Why Artificial Intelligence Is Emerging Further –– Trends
Artificial intelligence (AI) systems will be understood better and better by informatics scientists. AI analysts at consultancy firm PwC compiled AI key trends to be probably in the spotlight this year. What the experts expect in short: Man and machine are moving closer together.
The ten trends center on automated machine learning as “Democratization of Artificial Intelligence”, Capsule Networks –– and the ancient theory of probability (calculus) by English mathematician Thomas Bayes (born 1702, died 1762). A better understanding of AI is not just about researchers’ claim for themselves, but also about business benefits through more compliance. –– In detail:
– Source / Infographic: PwC
Higher Compliance by Advanced Insight into Neural Networks
So-called Deep Neural Networks (DNS) mimic the learning behavior of the human brain. You “learn” from audio, image and text files. So far, users understand comparatively little of how exactly this is going on. PwC observed that researchers are gaining more and more insight due to new theories of deep learning. The benefits, the analysts outline for companies, are more transparency in corporate networks and architecture, which in turn should lead to more compliance.
Better Data Classification by Capsule Networks
So-called capsule networks form a new type of DNA that processes visual information similar to the human brain. The special feature: Capsule Networks “understand” better than previous Convolutional Neural Networks, in which hierarchy they should order the information. PwC sees potential here for the more precise classification of data. Specifically, up to 50 percent fewer errors are possible.
Higher Availability of Proper Data by Lean, Augmented Data Learning
The biggest problem around Deep Learning and Machine Learning PwC sees in the availability of appropriate data to train the system. There are two possibilities: synthesizing data synthetically – lean data technology – and moving proven learning models from one task to the next – augmented data technology.
Digital Twins Stabilize the (Industrial) Internet of Things
A physical object gets its virtual twin to make maintenance and operation trouble-free and more efficient. The origin actually can be found in large plants. It will expand to non-physical objects and processes. Next stage will be to expand and stabilize the (Industrial) Internet of Things (IoT).
Automated Machine Learning Automates Model Development
Because Automated Machine Learning (AutoML) targets on automation of workflows for the development of intelligent models, AutoML contributes to the democratization of artificial intelligence. The analysts expect AutoML tools in the future to be part of larger machine learning platforms.
Gathering Insights: From Artificial to Explainable Intelligence
Many experts are talking about a black box in connection with AI, because it becomes not always clear exactly how results are accomplished. On the other hand artificial intelligence should be explainable, transparent and provable, PwC underlines – identifying this contradiction as a new movement: Explainable AI.
New Languages for Probabilistic Programming
The PwC analysts expect a growing use of programming languages to describe probabilistic models. It will make it easier to develop models. Specifically, these languages assist in dealing with insecure and missing information.
Hybrid Learning Models Improve the Handling of Probabilities
In the future, companies will combine different types of neural networks to make safer predictions. The focus is on so-called Bayesian Conditional GANs. Thomas Bayes (see intro) was a trailblazer for the probability calculus.
More Decisions Than In Chess: Deep Reinforcement Learning
When first chess computers defeated humans, it was considered as a sensation. Today PwC expects further progress in Deep Reinforcement Learning (DLR). Its potential is located in learning progress through observation, interaction with the environment, and reward. In the future, this type of AI will be used by decision-makers in a wide variety of applications.
Generative Adversarial Networks vs Cyber Fraud
In a Generative Adversarial Networks (GAN), two networks compete against each other – one as the generator, the other as the discriminator. The so-called generator creates fake data that looks like real. The other one, the discriminator, keeps it apart from each other.
Explore the latest developments regarding data, systems and networks, machine learning, and AI for the IIoT –– at automatica 2018, the leading exhibition for smart automation and robotics with IT2Industry as special subject area with IT solutions and services for digitalization and Industry 4.0, from 19 to 22 June, 2018, at Messe München.
Edited by Ingo Becker
AI,Artificial Intelligence,automated machine learning,AutoML,capsule networks,data,deep learning,deep reinforcement learning,Digital Twins,DLR,GAN,hybrid learning models,neural networks,probabilistic programming,PwC,technology,Thomas Bayes,trends