When is the moon? Ask the sky for wine. I don't know what year is the palace in the sky?

Echoing the poems written thousands of years ago, it is the dream of a group of "star chasers": in the vast space, pulsars millions of light-years away continue to send signals to the earth, which is also China's "sky eye" - the 500-meter-aperture spherical radio telescope FAST (Five-hundred-meter Aperture Spherical radio Telescope, FAST for short) detection boundary.

Fast was completed in Guizhou in 2016. It is equivalent to the reflection area of 30 standard football fields, which greatly expands the "starchaser"'s field of vision to observe the starry sky, and also brings back more valuable cosmic "heartbeats" for radio astronomers on Earth.

And these jumping sounds like an electrocardiogram are also creating more possibilities for humans to explore and understand the universe.

Wang Shen is a member of "Star Chaser".

He graduated from the National Astronomical Observatory of the Chinese Academy of Sciences and obtained a doctorate in astrophysics. He is currently doing postdoctoral research in the laboratory of Mingmin Chi, associate professor and doctoral supervisor of the School of Computer Science and Technology, Fudan University. His main research interests are radio astronomical observations, pulsars and radio transients. source search, and interdisciplinary research related to computational astronomy.

In 2014, Wang Shen visited FAST for the first time. After going around the winding muddy mountain road to the scene, Wang Shen saw the FAST on the leveling site and building the support tower, but he was still attracted by this majestic project at a glance.

The FAST in front of Wang Shen is huge. But compared to this, the signals it has collected millions of light-years away are very "weak".

Wang Shen described: "All the cosmic signals we receive are not enough energy to turn a page, and the signal strength is very weak. But pulsars have the advantage that they are periodic, so we can use a period The way of folding is also a Fourier transform, which accumulates multiple periodic signals for detection and data processing."

FAST solves the problem of collecting signals from pulsars, the weak signals that accumulate to produce about 500 terabytes of raw data per day.

But new problems have arisen—one, the preprocessing of the signal through professional astronomical software for graphic transformation requires huge computational resources; To complete the feature comparison of pulsars from the 30-100 million images generated by preprocessing, it may take a year to successfully find pulsar candidates, which is a lot of work and time-consuming.

In the face of the huge amount of data, using AI to "find the stars" was taken into consideration.

The real problem, however, is that the sample size of pulsars is extremely small. Since the launch of FAST, more than 600 pulsars have been discovered in China, but only about 100 samples can be used to train AI models. This makes it difficult for the AI screening model of the National Astronomical Observatory to perform well.

The turning point occurred in 2021, when the research team of Tencent Youtu Lab where Wang Chengjie and Wang Yabiao were located was seeking the landing scene of AI+Science.

After learning about the pulsar project that Fudan University participated in, Wang Chengjie and Wang Yabiao quickly started a tripartite cooperation. In 2021, Tencent will join hands with the National Astronomical Observatory and the School of Computer Science and Technology of Fudan University to officially launch the "Star Exploration Program". Relying on the computer vision technology of Tencent Youtu Lab and Tencent's cloud computing capabilities, they hope to improve the efficiency of exploring pulsars through "cloud + AI".

Star Exploration Program

What blessings can AI algorithms and cloud computing power bring to "star chasers"?

Chi Mingmin, an associate professor at the School of Computer Science and Technology of Fudan University, is one of the founding members of the "Star Exploration Program". In her view, the addition of Tencent Youtu has brought several breakthroughs, including AI solutions based on multi-modal + semi-supervised learning, AI + dynamic spectral signal solutions, etc.

Finding stars among 100 million pictures is not an easy task. In order to solve the problem of insufficient amount of sample learning, the Star Exploration Project team designed a solution for domain transfer semi-supervised learning—using labeled data obtained by non-FAST observation equipment, and simultaneously using a small amount of FAST labeled data and a large amount of unlabeled data for deep neural network Network modeling. Through active learning to obtain "pseudo-labeled" data, the available samples for model training are much larger than those for manual labeling, thus greatly improving the ability to automatically identify pulsars.

At the same time, the "Star Exploration Program" team also tried multimodal applications. After the space signals collected by FAST are converted into images, information in different dimensions can be obtained, such as dispersion, phase-time, frequency-time maps, and so on. Chi Mingmin said that the deep fusion of multi-modal input signals is much better than the direct analysis of constant source signals. The method of multi-modal fusion is to correlate and analyze the pulsar information of data of different dimensions, so as to improve the reliability of the confirmation of the pulse periodic signal, and also improve the efficiency and accuracy of finding pulsars.

With the joint efforts of industry and academia, and with the same computing power, the "cloud + AI" approach helps to improve the data processing efficiency of pulsar search by 120 times. Compared with the original AI screening model in the industry, the newly designed multi-modal + semi-supervised learning AI solution not only has a higher recall rate, but also reduces the false alarm rate by 98%.

Why Pulsar?

In less than a year, the "Star Exploration Program" has found 22 pulsars from the FAST sky survey observation data.

Among the 22 pulsars, there are 7 high-speed rotating millisecond pulsars, and 6 suspected end-of-life pulsars with intermittent radiation phenomena. At the same time, the joint team of the "Star Exploration Program" also used the self-developed dynamic spectrum AI model to complete the detection of the radio pulse of a magnetar for the first time, and will start the AI+ radio astronomy signal analysis and processing of the M31 Andromeda galaxy— — This will also be the deepest and most complete detection of a pulsar-like compact object in the galaxy by the astronomical community.

The discovery of a pulsar, a rapidly rotating neutron star produced by a supernova explosion, can be seen as a bridge that will deepen human understanding of the vast universe.

Wang Shen explained that the core research of astronomy is called "two darks, one dark and three origins". The "two darks" are dark matter and dark energy. "One black" is a black hole. The "three origins" are the origin of the universe, the origin of celestial bodies and the origin of life, and pulsars are involved in three of them at once.

There are currently three known products in the late evolution of massive stars, one is a white dwarf, the other is a neutron star represented by a pulsar, and the third is a black hole. Direct observation of black holes is difficult, and studying pulsars can help to understand its "cousin" black holes.

Secondly, the study of pulsars can also help us solve the problems of the origin of celestial bodies and the origin of life. The material left over from the supernova explosion is a neutron star represented by a pulsar. "When we study pulsars, we are actually studying the life process in the late stage of massive stars, and then studying the evolution of celestial bodies and the formation of life." Elements heavier than iron required for the formation of life are produced by supernova explosions. In other words, all humans came from stardust.

The innovation of research methods brought about by the "Star Exploration Program" is also being extended to a wider range of fields.

For example, FRBs have been discovered by the Star Explorer team using a single-pulse search technique for pulsars. The origin of fast radio bursts is still unclear. Its energy intensity is equivalent to the energy of a whole year of the sun erupting in one millisecond, and it is an extremely violent activity. Researchers speculate that it is likely to come from outside the Milky Way, and look forward to discovering more details of the fast radio burst.

These model algorithms based on pulsar exploration have been continuously optimized and are expected to be opened to the public for wider astronomical exploration in the near future.

Dynamic Spectrum Signal Plot

In Wang Shen's view, because the pulsar itself has a huge magnetic field and rotates at a high speed, and even the surrounding environment is a very extreme celestial body, many laboratory conditions that cannot be reached on the earth can be observed around it. "For example, we use pulsars to measure the magnetic field distribution of the Milky Way. Although it is only a preliminary magnetic field distribution of the Milky Way, it does allow us to have a better understanding of the galaxies in which we live, which will also give us the opportunity to explore pulsars. Xinghe 'Star Exploration Program' has more exploration value."

Also, we live in time and space, and travel is displacement in time and space. Any civilization that develops planetary navigation needs to have a positioning guide, an object reference that indicates its position in the universe. Millisecond pulsars have extremely stable periodic motion properties that are more accurate than atomic clocks on Earth.

The discovery and research of pulsars is an important scientific goal that FAST needs to accomplish. For "star chasers" like Wang Shen: "From my personal point of view, when I saw FAST at the time, I felt that this life was worth working hard for."

Star Navigation

The essence of a pulsar is a neutron star.

Some of the more massive stars exploded at the end of their evolution, and then burst into intense space fireworks. In the process, massive stellar remnants were left behind, and eventually became fast-rotating neutron stars. Neutron stars can emit pulse signals, and they are just pointing in the direction of the earth, and they become the "pulsars" we encounter.

"Its beam is very collimated and just aimed at the direction of the earth, so it can cross the 'thousands of waters and thousands of mountains' and then be received on the earth." Wang Shen said.

The first pulsar discovered by humans was discovered in 1967 by Jocelyn Bell and her mentor Hewish in the United Kingdom. It sends out a regular signal, and at first they thought it might be sent to us by an alien civilization, so they called it "Little Green Man". But after follow-up scientific research, it was found that it is actually a neutron star that was predicted more than 30 years ago.

But what does a star tens of thousands of light-years away have to do with us?

In the words of Chi Mingmin - GPS is used for earth navigation, and pulsars are used for interstellar navigation.

Because the pulsar rotates at a high speed, it periodically releases a strong radio pulse signal, which is very regular. The signal of each pulsar is different, and it is equivalent to a beacon in the universe. As long as there are three pulsars, a position can be located. The more pulsars are discovered, the larger the range of space that can be located.

Stars guide humans, and interstellar navigation needs lighthouses.

Distribution map of pulsars in the whole sky

"The 22nd pulsar we recently found is 38,000 light-years away," Chi Mingmin said, "We are working hard to discover more and more special pulsar samples with the help of Tencent Youtu Cloud + AI, such as whether there will be There are pulsing stars or other stars, which require a wider search."

This year, the "Star Exploration Program" has been advancing to the depths of the universe - the Andromeda galaxy outside the Milky Way.

"Andromeda is 2.5 million light-years away from us. Theoretically, the probability of finding a short-period pulsar is too low. Only when the signal is particularly strong, similar fast radio bursts are more likely to be detected by FAST." Chi Mingmin said , Andromeda is outside the Milky Way and is very far from the earth. The transmission of the signal passes through a complex cosmic environment, such as the intergalactic medium, which will delay the arrival of high-frequency signals and seriously distort the signals. Data preprocessing can be corrected to some extent, but the computational overhead required is greater and the search model is more complex.

"But if there is a discovery in the Andromeda galaxy, it's a 0-to-1 breakthrough, which means a lot to us, and it's the most anticipated moment."

There is a sentence in the science fiction work "Three-Body Problem", which gives civilization to the years, not years to civilization.

From the perspective of astronomical observations, civilizations are sometimes full of chance. The Maori civilization has developed very powerful navigation technology, and even has been able to cross the sea from New Zealand to Hawaii very early, and have ships to reach the Americas, but they have never formed a very powerful civilization and have not ruled the entire southern hemisphere. An interesting point is that there was a lack of "South Pole stars" visible to the naked eye so they couldn't navigate precisely before telescopes were invented. Therefore, the development of civilization is locked in a relatively small area, and it is difficult to further evolve to a higher civilization.

And people are expecting that this navigation of pulsars can lead mankind to see the vast civilization of the universe.