Neural network

A neural network is a machine learning model composed of layers of simple connected computational units, loosely inspired by biological neurons, which jointly learn to represent complex patterns in data by adjusting connection strengths during training. Deep learning refers to neural networks with many layers, which have driven much of the recent progress in AI across image recognition, language and other domains.¹

Why data availability often limits neural network capability more than architecture does

A neural network's capability generally scales with the amount and quality of training data and computational resources available, which is why data availability, not just model architecture, is often the limiting factor in a specific application.

Where neural networks apply

Not on the drawing; a neural network is a modeling technique used within software, not something specified by a part's requirements.

Common mistakes

Assuming a larger or more complex neural network is automatically better for a given task; the amount and quality of relevant training data, and the appropriateness of the architecture for the task, matter as much as raw size.

Is a neural network the same as deep learning?

Deep learning specifically refers to neural networks with many layers; neural network is the broader term covering both simple and deep architectures.

Do neural networks actually work like the brain?

Only loosely inspired by biological neurons; the actual computation and learning process differs significantly from how biological brains work.

What limits a neural network's performance?

Often the amount and quality of relevant training data and computational resources, more than the specific architecture chosen.