Source-linked AI summary
Building machines that adapt and compute like brains
Nikolaus Kriegeskorte, Robert M. Mok
TL;DR
The paper addresses how to understand human-like learning and thinking as machines advance beyond component tasks. It proposes integrating cognitive and neural models, relating their representations, and testing both against brain and behavioral data.
Problem
As machines conquer more component tasks, the paper asks how to define human-like learning and thinking while explaining visual cognition mechanistically.
Method
The paper proposes combining cognitive-level and neural-level modeling, comparing internal model representations with brain representations, and applying both brain and behavioral constraints.
Results
Brain-inspired neural networks can recognize objects robustly under natural viewing conditions, while current models remain less capable of understanding visual scenes deeply than humans.
Takeaways & Limitations
A cognitive computational neuroscience should connect cognitive science and computational neuroscience by testing cognitive and neural models with both brain and behavioral data.
Takeaways & Limitations
Neural network models may take longer than humans to train, and brains can differ in neuron number and idiosyncratic neuronal specialization.
Abstract
from arXiv · showhide
Building machines that learn and think like humans is essential not only for cognitive science, but also for computational neuroscience, whose ultimate goal is to understand how cognition is implemented in biological brains. A new cognitive computational neuroscience should build cognitive-level and neural- level models, understand their relationships, and test both types of models with both brain and behavioral data.