Deep learning and neural networks

Deep learning refers to an approach in machine learning, which aims at teaching machines to recognise abstract concepts based on large datasets. The leading edge currently is unsupervised machine learning, where the machine is left to make sense of the data on its own. Deep learning has made huge leaps in pattern recognition possible. Google Deepmind is one of the prominent companies utilising deep learning.

Why is this important?

For platforms deep learning offers the possibility to make sense of and recognise patterns from large amounts of data. Google provides an open source library called TensorFlow for this. Another benefit are the services that deep learning provides, such as voice recognition, chatbots etc. These can provide new functionality to the platform. On a broader view, the motivation is to use the deep learning to solve global problems.

Things to keep an eye on

The focus is now especially on unsupervised machine learning and “differentiable neural computers”, which can make sense of complex structured data. Examples of what deep learning algorithms such as the Google DeepMind can do range from lip reading to advanced translation to making sense of a metro map. One interesting development is making APIs to enable artificial intelligence algorithms to play games such as Starcraft and learn through it. This also means that artificial intelligence might be the future user of a platform. The big question then is will it benefit or exploit the platform.

Selected articles and websites

DeepMind has conquered games, London’s Underground and now it wants to help save the planet
Deep Learning Papers
Google’s DeepMind AI Said to Outperform Professional Lip-Readers
Zero-Shot Translation with Google’s Multilingual Neural Machine Translation System
Google’s AI creates its own inhuman encryption
Google DeepMind to Use Blizzard’s StartCraft II for AI Research Platform
Differentiable neural computers

Mikko Dufva

Research Scientist VTT Technical Research Centre of Finland Ltd
Share this
TwitterFacebookEmailLinkedIn

Artificial Intelligence

Artificial intelligence has made tremendous progress over the last decades, moving from beating humans at games towards a wide range of fields. In 1999, IBM’s big blue defeated the world best Jeopardy champions, and 17 years later Google bested the world’s champion “go” player. Recently, Big Blue was employed to develop an antiviral to attack the Zika virus with great success. The new “macromolecule” was effective against not only Zika but multiple other viruses as well. LTP news estimates that “AI is going to change every endeavor of human activity from Medicine to Government to Manufacturing, Law, Finance and beyond.  By 2025, AI Will Have a 5-Trillion-Dollar Direct Impact on the Workforce“.

Why is this important?

  •  AI replaces jobs, and platforms enable this: platforms act as the mediator between the AI and the user, and offer the interface and central place to ask for e.g. law advice.
  •  AI provides new services. This is especially the case for healthcare, where improved pattern recognition can detect tumors, or evolutionary algorithms can design new medicine. Add to this the possibility to turn code into biological products via synthetic biology, and the range of services expands.
  •  AI provides the boost that data analytics needed to make sense of the data collected and created via platforms. Platforms enable the recording of every transaction, which results in vast amounts of data, which can be useful also outside the interests of marketing.

Things to keep an eye on

The range of application areas of AI will expand, covering new industries, such as financial analytics, advising, insurance and law. Bank of America estimates the market for AI-based analytics to grow to 70 billion dollars by 2020. At the same time, artificial intelligence is becoming more and more ubiquitous; people use AI-based services everyday without realizing it. As a counter trend to the hype and excitement over the potential of AI, concern about its malevolent potential has also been raised, ranging from AI causing mass unemployment (most probable) to unleashing “a global propaganda war that sets governments and populations in opposition”.

The Finnish government recently released its action plan for enhancing the innovation ecosystem around intelligent robotics and automation in Finland, with a focus on security, privacy, user centricity and service design. Currently, the main innovation hub around AI and robotics in Finland is Airo island. A good example of a Finnish company applying AI is Zenrobotics, which makes recycling robots based on machine vision. The Finnish “Curious AI Company” is focusing on the development of advanced artificial intelligence and is now trying to apply its unsupervised machine learning to various pilot areas. Finns also take part in the discussion around AI, see for example the predictions on AI by Jarno M Koponen.

Selected articles and websites

A Government resolution to promote the development of intelligent robotics and automation
The next AI is no AI
AI and Communication: Machines That Can ‘Hear’ and ‘Understand’ Voices
Siri creators unveiled a new AI platform that seems to blow Siri out of the water
New Infosys AI tool could transform the way companies maintain complex systems
The AI Revolution: The Road to Superintelligence
New AI security system cleverly combines machine learning and human intuition
DimensionalMechanics raises $4.7M for enterprise artificial intelligence platform

 

Mikko Dufva

Research Scientist VTT Technical Research Centre of Finland Ltd
Share this
TwitterFacebookEmailLinkedIn