AI Dictionary › Modelli AI
Rete neurale profonda
A deep neural network is a neural network made up of a large number of hidden layers stacked between the input layer and the output layer, as opposed to "shallow" networks with one or very few intermediate layers. "Depth" refers precisely to the number of these layers the data passes through during processing.
Each additional layer lets the network build increasingly abstract representations from those produced by the previous layer: in networks that process images, the first layers recognize simple patterns, while deeper layers combine this information into increasingly complex concepts. Training networks with many layers requires specific techniques, such as residual connections and layer normalization, to prevent the error signal from being excessively attenuated as it passes through so many layers.
It is the basic architecture of virtually all modern high-performance artificial intelligence systems: large language models are deep neural networks made up of dozens or hundreds of stacked blocks, as are computer vision systems, speech recognition systems and many other applications. The discipline that studies these architectures took its name precisely from this characteristic.
The term "deep" distinguishes these architectures from the shallow neural networks studied in previous decades, and it became common usage once the growth in available computing power and the availability of large amounts of data made it possible to effectively train networks with far more layers than before.
From our network
HSE Genius: AI for Safety Data Sheets
Extract SDS data, H phrases and ECHA compliance checks in seconds, powered by AI.
Visit hsegenius.com →From the Agora Intelligence blog
📱 Download the Android app (beta) iOS coming soon
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.