Results 321 - 330 of 24331

In this talk I will present my vision of how combining the power of Brains & Deep-Networks (DNNs) can lead to significant breakthroughs in both domains and potentially bridge the gap between Brains & Machines. I will show how combining the power of Multiple Brains (“the Wisdom of a Crowd of Brains”) may lead to new breakthrough discoveries in Brain-Science, allow mapping of information between different brains (with NO shared data), and lead to new ways of training and interpreting artificial DNNs.

Many natural motor skills, like speaking or locomotion, are learned through trial-and-error over development. It has long been hypothesized that dopamine plays a critical role in this process, motivated by artificial learning experiments. Dopamine in the basal ganglia is thought to guide reward-based trial-and-error learning by encoding reward prediction errors, decreasing after worse-than-predicted rewards and increasing after better-than-predicted ones. Similarly, by changing perceived song quality with distorted auditory feedback, our previous work in adult zebra finches showed that dopamine in Area X, the singing-related basal ganglia, encodes performance prediction error: dopamine is suppressed after worse-than-predicted (distorted syllables) and activated after better-than-predicted (undistorted syllables) performance. However, it remains unknown if the learning of natural behaviors, such as developmental vocal learning, occurs through dopamine-based reinforcement. Here we tracked song learning trajectories in juvenile zebra finches and used fiber photometry to monitor concurrent dopamine activity in Area X. We found that dopamine was activated after syllable renditions that were closer to the eventual adult version of the song and suppressed after renditions that were farther away. Furthermore, the relationship between dopamine and song revealed that dopamine predicted the future evolution of song, suggesting that dopamine drives behavior. Finally, dopamine activity was explained by the contrast between the quality of the current rendition against the recent history of renditions, consistent with its hypothesized role of encoding prediction errors. Reinforcement learning algorithms explain learning in reward-based laboratory tasks as well as drive autonomous learning in artificial intelligence. Our results suggest that complex natural behaviors in biological systems can also be learned through dopamine-mediated reinforcement.

Abstract not available.

We study agents playing a pure coordination game on a large network. Agents
are restricted to coordinate locally, without access to a global communication device,
and so different regions of the network will converge to different actions, precluding
perfect coordination. We show that the extent of this inefficiency depends on the
network geometry: on some networks, near-perfect efficiency is achievable, while on
others welfare is strictly bounded away from the optimum. We provide a geometric
condition on the network structure that characterizes when near-efficiency is attainable.
On networks in which it is unattainable, our results more generally preclude high
correlations between outcomes in a large spectrum of dynamic games.

Weaver ants build closed nests by collectively bending and stitching leaves using their own bodies as tools. Using controlled laboratory experiments, we show that in simple configurations the entire construction process can be quantitatively captured by a local rule—the “zipping” heuristic—which reliably produces closed, rigid nests. We then challenge the colony with construction puzzles and find that it can still converge to viable solutions even when this heuristic predicts failure. These results suggest that information processing emerges from the coupled dynamics of ants and leaves, rather than from individual cognition. This system provides a tractable model for studying the mechanisms and limits of collective cognition.