Contents

I planned it with reference to the following site. https://blog.kikagaku.co.jp/2020/04/06/how-to-learn-ai/

• Decide what you want to be with AI
  First of all, I think it is necessary to fully understand what AI is and what is different from conventional simulations. If you understand this well, you will understand that you want to do what you can do naturally. A sense of crisis is more motivating than what you want to do ――

  Why did you decide to study AI? It's my personal feeling, so you can skip it. AI will appear frequently as we develop technology at the incumbent manufacturer. I can't write in detail, but the movement to use AI for business is getting closer and closer than it was a few years ago when I joined the company. It includes not only company information but also what other companies are doing.
  I think there is no doubt that AI will come as a means to solve problems efficiently in the future. Therefore, I think that people who can master AI will be able to carry out effective proposal activities.
  Therefore, I would like to study AI from now on and be able to use AI in the future.

• Understand the big picture for learning
  As for the basics, it seems that you should learn the following. Mathematics required for machine learning Python Knowledge of machine learning

  I also attended a programming school to learn about IT, but I am afraid that this kind of language will become a so-called mass-produced engineer because it is a field with low barriers to entry and a large competitive population when anyone can do it. I did. Therefore, it is good that the basics include highly specialized content.

• Minimal math It seems that the following contents should be suppressed. In my case, the content in bold seems to require new learning. Differentiation (up to partial differentiation) Linear algebra Basic statistics (mean, variance, standard deviation, etc.) Simple regression analysis mathematics Mathematics of multiple regression analysis
• Python There seems to be no problem with VS Code for the learning environment for the time being. An open source library called chainer is introduced, and it seems to be good to learn while actually programming.
• Machine learning It seems good to suppress the curriculum below. Since it is a technical book level, I wonder if it is necessary to purchase Typical algorithms for supervised learning Multiple regression analysis, ridge regression, lasso regression, logistic regression, k-nearest neighbor method, support vector machine, decision tree, random forest, typical algorithm of unsupervised learning, k-means clustering, principal component analysis, representative of hyperparameters Adjustment method, grid search, random search, Bayesian optimization, typical evaluation index of classification, correct answer rate, precision rate, recall rate, F value

  You also need to study deep learning ... Neural network calculation (forward propagation), linear transformation, non-linear transformation, neural network calculation (back propagation) Error back propagation method, gradient descent method, mini-batch learning, neural network implementation (TensorFlow or PyTorch) </ b>

  Machine learning is too dark and scary About the data to be handled below Image data, convolutional neural network (CNN), object detection algorithm (R-CNN, YOLO, SSD, etc.), semantic segmentation algorithm, sentence data, sentence data feature extraction method (Bag of words, Word2Vec, etc.) , Machine translation algorithms (Seq2Seq, Attention, etc.)
  Time-series data (1/1 number of visitors is data that has a context in the data of 100 people) Recurrent neural networks (RNN, LSTM, GRU, etc.) Convolutional Neural Network (CNN)
  Table data (data as described in an Excel sheet) Feature engineering Evolving machine learning algorithms (XGBoost, LightGBM, etc.) </ B>