How DeepMind AI is Changing the Way We Look at Machine Learning

Tags – DeepMind AI


DeepMind is a London-based artificial intelligence company that was founded in 2010.

DeepMind is best known for creating the AlphaGo program, which was able to defeat a professional Go player in 2016.

As you can imagine, DeepMind’s success has brought them a lot of attention, and they are now considered one of the leaders in the field of machine learning.

In this blog post, we discuss DeepMind’s contributions to the world of machine learning and how their work is changing the way we look at AI, with 3 distinct examples.


DeepMind AI (1)


1. Algorithms Mimicking AI

To get the best of both worlds, DeepMind wants to have artificial neural networks mimic algorithms, and it’s testing Google Maps as a testbed.

To provide some context here, traditional algorithms have allowed software to be utilised in a wide range of applications, however the data they operate on does not always represent reality.

Taking its application to the next level, deep learning is what enables some of today’s most famous AI applications, but retraining deep learning models for areas they were not intended for is necessary. DeepMind is attempting to combine deep learning with algorithms: a deep learning model that can learn how to replicate any algorithm, resulting in an algorithm-equivalent model that can work with actual-world data.

The key point is that machine learning algorithms have distinct qualities from deep learning approaches.

The hypothesis that deep learning approaches may be improved to mimic algorithms implies that generalisation, such as that seen with algorithms, might be achievable with deep learning.


2. Controlling Nuclear Fusion Reaction

Researchers use strong magnetic coils to trap the nuclear fusion reaction and guide it into the intended path.

When the fusion reaction is taking place, the gas in the vessel must be controlled to prevent the plasma from contacting the sides of the container. This might cause damage to the walls and impede the fusion reaction.

If you are wondering, there’s little danger of a nuclear explosion since the fusion reaction cannot function without confinement.

However, each time scientists wish to alter the plasma’s configuration in order to test out new forms that may produce more power or a cleaner plasma, it necessitates a significant amount of engineering and design work.

So far, conventional methods are computer-based and rely on models and elaborate simulations, yet they may be complex.

Alternatively, AI from DeepMind has been able to operate the plasma independently.

The neural network, an example of AI technology that emulates the structure of the human brain, was initially trained in a computer simulation.

The researchers noticed how changing the settings on each of the 19 coils affected the form of the plasma contained within the tank. It was then given various forms to try to duplicate in the plasma.

The magnetic coils in the right way were used by ‘DeepMind’ to create these forms autonomously.


3. Deciphering Ancient Inscriptions

A computer scientist and classical studies specialists at DeepMind and Ca’ Foscari University of Venice developed a transformer-based neural network to decipher Greek inscriptions written between the 7th century BC and the 5th century AD.

In addition, the model may also tell when the text was created and where it came from.

To get the needed information, researchers may use broken pots, faded scripts, and other such discoveries to help them examine ancient civilizations.

To begin the process, the text must first be transcribed by scanning an image of a historical item or script.

Then, Ithaca analyses the text. It works by predicting missing or blurry characters to generate output words.

The program generates and ranks a list of its greatest predictions; epigraphists can then scroll through them and determine whether the model’s conclusions seem correct or wrong.

When humans and machines collaborate, the greatest outcomes can be achieved. When experts worked alone, they were 25% successful in reconstructing ancient artefacts, but when they collaborated with Ithaca, their success rate increased to 72%.

Ithaca’s performance on its own is around 62%. For comparison, it’s 71% accurate in determining the place where the text was written, and it can date works to within 30 years of their creation between 800 BC and 800 AD.

DeepMind researchers are now modifying their model to account for various sorts of ancient writing systems, including Akkadian from Mesopotamia, Demotic from Egypt, Mayan from Central America, and ancient Hebrew.


To learn more, get in touch with us today.


You may also like:

  1. How to Create Viral Marketing Campaigns That Engage and Convert
  2. Marketing Needs Purpose; Not Just Campaigns: Unilever Marketing
  3. The Role of Artificial Intelligence (AI) In The Future of Marketing
  4. LinkedIn Marketing – A Detailed Future-Proof Look
  5. 9 Personality Traits of a Good Salesperson