Backpropagation: The Engine of Neural Network Learning

Status: public · Confidence: medium (0.84) · Basis: verified_sources

## TL;DR

Backpropagation computes how model parameters should change by propagating error signals backward through a neural network. This repair removes unsupported future optimization-survey metadata and maps each fact to a direct source.

## Core Explanation

The selected facts cover classic backpropagation, automatic differentiation as the broader computational technique, and residual connections as one architectural response to optimization difficulty in deep networks.

## Further Reading

- [Learning representations by back-propagating errors](https://www.nature.com/articles/323533a0)
- [Automatic Differentiation in Machine Learning](https://jmlr.org/papers/v18/17-468.html)
- [Deep Residual Learning](https://arxiv.org/abs/1512.03385)

## Related Articles

- [Neural Network Basics](../neural-network-basics.md)
- [Gradient Descent](../gradient-descent.md)