What are neural networks?
Posted: Thu Jan 30, 2025 6:51 am
Neural Network Neural networks are groups of algorithms that are designed like a human brain to recognize recurring patterns and then sort or label them. The recognized patterns are translated into mathematical vectors. All information from the real world, such as images, sound, text or time sequences, is taken into account.
Neural networks help to classify new information based on similarities across multiple levels for the respective system and to group it into model groups. Labels help to name these groups. Examples of labels can be: Spam, Not Spam, Satisfied Customer, Unsatisfied Customer, Purchased Link, Not Purchased Link .
The following slide from a presentation by Google's Jeff Dean illustrates cambodia phone number data that certain patterns, for example in an image of a lion, are recurring. Based on these recurring patterns, machine learning can be used to automatically interpret and label the image of a lion.
Neural networks consist of several levels or layers that, when connected in series, contribute to the refinement and accuracy of the assumptions. If you want to delve deeper into the topic, you can find a good introduction here .
types of machine learning
There are basically three different types of machine learning:
Supervised Learning
Unsupervised learning
reinforcement learning
Here is an excerpt from the German Wikipedia :
Supervised learning The algorithm learns a function from given pairs of inputs and outputs. During the learning process, a "teacher" provides the correct function value for an input. The aim of supervised learning is to train the network to make associations after several calculations with different inputs and outputs. One area of supervised learning is automatic classification. An example of an application would be handwriting recognition.
Unsupervised learning The algorithm creates a model for a given set of inputs that describes the inputs and enables predictions to be made. There are clustering methods that divide the data into several categories that differ from one another by characteristic patterns. The network thus independently creates classifiers according to which it divides the input patterns. An important algorithm in this context is the EM algorithm, which iteratively sets the parameters of a model so that it optimally explains the data seen. It assumes the existence of unobservable categories and alternately estimates the belonging of the data to one of the categories and the parameters that make up the categories. One application of the EM algorithm can be found, for example, in Hidden Markov Models (HMMs). Other methods of unsupervised learning, e.g. principal component analysis, do not involve categorization. They aim to translate the observed data into a simpler representation that reproduces it as accurately as possible despite drastically reduced information.
Reinforcement learning : The algorithm learns a tactic through reward and punishment on how to act in potentially occurring situations in order to maximize the utility of the agent (i.e. the system to which the learning component belongs). This is the most common form of learning for humans.
I also found some good graphics in this presentation by Rahul Jain . (Link to the presentation at the end of the article).
supervised learning
Machine Learning Process: Supervised Learning
Supervised learning requires a lot of preparatory work, because example models must be defined and labeled in advance in order to identify incoming information and assign it to this model group or classify it. For quality assurance reasons, this labeling is usually carried out by humans. Based on certain recurring patterns, the system can then independently recognize information with the same or similar pattern properties in the future and assign it to the respective model group.
unsupervised learning
Machine Learning Process: Unsupervised Learning
In unsupervised learning, pre-labeling does not take place and the model groups are formed automatically based on patterns.
reinforcement learning
Machine Learning Process: Reinforcement Learning
Similar to the term Artificial Intelligence, the term Machine Learning is often equated with Deep Learning and Semantics or mentioned in the same breath. Below is an attempt to differentiate.
Neural networks help to classify new information based on similarities across multiple levels for the respective system and to group it into model groups. Labels help to name these groups. Examples of labels can be: Spam, Not Spam, Satisfied Customer, Unsatisfied Customer, Purchased Link, Not Purchased Link .
The following slide from a presentation by Google's Jeff Dean illustrates cambodia phone number data that certain patterns, for example in an image of a lion, are recurring. Based on these recurring patterns, machine learning can be used to automatically interpret and label the image of a lion.
Neural networks consist of several levels or layers that, when connected in series, contribute to the refinement and accuracy of the assumptions. If you want to delve deeper into the topic, you can find a good introduction here .
types of machine learning
There are basically three different types of machine learning:
Supervised Learning
Unsupervised learning
reinforcement learning
Here is an excerpt from the German Wikipedia :
Supervised learning The algorithm learns a function from given pairs of inputs and outputs. During the learning process, a "teacher" provides the correct function value for an input. The aim of supervised learning is to train the network to make associations after several calculations with different inputs and outputs. One area of supervised learning is automatic classification. An example of an application would be handwriting recognition.
Unsupervised learning The algorithm creates a model for a given set of inputs that describes the inputs and enables predictions to be made. There are clustering methods that divide the data into several categories that differ from one another by characteristic patterns. The network thus independently creates classifiers according to which it divides the input patterns. An important algorithm in this context is the EM algorithm, which iteratively sets the parameters of a model so that it optimally explains the data seen. It assumes the existence of unobservable categories and alternately estimates the belonging of the data to one of the categories and the parameters that make up the categories. One application of the EM algorithm can be found, for example, in Hidden Markov Models (HMMs). Other methods of unsupervised learning, e.g. principal component analysis, do not involve categorization. They aim to translate the observed data into a simpler representation that reproduces it as accurately as possible despite drastically reduced information.
Reinforcement learning : The algorithm learns a tactic through reward and punishment on how to act in potentially occurring situations in order to maximize the utility of the agent (i.e. the system to which the learning component belongs). This is the most common form of learning for humans.
I also found some good graphics in this presentation by Rahul Jain . (Link to the presentation at the end of the article).
supervised learning
Machine Learning Process: Supervised Learning
Supervised learning requires a lot of preparatory work, because example models must be defined and labeled in advance in order to identify incoming information and assign it to this model group or classify it. For quality assurance reasons, this labeling is usually carried out by humans. Based on certain recurring patterns, the system can then independently recognize information with the same or similar pattern properties in the future and assign it to the respective model group.
unsupervised learning
Machine Learning Process: Unsupervised Learning
In unsupervised learning, pre-labeling does not take place and the model groups are formed automatically based on patterns.
reinforcement learning
Machine Learning Process: Reinforcement Learning
Similar to the term Artificial Intelligence, the term Machine Learning is often equated with Deep Learning and Semantics or mentioned in the same breath. Below is an attempt to differentiate.