Artificial intelligence can increase productivity, but everyone knows that scientists have begun to use AI technology to reproduce the brain's advanced cognitive functions! This time, they created a complex neural network to simulate the spatial navigation capabilities of the brain. This achievement shows the world that AI algorithms can help neuroscientists to verify their various theories about how the brain works. Of course, this powerful computer-based verification tool will not let traditional neuroscientists lose their jobs. On May 9th, US local time, the UK's Deep Mind team and University College London (UCL) published a paper on Nature, which shocked the academic world. The paper, titled Vector-based navigaTIon using grid-like representaTIons in arTIficial agents, in this study, the research team used a deep learning approach to train computers to simulate the location of rats in a virtual environment. Paper link: http:t.cn/R32YrKS One of the mammalian brains used for navigation is called grid cells, which are activated when mammals record their position in space. Cellular activity presents a six-sided appearance after denoising. Shape arrangement mode. To the surprise of scientists, the AI ​​simulation system jointly developed by University College London and DeepMind automatically generates a hexagonal pattern that is very similar to brain cell activity and guides virtual rats to take shortcuts. The “grid cells†obtained by the researchers using AI are highly similar in pattern to the “grid cells†in the mammalian foraging state. What's even more amazing is that computer-simulated rats can navigate well in virtual maze through grid-like cell coding, and even find shortcuts to get out of the maze! “This paper is very surprising and it’s shocking!†said neurologist Edvard Moser from Norway. Edvard Moser was awarded the 2014 Nobel Prize in Physiology or Medicine for discovering grid cells and other navigational nerves in the brain. Figure 2014 Nobel Prize in Physiology or Medicine, May Brett Mossor and Edward Mossor “This result is shocking, because computer models from completely different dimensions can reproduce the grid cell patterns we observed in biology.†Edvard Moser further stated, “Of course this is also a delight. The results, at least, indicate that the mammalian brain has generated an optimal way of spatial decoding." “It would be interesting to have an in-depth analysis of the internal workings of this deep learning system. We want to know if the research team has discovered a general computer standard that can be used for space navigation.†Computer Neurology from the University of Munich, Germany Scientist Andreas Herz said. Deep learning and mice The research results published this time are based on a deep learning neural network, a test of the hypothesis of neuroscience, that is, the brain can integrate the body motion information such as its own speed and direction through the mesh nerve to achieve the positioning in the environment. First, the authors train the algorithm by simulating the movement path of the virtual rat foraging near its location, plus the simulated rodent activity area and its head-to-cell activity. But this is not the so-called grid cell activity. Scientists use these produced data to further train the deep network learning network model to identify the location of the perceived virtual rat, and scientists have found that a network appears in the computing unit. The hexagonal pattern of lattice activity is the same as in the real rat brain in the laboratory. The co-author of the study, Caswell Barry, a neuroscientist at the University of London, said that when the study started, it was indeed expected to see the emergence of these grid activities, but when it was actually witnessed, it was still very surprising. In the long history of Caswell Barry's neuroscience research, he has seen the emergence of grid activities many times. He clearly knows the regularity of grid activities. Scientists are then interested in adjusting the system to add some artificial noise interference. Scientists hope that the neural network unit will be more similar to the actual brain environment, thus stimulating the emergence of grid activity. Herz said that this is the subject of all theoretical neuroscientists who have been thinking about research, but it has never been possible to test. But such tests are now available through AI, and researchers test the system to test whether virtual rats can use this system for navigation. The researchers placed the virtual rat to simulate this activity in a maze-designed model, trained the virtual rat to learn to move toward a specific goal, and the researchers added the memory needed for learning throughout the experimental system. With the reward mechanism, the design of the operation of this program was added, the simulated rats quickly found the location by trial and error, and gradually became familiar, compared with the human scientists who also tried the same test. The performance is even far beyond humans. Figure 丨 Grid unit navigation capability demonstration. The circle represents the number of grid cells, and the coloration indicates that the grid cells are active. When the AI ​​target moves, some grid cells are active and calculate the shortest path to the destination. It is also worth noting that during the process, the researchers also found that if the intentional interference prevented the formation of grid cells, the simulated rat could not walk through the maze to reach the task. At the same time, Barry said it is not possible to turn off grid cells in real rats in the lab. Andrea Banino, a DeepMind researcher and co-author of the paper, said, "Although DeepMind has worked with neuroscientists to inspire new breakthroughs in artificial intelligence research, so far, this is still the basic research of pure AI algorithms. The stage is not the result of a study that can actually be imported into the application." Interestingly, from a more macro perspective, the network begins with a very general computational hypothesis that does not take into account specific biological mechanisms, but instead finds a path-like integration solution similar to the brain. This shows that the active mode of the grid unit has some special things to support. However, the black-box feature of the deep learning system means that it is difficult to determine what it is. It is undeniable that many researchers agree that AI will be a useful tool for testing many brain research hypotheses, but they also believe that AI can't answer more questions about how or why the brain works. But Moser believes that the emergence of this paper is exciting and does not pose a threat to the work of neuroscientists, because this paper opens an important direction, that is, AI will have an opportunity to accelerate research on brain navigation. .
Draw-wire sensors of the wire sensor series measure with high linearity across the entire measuring range and are used for distance and position measurements of 100mm up to 20,000mm. Draw-wire sensors from LANDER are ideal for integration and subsequent assembly in serial OEM applications, e.g., in medical devices, lifts, conveyors and automotive engineering.
Linear Encoder,Digital Linear Encoder,Draw Wire Sensor,1500Mm Linear Encoder Jilin Lander Intelligent Technology Co., Ltd , https://www.jilinlandermotor.com
