Dim(N) Week 1 - Neural state space alignment for magnitude generalization in humans and recurrent networks

Dim(N) Week 1 - Joonoh Park (07/04/24) Neural state space alignment for magnitude generalization in humans and recurrent networks

[Paper] Sheahan H, Luyckx F, Nelli S, Teupe C, Summerfield C. Neural state space alignment for magnitude generalization in humans and recurrent networks. Neuron. 2021 Apr 7;109(7):1214 - 1226.e8 https://doi.org/10.1016/j.neuron.2021.02.004

[Abstract] As my first presentation for the community members, I want to share the intriguing research about magnitude normalization and the definition of 'intelligent behavior' proposed by the C. Summerfield group. The authors coined the term 'number lines,' which are aligned in internal space, to describe the ability of intelligent agents to understand relationships and generalize knowledge across different contexts. They studied two agents, humans and RNNs (Recurrent Neural Networks), to solve a sequential magnitude comparison task while manipulating the local context of the magnitude and the RNN's memory. I hope this presentation helps you grasp the 'connection' between the domains of 'behavior,' 'neural,' and 'task' easily, while advancing our understanding of human intelligence.