1. Introduction to Jiuzhang’s intelligent decision-making products 2. Statistical models and symbolic reasoning 3. Links related to machine learning 4. Recommended reading 5. Tai Chi fitness

The second part is mainly about some related discussion articles on the two paths of connectionism and symbolism. How can we combine the strengths of the two to quickly realize industry intelligent decision-making solutions in the field and improve production efficiency. Here is a quote from Cai Shaowei on Weibo. Please pay attention to the work of Cai Shaowei and Xide in China, as well as the work of the Wolfram team abroad.

The third part introduces some materials related to machine learning, including a recent book published by a teacher from MIT, which has good content and some coding practices, which I recommend to everyone. Xiazi User Guide provides many links to tools to assist scientific research. There is also a graph tool that can query the relationships between papers.

The fourth part recommends some recently read and reprinted public account articles, including content about industrial software, as well as discussions on how to choose the direction of learning as artificial intelligence advances by leaps and bounds, and thoughts on how people and silicon-based silicon can collaborate.

The fifth part is about Tai Chi fitness, including Teacher Gu’s video and Li Xiangyu’s video account.

Jiuzhang product introduction

The following products include videos and some case introductions:

Introduction to Jiuzhang Intelligent Products and Cases

Can be collected directly into a module in TMS, (Integrated with Fuller TMS):

Instructions for calling the path planning module

The three-dimensional packing module API call can be linked to a small program or your own program module:

3D binning API call
Welcome to cooperate:

Recruitment of interns and collaborators

two

Statistical models and symbolism

Will new mathematical and physical concepts, formulas and conclusions be born? Can it help solve some mathematical problems, such as the Riemann Hypothesis?

The end of the universe is "calculation"! AI tycoon Wolfram’s latest speech: LLM independently explores the computing space, and the singularity is coming now

How to combine the research results of symbolism and machine learning to establish modeling methods and model libraries for different industries requires the joint efforts of academia and industry:

The combination of deterministic solving system and statistics-based learning system: symbolism + connectionism, mutual promotion and common progress, helping to solve problems in the industry

The following text comes from Cai Shaowei's Weibo. They have made useful explorations in formal systems and hope to get support from the market:

While ChatGPT is on fire, I can't help but think of another methodology counterpart, a software with a similar mission, Wolfram|Alpha based on symbolism. Here are some thoughts on reading the linked article:

https://writings.stephenwolfram.com/2023/01/wolframalpha-as-the-way-to-bring-computational-knowledge-superpowers-to-chatgpt/

While ChatGPT is a remarkable achievement in imitating human thinking, not all useful tasks are amenable to human-like solutions. Some tasks require a more formal and structured approach. One of the most important achievements of our civilization over the past few centuries has been the establishment of paradigms such as mathematics, exact science, and computation, and the creation of a tower of abilities distinct from those of purely human thought.

Wolfram|Alpha attempts to represent as many things in the world as possible through a formal symbolic method and perform knowledge calculations on this basis. Using the Wolfram Language and Wolfram|Alpha, you can do many things that are beyond the capabilities of the human mind. This approach is also important for artificial intelligence. It allows AI to gain knowledge computing capabilities and make up for the shortcomings of current AI.

ChatGPT does a great job with the "human-like part", there is no precise "right answer". But it doesn't perform well when faced with certain tasks that require precision. To solve this problem, ChatGPT and Wolfram can be combined. You can consider inputting the natural language generated by ChatGPT to Wolfram|Alpha, and Wolfram|Alpha converts the input into Wolfram language code for calculation; another way is for ChatGPT to directly learn the Wolfram language and let GPT run directly using Wolfram.

ChatGPT and Wolfram|Alpha are two very different systems, but they share a common interface: natural language. This means that ChatGPT can "communicate" with Wolfram|Alpha like a human, and Wolfram|Alpha can convert the natural language obtained from ChatGPT into a precise symbolic computing language and apply its computational knowledge capabilities.

For a long time, there has been a binary way of thinking between "statistical method" and "symbolic method" in the field of artificial intelligence. But now, thanks to the success of ChatGPT and the work of Wolfram|Alpha, there is finally an opportunity to combine them and create something more powerful than either approach alone.

==================================

In our long-term scientific research on constraint solving, we have studied different solving methods and developed solvers. However, when we connected with the industry, we found that the biggest bottleneck was actually in mathematical modeling: how to express the enterprise's problems as mathematical problems. This requires both business understanding and mathematical knowledge, and often relies on expert experience. Different "sects" have different approaches, which feels like a sword debate on Huashan Mountain. There is a very lack of talent in this area, and there is no corresponding training plan. It is still a "craft". And this shortcoming also seriously hinders the development of mathematical power.

I think many people, like me, have once imagined the automation of mathematics. As for the solution part, I have seen that AI is gradually becoming more involved, and I am still confident. The arrival of GPT has also shown us that there is hope for automation in mathematical modeling. In this way, the two tiles of mathematical power are complete.

There is a natural idea: in the future, mathematical modeling can be automated to a large extent with the combination of large models and mathematical languages. For example, large models can be used to parse natural language (including guiding modification requirements), generate mathematical models under the specifications of mathematical languages/templates, and rapidly test by introducing solvers to form feedback and iteratively modify the mathematical models.

Modeling can be turned into a programmable process that can be debugged. Once it can be mechanized, even if the initial model is inaccurate, there will be a repair process, and then iteration will be faster.

This seems to be the general trend. I think this will first affect the current status of intelligent decision-making. For many companies, the current approach to their demand for intelligent decision-making is to purchase technical services. Companies that engage in intelligent decision-making technology generally rely on experts to conduct mathematical modeling, and then design specialized algorithms or use solvers to solve the problem.

If automatic mathematical modeling comes and the role of experts changes from building mathematical models to reviewing models, productivity will be greatly improved and costs will be greatly reduced. This could make mathematics more accessible to more industries. Maybe this can detonate the power of mathematics. The mathematical power that everyone has been waiting for is finally coming.

Xide Software: Automatic Modeling

https://leandojo.org/

How to achieve teamwork and human-machine cooperation in the era of ML:

three

some study materials

Machine learning helps solve optimization problems and machine learning solves combinatorial optimization problems: https://event.baai.ac.cn/live/725

Book: Understanding Deep Learning:

https://udlbook.github.io/udlbook/

by Simon J.D. Prince
Published by MIT Press Dec 5th 2023.

30 free machine learning books in English:

https://newsletter.theaiedge.io/p/30-free-machine-learning-e-books

Keyanxiazi usage guide (scientific research): https://keyanxiazi.bepass.cn/

Website to explore academic papers in a visual graph: https://www.connectedpapers.com/

Generative AI for Everyone (Andrew Ng):

https://www.deeplearning.ai/courses/generative-ai-for-everyone/

Four