Chapter 1906 Make research romantic


Chapter 1906: Let Research Be Romantic

It took nearly an hour to finish all this. Zhou Zhi was too lazy to go back to the bedroom. According to his old habit, he poured down on the big couch next to the seat to finish the work.

In the next two days, Zhou Zhi and Mai Xiaomiao began to become familiar with the oracle bone conjugation project and the graph database architecture of the digital library.

These two projects are also the focus of the next experimental research on cell-based neural networks.

The oracle bone conjugation project originated from the automatic recognition technology of converting picture characters into vector characters in the Hanwen large font library, as well as the root deconstruction and splicing technology. The success of these two sub-projects has also saved money for the research and development of the Hanwen large font library. It took a lot of time, especially for the second and third phases of the project, to complete the recognition, input, vectorization, warehousing and other work in a very short time after getting the cards of "Sea of ​​Words" from two senior brothers.

Also due to the rapid enrichment of the second-phase font library, it directly blocked the possibility of independent research and development of full-code font libraries in Chinese character-using areas such as Hong Kong Island, New Island, Bay Island, and even island countries, attracting Japan and South Korea to send working groups to "jointly support each other." "A grand event", Hanwen Character Library finally became the only and unified Chinese character library in the world.

Of course these technologies will not be shelved after completing the development of the large font library. Instead, they will continue to be developed in depth, and Zhou Zhou soon found an application scenario for them, which is oracle bone fusion.

So the two previous technologies developed into the edge recognition matching algorithm and the changing radical matching algorithm.

These two algorithms are simple to say, they just realize some functions of the human eye and human brain.

The former algorithm allows the computer to identify the edge features of the fragments in the picture regardless of whether they are large or small, their location or angle, and match them with the edge features of other fragments in the database. If a matching position is found, an attempt will be made to piece together the oracle bones.

After the oracle bones are pieced together, the text on them often changes from a few random words of unknown meaning to a paragraph, and a lot of historical information hidden in the fragments will be presented to researchers for the first time.

The value is of course self-evident.

The second algorithm is more discrete and nonlinear.

Although oracle bone writing is very mature, it has been popular for a long time, and there is no fixed structure when forming characters. Therefore, one character often has dozens of ways of writing.

For example, "bu" is actually the root of a left foot and a right foot. There are many ways to write these two roots, some are traditional and some are simple, and these two roots are used by Zhen people. After being carved into the oracle bone to form the word "step", its positional relationship and angle relationship can be described as varied.

For a trained human being, as long as these two characters are engraved with similar accuracy, the concept will naturally pop up in their mind: This is the left foot and this is the right foot.

Then as long as the two root words are placed together, roughly staggered up and down, another concept will naturally pop up in the mind: Well, this is most likely the word "step". However, such thinking activities are too "advanced" for current computer technology, because current computer technology is mainly still processing linear logic operations, and for this type of one input with multiple outputs, or multiple The complex nonlinear relationship in which two outputs jointly lead to an input is difficult to handle.

The discipline of wheat seedling research is used to solve this problem.

On a larger scale, this is the direction of artificial intelligence. On a smaller scale, this is a rudimentary attempt to let computers solve nonlinear problems.

Of course, to solve practical problems, mathematical theories and tools alone are useless. To integrate with a complex project like Oracle Bone Fusion, it is absolutely impossible to understand how the project currently operates.

After having an in-depth understanding of the oracle bone conjugation and digital library projects, Mai Xiaomiao raised a question: "I have a question for you, did you have the idea of ​​developing a neural network system from the beginning? Because judging from the current stage of these two projects, they are completely cellular architectures. The system is completely in line with the bottom-level system structure of the neural network system. Next, we only need to study how to construct the interrelationship algorithm between the various elements."

"Xiaomiao, you really think highly of me. "Zhou Zhi smiled and said: "The reason why these two projects are so special is that The whole thing is that they are all literature and history research projects, which are completely different from the science and engineering projects that you are good at.”

“I have been instilling a point of view in Tianyu Lidong Chunjia and others, and that is the logic of science and engineering. The system is often like a winding mountain road with a clear path The road reaches the top of the mountain, which is the distinctive feature of linear logic. "

"The logical system of cultural and historical projects is actually a spherical network structure. In such a structure, winding mountains are also used. Can a road-like system be used to solve the problem successfully? ”

"It can indeed be successful on some small and local problems. Again, for example, keyword search, search for the word Su Shi, and find all the passages with Su Shi in all the classics in history. At most, the word Su Shi Under the article It is already difficult to get a biography and historical evaluation of Su Shi by searching various labels such as Dongpo, Huangzhou Tuanlian, Chibi Fu before and after, When will the moon come out, etc. ”

“ If you can do this Although this is already remarkable, there is often another type of problem that researchers in literature and history want to explore. Let me give you a topic as an example: "On the Interrelationship between Su Shi's Personality Charm and Song People's Aesthetics, and Its Impact on Later Aesthetics" "Influence of". For such questions, if you want to find answers in the current digital library, there is no ready-made answer. It requires a lot of study and data collection to complete the correlation between the elements in the scholar's mind, and finally After complex thinking Only then can we get a reliable conclusion."

"Theoretically, the information system can realize the answer to this question by itself." Mai Xiaomiao nodded: "As long as the two elements of cell sample and relationship algorithm are rich enough, from then on. If you spread out a point across the entire net, you can get a lot of event probability combinations, and finally form a... In your words, a reliable conclusion."

"Don't ask me," Zhou Zhi said. : "It's my turn to ask you Yes, is this the goal of your laboratory?”

Mai Xiaomiao thought for a while and said: “I actually didn’t think about it well before, because my research was mostly theoretical, and I understood it before returning to China. Looking at the current development situation in China, I thought that the best way to deploy it was to use supercomputing to conduct some physics or astronomy research and try to perform non-linear processing of data. In other words, the computing device is single and the data is distributed. .”

"It's your liberal arts imagination that makes my research more romantic."

After speaking, he made a bitter face again: "But the research and development expenses, maybe..."


Previous Details Next