What is AI and ML

Big data

Buzzwords that do not mean anything are widespread in the IT world - a further complication is that hardly two experts understand exactly the same term as we often find out in our web TV section "Ready when you are - the buzzword interrogation" were allowed ...

This is a little different in the field of artificial intelligence, cognitive computing and thinking robots - here the meaning is mostly clear, but not the appropriate term itself. Sometimes two terms mean the same thing - or do you know a difference between machine intelligence and artificial intelligence ? What about machine learning and deep learning? We clarify.

Generic term KI / AI

"Artificial Intelligence" (AI) or the English expression "Artificial Intelligence" (AI) refers to "a wide range of methods, algorithms and technologies to make software so smart that it appears to outsiders like human intelligence", describes it by Lynne Parker, head of the Information and Intelligent Systems department at the American National Science Foundation. In other words, machine learning (ML), machine vision (computer vision), natural language processing, robotics and all other related topics are part of AI.

Machine intelligence = AI

"Some people will distinguish between machine and artificial intelligence, but there is no consensus that the two terms have different meanings," said Parker. The use of the two terms is regionally different - "machine intelligence" goes back more to classic engineering work, which can be found mainly in Europe, explains Thomas Dietterich, professor at Oregon State University and chairman of the Society for the Promotion of Artificial Intelligence AAAI. "Artificial intelligence", on the other hand, has a kind of "science fiction coating" and is more common in the USA. In Canada, for example, the term "Computational Intelligence" is also common.

Machine learning as a common word

As part of AI, the term machine learning (ML) describes a wide range of algorithms and methods to improve the performance of software with growing amounts of data. This is about neural networks as well as deep learning - both terms will play a role later.

"Basically, machine learning is about reading developments from amounts of data or recognizing categories in which the data can be classified. As soon as the software comes into contact with new data, it can make appropriate decisions," explains Parker.

  1. Apache Spark MLlib
    Formerly known as part of the Hadoop universe, Apache Spark is now a well-known machine learning framework. Its extensive range of algorithms is constantly being revised and expanded.
  2. Apache Singa
    Singa, which recently became part of the Apache Incubator, is an open source framework that is intended to "train" deep learning mechanisms for large volumes of data. Singa provides a simple programming model for deep learning networks and supports various development routines.
  3. Caffe
    Caffe comprises a whole set of freely available reference models for common classification routines; the grown caffe community contributes further models. Caffe supports the Nvidia programming technology CUDA, with which program parts can optionally also be processed by the graphics processor (GPU).
  4. Microsoft Azure ML Studio
    Because the cloud is the ideal environment for ML applications, Microsoft has equipped its Azure cloud with its own ML service based on “pay as you go”: With Azure ML Studio, users can develop and train and train AI models then convert them into APIs in order to make them available to others.
  5. Amazon machine learning
    Amazon Machine Learning works with data that is in an Amazon cloud such as S3, Redshift or RDS and can build new AI models with the help of binary classifications and multi-class categorization of given data.
  6. Microsoft DMTK
    Microsoft's DMTK (Distributed Machine Learning Toolkit) is designed to scale ML applications across multiple machines. It is intended more as an "out of the box" solution and less as a framework - the number of supported algorithms is correspondingly small.
  7. Google TensorFlow
    TensorFlow is based on so-called data flow graphs, in which bundles of data (“tensors”) are processed by a series of algorithms that are described by a graph. The movement patterns of the data within the system are called “flows”. The graphs can be assembled using C ++ and Python and processed via CPU or GPU.
  8. Microsoft CNTK
    The Microsoft Computational Network Toolkit works similarly to Google TensorFlow: Neural networks can be generated using directed graphs. According to Microsoft's own description, CNTK can also be compared with projects such as Caffe, Theano and Torch - but it is faster and, in contrast to those mentioned, can even access processor and graphics processor performance in parallel.
  9. Samsung Veles
    The Samsung framework is intended to analyze and automatically normalize data sets before they go into productive operation - which in turn is immediately possible thanks to its own API called REST - provided the hardware used has enough power. The use of Python in Veles also includes its own analysis and visualization tool called Jupyter (formerly IPython) for the representation of individual application clusters.
  10. Brainstorm
    Brainstorm relies on Python to provide two data management APIs (called “handers”) - one for CPU processing using the “Numpy” library and one for GPU processing using CUDA. A user-friendly GUI is in the works.
  11. mlpack 2
    The new version of the mlpack machine learning library, which was written in C ++ and first appeared in 2011, has a lot of innovations - including new algorithms and revised old ones.
  12. Marvin
    Marvin's source code is very clear - the pre-trained models included (see picture) allow extensive further development.
  13. neon
    Neon by NervanaSystems is an open source framework based on modules that can be switched on and off and enables AI processes via CPU, GPU or Nervana's own hardware.

Face recognition is used as an example. "I don't know exactly how it works that I recognize my wife's face," says Dietterich. "This is what makes it so difficult to program a computer to do just that." Machine learning therefore works with examples. "It's more about input-output than coding," says the AAAI boss.

According to Parker, common ML varieties are artificial neural networks, support vector machines, decision trees, Bayesian networks, closest-neighbor classifications, self-organizing maps, case-based reasoning, instance-based learning, the hidden Markov model and various types of regression analysis. If you want to know more about the individual terms, you will find detailed information on the stored Wikipedia links.

Neural Networks vs. Deep Learning

Artificial neural networks are a special type of ML that is based on how the human brain works - even if, according to Parker, the real comparability is extremely poor. There are different types of neural networks - but essentially all of them are based on a system of nodes that are connected to one another by cables of different weights. The nodes are also called "neurons" and are arranged in several layers - including an input layer, through which data enter the system, and an output layer, through which the responses are made. In addition, there is one or more hidden layers on which the actual learning takes place. Typically, neural networks learn by changing the weight of the cross connections between the nodes, says Parker.

The term "deep learning" now refers to a "deep neural network", namely one that comprises a very large number of neurons in various hidden layers. A "flat" neural network, on the other hand, usually only consists of one or two hidden layers.

Parker explains, "The idea behind deep learning is not new, but has recently become popular because we have very large amounts of data today and also have fast processors that enable successful solutions to difficult problems."

  1. Facebook Big Sur
    The AI ​​system, which is under an open source license, relies on the Nvidia Tesla Accelerated Computing Platform and today performs complex tasks on Facebook that previously required third-party hardware.
  2. Google RankBrains
    For search queries that appear for the first time, RankBrains is supposed to translate human written language into mathematical vectors, which the search engine can then process. This form of machine learning is getting better and better as the number of previously unknown search queries increases. Inquisitive internet users train the system almost unconsciously.
  3. Google Deepmind AlphaGo
    Recently defeated the world and European champions in the Asian board game Go: the Alpha Go AI system, designed by Google Deepmind.