DeepMind Technologies is a UK-based AI company. It was founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman, and was acquired by Google in 2014 for £400 million so it is now part of the Alphabet group. The company is well-known for developing AlphaGo, which became the first computer program to beat a professional Go player on a full-sized board in 2016.

The majority of DeepMind’s patents relate to the architectural details of neural networks, and to aspects of training a neural network. Their inventions therefore represent innovations that have a potentially far-reaching impact in a variety of fields. DeepMind have also pursued patents relating to some of these fields, examples of which are described below:

• WO 2018/048934 relates to the synthesis of audio data using a neural network, which could be used in text-to-speech systems.

• WO 2018/081089 relates to a technique of performing language modelling tasks on text sequences using neural networks, and has applications towards machine translation, text summarisation, and speech recognition. In each of these application fields, an understanding of linguistic context enables more accurate outputs to be generated.

• US 2014/0185959 relates to an image processing algorithm which identifies the texture or patterning of objects in an image (such as striped or spotted items of clothing), which has applications towards image classification.

• EP 3398114 relates to a method of image compression, where salient features are recognised using a neural network and then used to summarise the contents of the image.

DeepMind have typically applied for patents via the WIPO (PCT) route, indicating that they seek to commercialise their inventions in the global marketplace.