In the partnership with Harvard University, DeepMind scientists have tested neuron behaviour in mice to prove the validity of the initial hypothesis that brain uses distributional reward predictions to strengthen learning algorithm.
In the article “A distributional code for value in dopamine-based reinforcement learning” published in Nature on January 15, scientists claimed that their findings “provide strong evidence for a neural realization of distributional reinforcement learning”.
During the press briefing, Matt Botvinick, DeepMind’s director of neuroscience research and one of the lead authors on the paper, has further explained the possible implications of their research, starting from a fundamental level:
“It gives us a new perspective on what’s going on in our brains during everyday life”
It basically updates one of the key neuroscience theories on reward brain system, giving hope to find new practical applications in mental health discipline, by providing doctors with more understanding on different types of neurons.
For AI, the future use of discovered distributional reinforcement learning could help humans to make the next big step forward to real world applications, added Botvinick:
“If the brain is using it, it’s probably a good idea. It tells us that this is a computational technique that can scale in real-world situations. It’s going to fit well with other computational processes”
Photo credit: Liu.se