Definition and Classification of ML
ML refers to the discipline that specializes in how computers simulate or implement human learning behaviors to acquire new knowledge or skills, allowing computers to reorganize the existing knowledge structure and continuously improve their performance. ML is based on data, searching for rules by studying sample data, and predicting future data based on the rules obtained. ML is the core of artificial intelligence (AI), a wide range of AI fields such as data mining, computer vision, NLP, and biometric recognition.
① The training data of supervised learning has classification labels. The higher the accuracy of the classification label, the higher the accuracy of the learning model. Supervised learning establishes a function model based on the given training data to realize the annotation mapping of the new data. Supervised learning algorithms include regression and classification, and application areas include NLP, information retrieval, text mining, handwriting recognition, spam detection, etc.
② Unsupervised learning uses unlabeled limited data to describe the structure or law hidden in the data. Its typical algorithm is clustering. Unsupervised learning does not need to use manually labeled data as training samples, and avoids classification errors caused by positive sample offset and negative sample offset. Application areas of unsupervised learning include economic forecasting, anomaly detection, data mining, image processing, pattern recognition, etc.
③ Reinforcement learning refers to a learning mode that maximizes the return of the subject in the process of interaction with the environment. The purpose of reinforcement learning is to achieve the best evaluation of the agent through the external environment. Reinforcement learning is widely used in robot control, unmanned driving, industrial control and other fields.
① The number of hidden layers of the shallow learning algorithm network is small, the algorithm framework is simple, and there is no need to extract multi-level abstract features. Typical shallow learning includes support vector machines, logistic regression and so on.
② Deep learning is a self-learning method based on multi-layer neural networks and large amounts of data as input rules. It relies on a large amount of actual behavior data provided to it, that is, the training data set, to adjust the parameters and rules in the rules. The deep learning algorithm network has many hidden layers and complex algorithms. Compared with shallow learning, deep learning pays more attention to the importance of feature learning. Typical deep learning algorithms include convolutional neural networks, recurrent neural networks and so on.