Do you know deep learning? Then do you know the function and how it works?
But you are certainly familiar with the term machine learning, right? yup, deep learning is a subfield of machine learning whose algorithms and programs are inspire by the structure and workings of the human brain. These structures are call Artificial Neural Networks or abbreviated as ANN. Generally, ANN has three or more layers of neural networks. These structures can learn and improve their capabilities and can adapt from the data obtained, so they can create solutions to various problems that are difficult to solve by other machine learning algorithms.
Types of Deep Learning Algorithms
The following is the algorithm of deep learning’s interconnect neural network:
- Convolutional Neural Network (CNN)
CNN consists of many layers to process and extract features from data. It is usually use to process images and detect objects. Currently, CNN is widely use to identify satellite imagery, medical imagery, and detect anomalies.
- Recurrent Neural Network (RNN)
Recurrent Neural Networks (RNN) is a form of Artificial Neural Networks (ANN) architecture which is specially design to process sequential data. RNN is usually use to solve historical data or time series problems, for example, weather forecast data. In addition, RNN can also be implement in the field of natural language understanding, such as language translation.
- Long Short Term Memory Network (LTSM)
LSTM is a type of Recurrent Neural Network that can study historical data or time series. It is a complex deep learning algorithm and can learn long-term information very well. LSTM is very powerful to solve complex problems such as speech recognition, speech-to-text application, music composition, and development in the pharmaceutical sector.
- Self Organizing Maps (SOM)
The final type of deep learning neural network is self-organizing maps or SOM. This algorithm is able to make data visualization independently. SOM was create to assist users in understanding high-dimensional data and information.
Application of Deep Learning
Deep learning is the AI technology that most often drives the performance of the software. We find many applications of deep learning, from computer applications to smartphones.
- Analyze photos (Image recognition)
Image recognition systems are useful for identifying and detecting features and patterns such as letters, numbers, objects, places, and people’s faces. Like the Face Authentication Tagging function on Facebook which works by recognizing a person’s face from a photo so that you can add a personal name and an explanation of the photo.
This system is often also apply to support vehicle steering as a security system by detecting dangerous objects ranging from image analysis, character recognition systems, and so on.
- Voice Recognition
This system functions to analyze and identify human voices which can then be converte into text. Usually, this system is apply to devices that use voice commands such as Google Assistant, Alexa, Siri, and Microsoft Translator.
- Natural language processing
This technology works by processing the language used every day. Processing is achieve by dividing the text to be process into words, clarifying the sentence structure, and thinking about the meaning of the whole text.
This system must be process using all information in an integrate manner. Source Engine in a browser, Google translate, Siri Apple, MS. Office Word, to Extensions and Applications such as Grammarly use a natural language processing system developed to solve problems such as ambiguous words, syntactic meaning, and so on.
Benefits
It was first launched in 2010 and became the momentum that led to a breakthrough in the development of AI (artificial intelligence). In recent years this technology from AI has continued to develop, we can even feel some of its benefits, such as:
- Can process unstructured data such as text and images.
- Can automate the feature extraction process without the need to do the labeling process manually.
- Provide a quality end result.
- Can reduce operating costs.
- Can perform data manipulation more effectively