What is a "black box" in Deep Learning?

AIGlossary

Deep Learning

Related services

If Machine Learning (ML) is when you teach a program to distinguish spam from an ordinary letter, then Deep Learning is when a program learns to drive a Tesla car by itself. It is "deep" because it uses neural networks with hundreds of layers, which are able to analyze gigantic volumes of unstructured data (video, audio, text).

What is the difference between ML and Deep Learning?

In classic ML, a person must specify features to the program. For example, to teach the program to recognize a cat, you say: "look for whiskers and pointed ears." In Deep Learning, you don't explain anything. You simply "feed" the program a million photos of cats and dogs, and it itself finds signs by which they can be distinguished.

Why has Deep Learning exploded now?

The mathematics of deep networks itself was known back in the 80s. But two things are needed for their work: 1) Huge volumes of data (Big Data). 2) Huge computing power (GPU video cards). Only in the 2010s did these two factors converge.

Where is it used?

01Generative AI

All modern LLMs (ChatGPT, Claude) and image generators (Midjourney) are built on Deep Learning (Transformers or Diffusion architecture).

02Autopilots

Cars use real-time convolutional networks (CNNs) to analyze video from cameras and make decisions on the road.

/ FAQ

This is the main problem of deep learning. The network can diagnose cancer with 99% accuracy from the image, but it can't explain to the doctor WHY it made that decision. Its internal logic is too complex for a human to decipher.

Deep Learning

/ Deep Learning

Ready to implement?

We'll help you implement it. 30-minute process audit - free.

Book an audit