Definition
Neural network is a neural network is a computing system made of layers of simple connected units that learn patterns from data by adjusting the strength of their connections, and it is the basis of modern AI.
A neural network is a type of software loosely inspired by the brain. It is built from many small units, often called neurons, arranged in layers and connected to each other. Each connection has a number attached to it, called a weight, that controls how strongly one unit influences the next. Learning means finding good values for those weights.
Data enters the first layer, such as the pixels of an image or the words of a sentence turned into numbers. Each unit combines the signals it receives, applies a simple function and passes the result on. After several layers, the final layer produces an output: a label, a prediction or the next word.
At the start, the weights are random and the answers are poor. During training, the network makes a guess, compares it with the correct answer from the training data, and measures the error. Gradient descent then adjusts every weight slightly in the direction that would have reduced that error. Repeat this across millions of examples and the network gradually gets good at the task.
An everyday comparison is a recipe with thousands of dials. Nobody sets the dials by hand. The network tries, tastes the result and turns each dial a tiny bit, over and over, until the dish comes out right.
A network with many layers is called deep, and training one is called deep learning. Extra layers let it build up understanding in steps. In an image model, early layers might detect edges, middle layers shapes, and later layers whole objects. The total count of adjustable weights is the model's parameter count, which is the figure people quote when they say a model has billions of parameters.
A neural network does not follow written rules. It finds statistical patterns, which makes it flexible and also hard to inspect. Its mistakes can be confident and strange, and it can only be as good as the data it learned from. Understanding this explains many behaviors of modern AI, from impressive fluency to occasional hallucination.
Not exactly. Deep learning means training neural networks with many layers. A neural network with only a few layers is still a neural network but is not usually called deep.
Only loosely. They borrow the idea of connected units, but they are mathematical functions trained by calculus, and the brain works very differently.
Amazon is in talks to buy Decart, whose software ports AI models between chips, for about $7 billion, the Wall Street Journal reported Sunday. Read it as a bet on Amazon's Trainium chips, and on making your workload easier to move onto them.
MIT CSAIL researchers say a frontier LLM, used as is, read and rewrote neural network weights without gradient descent, lifting a small Qwen model's MATH-500 score from 7.9% to 48.1%. It is an early blog post, not a peer-reviewed paper, and still the strangest result of the week.