The latest generation neural networks are partly replacing scientists: they conduct experiments, diagnose diseases, reveal patterns, put forward and test hypotheses. They are used where data volumes exceed any human capacity. What scientific issues helped to solve the artificial intelligence - in the material RIA Novosti.
"Adam" and "Eve"
The first robot scientist was created in 2009 by British specialists under the leadership of Professor Ross King, then an employee of the University of Aberystwyth. His “brain” was a neural network program using four PCs and controlling laboratory equipment. The virtual creature was named "Adam".
A neural network is a computer program that analyzes large amounts of data at great speed, looking for common features and patterns in them. Unlike modeling, neural networks do not need scientific hypotheses - they build them themselves and test them themselves. Scientists use this property to find out how likely a scenario is. This significantly saves time and computing power, which is required much more in, for example, computer simulation. Scientists provided Adam with strains of baker's yeast with various genes disabled. The robot itself grew cultures of these mutant strains and monitored how they develop without certain enzymes, for which the switched off genes are responsible. The artificial brain learned from the first experiments and subsequently planned new ones more efficiently. The robot could conduct a thousand experiments per day. As a result, he put forward two dozen hypotheses about genes encoding 13 enzymes. The scientists then performed manual experiments and confirmed Adam's guesses for 12 genes. Nearly a decade later, King and his colleagues developed another robot scientist, Eve. She sorts out various compounds and looks for which ones are promising as drugs. The machine is capable of examining ten thousand substances per day. The first discovery of "Eva" was a chemical compound with anti-cancer properties, which was also effective against the causative agent of malaria. For screening "Eva" uses smart systems based on genetically modified yeast. Nearly a decade later, King and his colleagues developed another robot scientist, Eve. She sorts out various compounds and looks for which ones are promising as drugs. The machine is capable of examining ten thousand substances per day. The first discovery of "Eva" was a chemical compound with anti-cancer properties, which was also effective against the causative agent of malaria. For screening "Eva" uses smart systems based on genetically modified yeast. Nearly a decade later, King and his colleagues developed another robot scientist, Eve. She sorts out various compounds and looks for which ones are promising as drugs. The machine is capable of examining ten thousand substances per day. The first discovery of "Eva" was a chemical compound with anti-cancer properties, which was also effective against the causative agent of malaria. For screening "Eva" uses smart systems based on genetically modified yeast. For screening "Eva" uses smart systems based on genetically modified yeast. For screening "Eva" uses smart systems based on genetically modified yeast.
Longevity and smoking markers
Last year, scientists from several countries, including Russia, represented by the staff of ITMO University (St. Petersburg), published a paper on how to determine a person's age using a biochemical blood test. To do this, they trained the neural network and then gave it samples of more than 120,000 blood tests from patients from Canada, South Korea and Eastern Europe for research. The program knew only nationality, gender and two dozen biochemical parameters of blood. This was enough to establish the age of each patient with good accuracy. In January of this year, the same team of scientists presented new results: the artificial intelligence they trained was able to calculate, based on the biochemical parameters of blood, whether a person smokes or not. Scientists have made available to the neural network a database of almost 150 thousand blood tests of patients in the province of Alberta (Canada), which were previously made anonymous. The program only knew the sex of the people. The neural network successfully coped with the task and learned to isolate smokers. Moreover, she found signs that indicated the true, that is, the biological age of the person, and not chronological (according to the passport). It turned out that female smokers were biologically aging twice as fast as non-smokers, and men - one and a half times. It turned out that female smokers were biologically aging twice as fast as non-smokers, and men - one and a half times. It turned out that female smokers were biologically aging twice as fast as non-smokers, and men - one and a half times.
Anti-cancer neural network
Promotional video:
Scientists from Stanford (USA) used the ability of neural networks to analyze images, which are essentially a set of digital data. They trained the photo program to distinguish between carcinoma and melanoma, malignant growths that indicate cancer.
The program examined nearly 130 thousand images of various formations on the skin, which were marked by the type of disease or as ordinary moles, keratomas, and deduced patterns. The results were checked by two dozen dermatologists: they turned out to be quite accurate. Now, in order to conduct an initial diagnosis, it is enough to send a photo of a skin neoplasm to the doctor from a smartphone. And then - depending on the answer - decide whether to do a biopsy in order to accurately establish a diagnosis. Artificial intelligence is also used at the OncoTarget Center for Personalized Oncology at Sechenov University (Moscow). There they create a digital model of the patient - this is complete information about his disease, the genetic characteristics of the tumor. Scientists hope that the neural network, analyzing data arrays, will optimize treatment for each patient.
In search of the mysteries of the universe
Artificial intelligence opens up great prospects for astronomers, who literally choke on the abundance of data obtained as a result of observations. Numerous space missions, orbiting and ground-based telescopes have generated far more of them than humans will be able to process soon. Kevin Schawinski of the Institute for Particle Physics and Astrophysics at the Swiss Higher Technical School in Zurich believes neural networks will revolutionize astronomy. He and his colleagues tested artificial intelligence in analyzing data on the rate of formation of binary stars to understand why it decreases in galaxies when the external conditions change. Astronomers trained the neural network using an array of galaxy images. Similarly to how the program can depict what the face of a person will be in old age,it can also change the appearance of a galaxy as it enters a group or cluster. The results of the neural network work coincided with the observations. In 2017, a self-learning neural network created by Google helped NASA discover a new exoplanet. Analysis of data from the Kepler orbiting telescope revealed a rocky planet just thirty percent larger than Earth, orbiting the star Kepler-90 in the constellation Draco. However, the planet was too close to the star for life. Earlier, the neural network has already found the sixth planet in the Kepler-80 star system. All this is the result of processing weak light signals that only a computer program can catch.helped NASA discover a new exoplanet. Analysis of data from the Kepler orbiting telescope revealed a rocky planet just thirty percent larger than Earth, orbiting the star Kepler-90 in the constellation Draco. However, the planet was too close to the star for life. Earlier, the neural network has already found the sixth planet in the Kepler-80 star system. All this is the result of processing weak light signals that only a computer program can catch.helped NASA discover a new exoplanet. Analysis of data from the Kepler orbiting telescope revealed a rocky planet just thirty percent larger than Earth, orbiting the star Kepler-90 in the constellation Draco. However, the planet was too close to the star for life. Earlier, the neural network has already found the sixth planet in the Kepler-80 star system. All this is the result of processing weak light signals that only a computer program can catch. Earlier, the neural network has already found the sixth planet in the Kepler-80 star system. All this is the result of processing weak light signals that only a computer program can catch. Earlier, the neural network has already found the sixth planet in the Kepler-80 star system. All this is the result of processing weak light signals that only a computer program can catch.
Tatiana Pichugina