papers from the Massachusetts Institute of Technology provide evidence for this hypothesis. Scientists have trained neural networks for a specific type of self-observed learning. The resulting models generated activity patterns very similar to those observed in the brains of animals performing the same tasks.
The results show that such models can study representations of the physical world and use them for accurate predictions. The mammalian brain may be using the same strategy, the researchers say.
"Our results seem to suggest an organizing principle across different brain regions and scales," says Aran Nayebi, co - author of one of the studies.
In the second study, the researchers focused on special neurons called grid cells that are involved in navigation. They trained a machine learning model based on introspection to perform the task of integrating the trajectory of movement and effectively represent space. After training, the activation patterns of the model formed lattice structures very similar to those that form grid cells in the brain.
Studies show similarities between models of self-observed learning and mammalian brain performance in performing cognitive tasks. The study may help to better understand how the brain functions, the authors say.
The brain develops an intuitive understanding of the world around it to interpret incoming sensory information. How exactly does this happen? Many scientists believe that the brain can use a process similar to "self-observed learning" in machine learning. This method allows models to learn from the similarities and differences between images without labels.
Two new papers from the Massachusetts Institute of Technology provide evidence for this hypothesis. Scientists have trained neural networks for a specific type of self-observed learning. The resulting models generated activity patterns very similar to those observed in the brains of animals performing the same tasks.
The results show that such models can study representations of the physical world and use them for accurate predictions. The mammalian brain may be using the same strategy, the researchers say.
"Our results seem to suggest an organizing principle across different brain regions and scales," says Aran Nayebi, co - author of one of the studies.
In the second study, the researchers focused on special neurons called grid cells that are involved in navigation. They trained a machine learning model based on introspection to perform the task of integrating the trajectory of movement and effectively represent space. After training, the activation patterns of the model formed lattice structures very similar to those that form grid cells in the brain.
Studies show similarities between models of self-observed learning and mammalian brain performance in performing cognitive tasks. The study may help to better understand how the brain functions, the authors say.
