Neural Networks

Computational models composed of interconnected layers of mathematical functions called neurons, loosely inspired by the structure of biological brains, that learn to perform tasks by adjusting their parameters on training data. Each neuron takes a weighted sum of its inputs, applies a nonlinear activation function, and passes the result to the next layer. Networks learn by comparing their outputs to correct answers and propagating the error signal backward through the layers using calculus, adjusting weights to reduce future errors in a process called backpropagation. Shallow networks with one or two layers were studied from the 1980s, but the modern era began around 2012 when deep networks with many layers, trained on GPUs with large datasets, began dramatically outperforming prior methods in image recognition. The transformer architecture, introduced by Google in 2017, became the dominant neural network design for language tasks and is the foundation of all major large language models. Example: AlphaFold2, developed by Google DeepMind and released in 2021, used a specialized transformer-based neural network to predict three-dimensional protein structures with accuracy rivaling experimental methods. It solved a 50-year grand challenge in structural biology, demonstrating that neural networks trained on biological sequence data can make scientific discoveries at a scale and speed impossible through traditional laboratory methods. Why it matters for AI: Neural networks are the foundational technology of the current AI era. Every major AI product, from large language models to image generators to protein design tools, is a specialized form of neural network. Understanding their behavior, failure modes (including hallucinations and distributional shift), and internal representations is central to both building effective AI applications and governing them responsibly, which is why interpretability research into neural networks has become a growing priority.

Explore the full Web3 Glossary — 2,062+ expert-curated definitions. Need guidance? Talk to our consultants.