Leveraging Open Source Hardware for AI Experiment Education: A Case Study of Xugu Number

Click on the above “Mushroom Cloud Creation” to follow us!

Leveraging Open Source Hardware for AI Experiment Education

— A Case Study of Xugu Number

Xie Zuoru, Wenzhou Middle School, Zhejiang Province

Artificial intelligence is an emerging technology that extends and enhances human capabilities to transform nature and govern society through intelligent machines. In 2017, the State Council issued the “New Generation Artificial Intelligence Development Plan,” proposing to “implement a national smart education project and set up AI-related courses in primary and secondary schools.” The subsequent “New Curriculum Standards for High School Information Technology (2017 Edition)” (hereinafter referred to as “2017 Edition Standards”) not only included AI requirements in Compulsory Module 1 “Data and Computing” but also set up an independent elective compulsory module named “Introduction to Artificial Intelligence.”

● AI Education and Open Source Hardware

The 2017 Edition Standards set high requirements for AI learning, requiring students to build intelligent systems with “application scenarios oriented towards real life”. Since building intelligent systems requires more than just understanding basic AI algorithms, students also need to be familiar with common AI development tools and frameworks. The newly written high school textbooks all use Python as the foundational language, making Python-supporting AI frameworks the first choice. Python has the characteristics of being open-source and cross-platform, and open-source hardware that can run the Linux system, such as Raspberry Pi and Xugu Number, naturally becomes an important carrier for learning AI. Therefore, the 2017 Edition Standards frequently mention open-source hardware.

● Completing Classic AI Experiments with Open Source Hardware

AI course teaching in universities generally combines a series of experiments to allow students to experience the process and cultivate their abilities. Conducting experiments requires an environment, and setting up a Python environment capable of completing a series of AI experiments requires installing many extension libraries. However, most computer labs in primary and secondary schools use Windows, making it a challenging task to configure the programming environment, often leading to compatibility issues between modules, leaving teachers who are new to Python at a loss.

Open-source hardware has convenient firmware installation, which can be “flashed” using specialized tools. Based on the analysis of high school textbooks, open-source hardware can support the following three types of AI experiments.

① Basic AI Algorithm Experiments.

The vast majority of textbooks will include expert systems, K-nearest neighbors clustering, K-means classification, decision trees, regression analysis, Bayesian analysis, and neural networks as basic AI algorithms. The main Python libraries supporting these experiments include scikit-learn, Keras, and Tensorflow.

② AI Experiments Supported by Specific Environments.

Computer vision, OCR, speech recognition, and natural language processing are major application areas of AI, requiring specific libraries for support. Computer vision typically uses OpenCV, while face recognition may also rely on Dlib and face_recognition. Offline versions of speech recognition libraries often do not perform well; it is recommended to use the SDK from Baidu AI Open Platform or Tencent AI’s WebAPI. Natural language processing also requires libraries that support Chinese word segmentation, such as jieba.

It is important to emphasize that the speech wake-up experiment in speech recognition has not found particularly suitable libraries under Windows, while under Linux, Snowboy or Baidu AI’s SDK can be used.

③ Complex Interactive Experiments Combined with Sensor Control.

With the rapid development of AI technology, human-computer interaction is no longer just simple command input and output; it increasingly reflects the characteristics of “natural communication.” AI is gradually becoming more human-like and visual, making it real, touchable, and interactive.

On ordinary computers, it is almost impossible to achieve “sensing and control” because it requires various peripherals or smart terminals, where open-source hardware is a typical representative of smart terminals, its chip pin levels can be programmed for control, allowing it to output high and low levels, read level states, and also feature ADC and PWM functions, making it the best choice for achieving complex AI interactions.

● Conducting AI Experiments on Xugu Number

Relatively speaking, AI experiments that can be completed on ordinary computers can also be completed on Xugu Number. Xugu Number comes pre-installed with Python and common AI frameworks, and we successfully completed a series of neural network experiments on Xugu Number, as shown in the table below.

Generally speaking, a complete machine learning experiment is divided into several basic steps: data collection, data cleaning, model building, model training, model evaluation, and model application. On Xugu Number, not only can experiments be completed, experiencing all steps, but models can also be directly applied to build an intelligent information system. For example, in the handwritten digit recognition and gesture recognition experiments, once trained on Xugu Number, it can directly use a USB camera to input handwritten digits or gestures, and Xugu Number expresses the recognition results through voice (speaker) or actions (servo). This expands the experimental results, allowing students to not only complete an experiment but also develop a typical maker project.

● Advantages of Conducting AI Experiments with Xugu Number

1. Learn upon startup, convenient environment to carry. Xugu Number runs a complete Linux system, and students can access it via remote desktop or web. As long as there is a network, it can be connected and used upon startup, and the board has built-in a series of ipynb format learning courses for self-study convenience. Students can take Xugu Number home for research, equivalent to a portable programming server, reflecting the characteristics of OYBD (Bring Your Own Device).

2. Training is application, models are directly deployed. Common machine learning experiments end after completing the “model evaluation” phase. For primary and secondary school students, such learning is insufficient; they need to apply what they have learned and build intelligent information systems to gain a more authentic experience. Xugu Number can not only deploy models trained by oneself but also run models trained on other platforms, demonstrating good compatibility.

● Conclusion

In addition to Xugu Number, there are several other open-source hardware options for conducting AI experiments, such as Raspberry Pi, Jetson Nano, and LattePanda. These hardware options are reasonably priced and can effectively complete these classic experiments. To encourage more schools to use open-source hardware, the 2017 Edition Standards specifically proposed in the “Teaching Tips” section to “fully utilize the rich resources of open-source hardware and AI frameworks,” to “build application scenarios oriented towards real life,” and to organize teaching through group cooperation and project learning, encouraging students to explore actively and practice boldly.

In fact, not only the “Introduction to AI” module requires open-source hardware, but the compulsory module “Information Systems and Society” and elective compulsory modules “Open Source Hardware Project Design” and “Network Basics” also require open-source hardware. We hope more teachers will understand open-source hardware and use it to teach AI, allowing our students to better touch, understand, and master AI.

Neural network experiments completed on Xugu Number

Leveraging Open Source Hardware for AI Experiment Education: A Case Study of Xugu Number

Note: The Python version on Xugu Number is 3.5, and the core AI frameworks are Keras and Tensorflow.

Source: China Information Technology Education Magazine

Author: Xie Zuoru

——————— End of Text ——————

The purpose of education is to cultivate students’ collaboration skills, communication skills, critical thinking, and creativity, with creativity being the core. Mushroom Cloud Maker Education, as one of the few one-stop maker education service providers in China, aims to cultivate children’s creativity.

To better connect the knowledge learned in class with the real world, Mushroom Cloud will guide schools to plan, establish, and operate their own maker spaces. Based on different age groups of students, distinctions have been made in design and layout.

Elementary school maker spaces focus on fun,

emphasizing interactive scenarios;

Middle school maker spaces focus on practicality,

emphasizing learning scenarios;

High school maker spaces focus on technology,

emphasizing application scenarios;

In terms of content, Mushroom Cloud collaborates with leading maker teachers in China to compile a series of textbooks suitable for domestic maker education. It also has a complete maker education curriculum system, including course content and teaching aids, course training, and technical Q&A. Similarly, distinctions will be made in course classification and design based on different age groups of students.

Elementary school focuses on gamification and experiential learning, primarily based on “learning through play.”

Middle school guides students to engage in inquiry-based learning through hands-on experiences, promoting “learning by doing.”

High school learning is based on problem and design, requiring teachers to create relevant real-life situations for students to learn through “thinking.”

Leave a Comment Cancel reply