Programming and technology are created for a specific purpose and meaning, as well as deep learning. The main reason this technology is present in the digital era is none other than to facilitate daily activities with the help of machines and artificial intelligence. More specifically, why deep learning was created to optimize the performance of unstructured data on a website or application. In addition, deep learning for vision systems will directly or indirectly have an impact on operational costs and technology development. Then, deep learning can make manipulation techniques and feature engineering more effective.
Deep learning is a set of algorithms in machine learning that seeks to learn at different levels, according to different levels of abstraction. This usually uses an artificial neural network. The levels in this studied statistical model correspond to different concept levels, where higher level concepts are determined from lower levels, and lower level concepts can help to define many higher level concepts.
Deep learning is part of machine learning which is the most popular research reference today. Deep learning uses an Artificial Neural Network (ANN) which is an information processing engine modeled on the structure and actions of biological neural networks in the brain. ANN is flexible and can adapt itself to solve complex problems that are not clearly described by mathematical models, such as pattern recognition and classification, function, and control approaches. The principle of deep learning is Artificial Neural Network (ANN) which has many hidden layers.
Based on the architecture and techniques used, deep learning is broadly divided into 3 (three) categories:
1. Deep Networks for Unsupervised or Generative Learning
This category captures high-level correlations of observed or visible data for pattern analysis or synthesis purposes when no information about the target class label is available.
2. Deep Networks for Supervised Learning
This category directly exerts discriminatory power for pattern classification purposes, usually by characterizing the class-conditioned posterior distribution of the visible data. Target label data is always available in a direct or indirect form for such supervised learning. This class is also referred to as discriminative deep networks.
3. Hybrid Deep Network
This category aims at assisted discrimination, often significantly, with generative or unsupervised deep networks. Another goal is achieved when discriminatory criteria for supervised learning are used to estimate parameters in either deep generative or unsupervised deep networks in the above categories.