The project is the following six items:
Project 1: Actual combat of handwritten numeral recognition project
This project is based on TensorFlow, the most popular open source deep learning framework, to realize handwritten numeral recognition. Multilayer convolutional neural network is used to extract the features of handwritten digital images, and fully connected neural network is used to identify handwritten digital images. The whole project process includes data analysis and processing, model structure design, optimization and debugging, result analysis and so on. The final recognition accuracy rate is over 90%. This technology can be applied to text data recognition scenes, such as card text data recognition, bill text data recognition, automobile scene text recognition and so on.
Item 2: actual combat of vectorization of text features of literary works.
This project mainly studies the application of deep learning in natural language processing, and realizes word embedding learning and context reasoning in this field by using circular neural network and long-short memory network. The project will select some literary works and realize word embedding feature extraction and context inference based on long-term and short-term memory in turn. Related technologies can be used to process time-space series data, such as economic data prediction, stock data prediction, consumer behavior data prediction and so on.
Project 3: actual combat of face image generation project based on GAN
After learning, it can be directly applied to intelligent customer service dialogue generation, visual image synthesis, data enhancement and other tasks. This project will take face image generation as an example to introduce the principle and implementation of generative countermeasure network.
Project 4: The actual combat of face image generation project based on distributed GAN.
Through the parallel way, the data throughput of deep learning is improved and the learning and training process of the model is accelerated. Based on face image generation, this project introduces deep learning GPU and distributed cluster parallel mode. Related technologies can be directly applied to various scenarios of artificial intelligence+big data/cloud computing.
Project 5: actual combat of maze game project based on deep reinforcement learning
This project will briefly introduce the basic idea of reinforcement learning, and show the development and training process of deep reinforcement learning through the practice of game maze, so as to realize the independent exploration and learning and intelligent decision-making of AI system on the environment. Related technologies can be used for autonomous driving, AI quantitative investment, e-commerce product recommendation, robotics, human-computer interaction, optimal scheduling and other auxiliary decision-making tasks.
Project 6: Actual combat of enterprise license plate recognition project
This project will take license plate recognition as practical application, and guide students to complete the whole process of typical artificial intelligence projects, including project positioning, system architecture design, functional module realization, key algorithm application, testing and maintenance in demand analysis. The project will focus on the development and testing of core AI modules, and relevant practical links can familiarize students with the complete cycle of actual enterprise-level projects. The technical core of this project can be extended to identify other similar problems, such as container number identification, and can also be used as one of the core modules of the intelligent parking lot project.