1. Artificial Intelligence¶
This module takes the deep learning of artificial intelligence as the theme, and is based on Python, allowing readers to develop artificial intelligence on the Lubancat RK series board. Focus on the deployment of various deep learning frameworks on Lubancat. There are brief descriptions about the training and optimization of the framework model. For more detailed and systematic learning, please refer to various deep learning books and RKNN documents.
1.1. Introduction¶
1、 Artificial Intelligence （AI）: Different scholars and institutions may have different definitions and understandings. The common definition is: artificial intelligence is a new technical science for simulating, extending and expanding human intelligence theories, methods, technologies and application systems. As for the specific implementation, there are many methods and branches of artificial intelligence. The machine learning and deep learning we often hear are currently more effective implementation methods.
2、 Machine Learning （ML）: It is an important branch of artificial intelligence. It mainly studies how to let computers automatically improve performance through data or experience, by imitating human learning ability, learning from sample data to obtain experience (model), and then making predictions. Common algorithms for machine learning include supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.
3、 Deep Learning （DL）: It is the most popular branch of machine learning algorithms, which has made remarkable progress in recent years and replaced most traditional machine learning algorithms. Simply put, it is a machine learning algorithm based on artificial neural network. Deep learning has a wide range of applications in machine vision, speech recognition, natural language processing and other fields, and is especially suitable for solving perception problems.
The relationship between artificial intelligence, machine learning, and deep learning can be briefly referred to as follows:
4、 Deep Learning Framework: It is a software library that can help you build, train, and deploy deep learning models. It usually provides advanced abstraction, predefined layers and optimizers, and automatic differentiation, making writing deep learning code easier and more efficient. There are many deep learning frameworks to choose from, such as:
TensorFlow: One of the most popular deep learning frameworks open sourced by Google, based on the calculation method of data flow graph, it supports multiple platforms and languages.
PyTorch: One of the latest deep learning frameworks open sourced by Facebook, based on the Torch library, written in Python language, supports dynamic graphs and automatic differentiation.
PaddlePaddle: Baidu’s open source industry-level deep learning platform, which integrates deep learning core training and reasoning frameworks, basic model libraries, end-to-end development kits, and rich tool components.
MindSpore: Huawei’s open-source and self-developed AI framework supports deep learning, training, and reasoning in all scenarios on the device, edge, and cloud, and has features such as general automatic differentiation and distributed parallel training.
Some frameworks are listed above. In fact, there are many other deep learning frameworks, such as Caffe, keras and so on.
1.2. Artificial intelligence and Lubancat RK series board¶
Lubancat RK series boards use Rockchip rk356X processor, rk3568 and rk3566 NPU modules. The processing unit specially used for neural network accelerates the neural network algorithm in the field of artificial intelligence, up to 1TOPS, and supports integer 8 and integer 16 convolution operations, and supports the following deep learning frameworks: TensorFlow, TF-lite, Pytorch, Caffe, ONNX etc.