Vectorized instructive signals in cortical dendrites

Nature

by Valerio Francioni

February 26, 2026

AI-Generated Deep Dive Summary

A groundbreaking study reveals that cortical dendrites carry vectorized instructional signals crucial for efficient credit assignment in biological systems, akin to machine learning algorithms. The research demonstrates how neural circuits can implement single-phase vectorized learning at the cellular level by processing feedforward and feedback information streams in separate dendritic compartments. By training mice to modulate the activity of two spatially intermingled populations of layer 5 pyramidal neurons using a neurofeedback brain–computer interface (BCI) task, scientists observed that dendritic signals contained information about task-related variables like reward and error. These signals were predictive of changes in overall activity during learning and disrupted when interfered with, indicating their critical role in the learning process. The study addresses a long-standing mystery in neuroscience: how biological systems solve the credit assignment problem, which is essential for adaptive learning. Unlike artificial neural networks that rely on backpropagation, the brain's solution remains elusive. The research provides direct evidence of vectorized instructive signals in dendrites, suggesting that cortical circuits spatially segregate credit-related information in apical dendrites, avoiding interference with feedforward inputs. This mechanism aligns with theoretical models proposing that feedback signals are processed distally in dendrites while feedforward signals are received perisomatically. By leveraging a visually guided neurofeedback BCI task, the researchers were able to isolate and analyze these vectorized teaching signals in real-time. They found that the relative magnitudes of somatic and dendritic activity could be predicted based on network dynamics and encoded information critical for learning. The signs of these instructive signals correlated with each neuron's causal role in achieving task goals, further supporting their function as teaching signals. This discovery not only advances our understanding of neural computation but also opens new avenues for designing biologically inspired machine learning algorithms that incorporate spatial segregation of credit assignment. The findings are significant for readers interested in the intersection of neuroscience

Verticals

scienceresearch

Originally published on Nature on 2/26/2026