Datawhale202211—Li Hongyi "Machine Learning" (Deep Learning Direction)
The basic situation of ML + related technologies + learning objectives
(here is a combination of Li Mu's "Part of the content of hands-on deep learning")
Article Directory
foreword
With the continuous development of artificial intelligence, the technology of machine learning is becoming more and more important. This article introduces the basic content of machine learning. This note combines the two courses of "Machine Learning" and "Hands-on Deep Learning" (the former is mainly) to better understand ML and DL.
1. Basic introduction to Machine Learning
The essence of machine learning: determine a function from data.
brief development
Since AI technology has swept the world, many of our technologies have only been advanced at the theoretical level. There is still a long way to go to truly become a product in the hearts of the public and become a technology that everyone can use or benefit from. Even though AI is ridiculed or doubted, we practitioners should look at our work more rationally and see its value and future.
Machine learning actually has only three steps, which simplifies the whole process. It can be compared to: put the elephant in the refrigerator. When we put the elephant into the refrigerator, there are actually three steps: open the door; stuff the elephant in; finally close the door, and then it's over. Therefore, the three steps of machine learning are like saying that putting an elephant in the refrigerator only needs three steps.
2. Related technologies
supervised learning
Definition: According to the existing data set, know the relationship between the input and output results . According to this known relationship, an optimal model is obtained through training. That is to say, in supervised learning, the training data has both features and labels . Through training, the machine can find the connection between features and labels by itself . When faced with data with only features and no labels, Labels can be identified .
Application example: regression, classification
PM2.5 prediction house price prediction
unsupervised learning
Definition: We don't know the relationship between the data and features in the dataset, but we need to get the relationship between the data according to clustering or a certain model. It can be said that compared with supervised learning, unsupervised learning is more like self-study. There is no label for machines to learn to do things by themselves.
Application example: news classification
transfer learning
Transfer learning is to transfer the knowledge of one domain (source domain) to another domain (target domain), so that the target domain can achieve better learning effect. Usually, the amount of data in the source domain is sufficient, while the amount of data in the target domain is small. This scenario is very suitable for transfer learning.
Application Example: Sentiment Classification
Structured Learning in Supervised Learning
Structured Learning is not a specific technology, but a method of describing and analyzing problems.
Application Example: Summary Generation
reinforcement learning
Reinforcement Learning (RL), also known as Reinforcement Learning, Evaluation Learning or Enhanced Learning, is one of the paradigms and methodologies of machine learning, used to describe and solve the problem of learning through the process of interaction between an agent (agent) and the environment. Strategies to achieve the problem of maximizing returns or achieving specific goals.
Application examples: Go, StarCraft
3. Why study machine learning
Many people say that AI may replace humans. Of course, this is an important topic in many disciplines, and we will not discuss it here. However, with the development of AI technology, many new occupation types and even new industries will be born. Below we introduce several types of AI-related occupations.
model trainer
There is an interesting metaphor in the course, comparing the model trainer to the Pokemon trainer, you need to pay attention to the selection and optimization of the model, and choose the loss function and model for the application scenario. Therefore, a useful model is inseparable from a trainer with experience and innovative thinking. In fact, there is no clear differentiation of this position at present, because all practitioners must consolidate their basic skills through training models in order to advance better. Of course, we look forward to the birth of similar positions in the future to save students in the winter of employment.
domain expert
This career will put more emphasis on industry experience and insights. Maybe an industry expert who has been working in a certain industry for many years and has a knowledge background of ML will be very suitable for this position. Discover the specific application scenarios of AI in a certain industry through experience and insight, and the ML/DL method can solve related problems very well. It is a powerful force that can help the AI model to be well implemented. The future will be AL frontier theory and AI society bridge of value.
data scientist
This profession is more inclined to experiment and scientific research, and practitioners with rich experimental experience, knowledge base and practical ability may be able to exert more powerful energy. They often need to obtain data from domain experts and build models based on the data to solve corresponding problems. If they have to compare with traditional IT positions, this profession will be more similar to engineers, and domain experts are more similar to product managers.
AI expert
This career is more academic, and scientists who may have been engaged in scientific research for a long time will be more handy. It mainly focuses on the accuracy improvement and optimization of the model to better promote the theoretical development.
Of course, such a division is imprecise, and current practitioners often wear multiple jobs. On the other hand, if you want to better engage in AI-related jobs, you need to have the ability to do more than one job. But we will see some promising scenery in the future, which is also the driving force for our continuous learning and progress.
4. Reference documents
Feeding from Datawhale
Li Hongyi's "Machine Learning" open source content 1:
https://linklearner.com/datawhale-homepage/#/learn/detail/93
Li Hongyi's "Machine Learning" open source content 2:
https://github.com/datawhalechina/leeml -notes
Li Hongyi's "Machine Learning" open source content 3:
https://gitee.com/datawhalechina/leeml-notes
Feeding from the official
Li Hongyi's "Machine Learning" official address
http://speech.ee.ntu.edu.tw/~tlkagk/courses.html
Li Mu's "Hands-On Deep Learning" official address
https://zh-v2.d2l.ai/
Feeds from netizens
Zhizhi animation understands related concepts
https://www.bilibili.com/video/BV1DW411v7TF/?spm_id_from=333.337.search-card.all.click
Summarize
The content of this section is comprehensible, and it will play a leading role in the follow-up study, and have a certain understanding of some basic concepts. Teacher Li Hongyi's courses are very interesting and clear. Just from the content of this section, we can understand some of the teacher's attitudes and opinions on AI. We hope to gain more surprises by moving forward.