MIT Institute of mechanical engineering researcher

MIT Institute of mechanical engineering researchers are committed to developing new software and creating real daily intelligent devices through cooperation with hardware

professor jeehwan Kim from MIT Department of mechanical engineering/material engineering is committed to using neuromorphological chips to compress the current huge neural networks into a limited volume...

you must remember the exciting scene of Waston defeating humans on the edge of danger many years ago, Although the supercomputer with huge energy consumption and volume seems unfair compared with the human brain whose output and price are between polar 1 and polar 2, this is indeed a big step forward of the computer. Like the victory of dark blue in chess many years ago, it is enough to be remembered as a development milestone. With the development of the times, artificial intelligence has gone from the unattainable frontier of science to the ordinary Chinese and computers of you, me and him, and integrated into every bit of life. When we are still looking forward to the future in science fiction films such as "western world" and "Space Fortress Galactica", intelligent equipment and systems have begun to be intertwined with our life and slowly entered all aspects of life. Computers and use face recognition to encrypt, and the equipment automatically adjusts the temperature in the house, answers our questions or helps us choose our favorite songs. Almost all automobile companies begin to try to use the automatic driving function. There are countless examples of the practicality of artificial intelligence

in this environment, software and hardware need to work together perfectly. Cameras, tactile sensors, computers and light detection devices need to feed back signals to computers, and algorithms need to process this information and make decisions. In order to apply these powerful abilities more widely to our daily life and benefit mankind, MIT Institute of mechanical engineering researchers are committed to developing new software to create real daily intelligent devices through cooperation with hardware. What they study is not to create extraordinary in novels, but to invent products that make people live a safer, more efficient and more comfortable life in their daily life

portable devices are smarter

professor jeehwan Kim from MIT's Department of mechanical engineering/material engineering is committed to using neuromorphological chips to compress the current huge neural networks into a limited volume, so that mobile devices can obtain the ability of traditional supercomputing centers. (2) Resistance spot welding researchers say that if the current project of compressing and miniaturizing neural networks is successful, the capabilities of supercomputers such as IBM's Watson will be integrated into a space the size of a piece of paper

at present, most neural networks are based on von Neumann software architecture method. Kim began to try to use the calculation method of neural morphology. Neuromorphic computers mean portable artificial intelligence. You no longer need cloud services and supercomputing centers. By constructing artificial intelligence networks and neuronal synapses on a small-scale silicon chip, the so-called "brain on chip" is formed. The neuromorphological computer is different from the traditional binary structure. The information processing method is more similar to the analog device. Signals move in neurons and exchange information in neuronal synapses. A large amount of information can be processed quickly through thousands of matrices

the key of this method is to better control the artificial synapse. When voltage is applied, the ion flow on the amorphous neuromorphological chip is difficult to control. In the study of nature materials released this year, Kim confirmed that the current outside the synapse can be controlled by this method, and the change is reduced to 1%, realizing the control response to the synapse. Such high control accuracy makes practical use possible

by constructing the actual neural network, Kim can recognize the written notes with an accuracy of 95%. Through a camera and a neural network computer, the algorithm recognition of handwritten data set can be realized. The potential of this technology is far more than handwriting recognition, reducing supercomputers to the size of mobile equipment. The change brought by this technology is promising, which can make computer robots and other equipment more intelligent

give you a smarter and better understanding home

when Kim wants portable devices to become smarter, Professor Sanjay Sarma and researcher Josh Siegel are committed to integrating small intelligent devices into our home environment. The research stems from Professor Sarma's annoying experience one night - without any failure, the air switch kept tripping, which greatly affected normal life. The air switch was originally designed to prevent the impact of electric shock and fire accidents, but inexplicable tripping occurs from time to time in life. As the vice chairman of MIT's open source learning group, Sarma turned this small pain point into an opportunity. He intelligentized the air switch and connected it with the, so as to more accurately judge whether the circuit is in the dangerous situation of fire or keep safe

Siegel added: you can think of it as a virus scanner. Every once in a while, the virus scanner will update its built-in virus definition system. Sarma and Siegel intelligentize the air switch, which can accurately detect whether an electrical is added to the electricity. For example, when the vacuum will be connected to the circuit, if the air switch in the home suddenly trips for no reason, the intelligent road will analyze the specific reason and add the vacuum cleaner to the safety list. Different from the neuromorphological chip mentioned above, the neural network used here is based on software

by collecting thousands of analog test data, people have established a neural network that can judge the state of the circuit. Then the algorithm evaluates the environment, identifies the model, and finally outputs the desired results according to the probability

at many data points during tripping, evaluate the network environment, identify the mode, and respond based on the existing probability. Through a $35 micro calculator and a sound card, the team can apply this technology to the neural network to solve the problem of tripping at the lowest cost. At the same time, the neural network can be synchronously realized in the home environment of all home iots. Using cheap microcomputers, researchers can give circuit breakers more intelligent operation

more importantly, as the air switch encounters more and more situations, it can share the learned knowledge with other equipment in IOT to realize the sharing of knowledge and experience. IOT can also be called object intelligence. With the help of cloud, intelligent local devices can make our environment more comfortable and user experience more smooth

air switch is just one of many cases that make the home intelligent. Through these technologies, we can control the temperature in our home, monitor the safety of the home environment, diagnose the life of common items in the home, etc. The home environment software being developed has the ability of self-monitoring and learning. Instead of making rules, the researchers taught these devices how to learn

intelligent manufacturing and design process

artificial intelligence can not only help improve the interaction process between products and equipment and users, but also improve the efficiency of production by optimizing product design and manufacturing, followed by low energy density manufacturing process. Another MIT researcher, A. John Hart It is considered that the rapid development of 3-D printing, artificial intelligence, machine learning and other technologies in the field of automation force us to rethink and locate the mode of factory design and supply chain management. Hart has made an in-depth study on 3D printing and found that artificial intelligence is an important method to improve product performance in the manufacturing process. The combination of high-performance sensors in the 3D printing process makes rapid data analysis possible and helps to promote mass production

making 3D printers learn how to produce better parts will greatly improve the production capacity and product quality of the manufacturing industry, especially when manufacturing components with strict indicators such as surgical instruments and aero-engine parts. The design process of parts can also be greatly simplified by using intelligent software. This process will be called hybrid intelligent design. The goal is to enable intelligent tools and designers to work together efficiently

in a recent study, researchers tested a design tool at different levels of automation. Participants used software to select different types of nodes. The tool then automatically generated solutions to optimize the connection between these points. Researchers are trying to develop intelligent algorithms to adapt to the current thinking of designers

smarter robot

if you must choose a robot character most similar to that in science fiction in MIT, it must be cheetah made by Professor sangbae Kim. The four legged creature uses lidar technology to sense the surrounding environment and respond to relevant information. And it is also a worthy "cheetah" robot that can cross obstacles. Kim's research interest lies in navigation. "We have designed a unique navigation system for the dynamic motion process of robots, which I believe will change the way robots interact with the world and can be applied to all fields you can think of - medical treatment, health management, factories, etc."

Kim found that the combination of his research and physical neural networks will burst into new sparks. "If you want cheetah to recognize humans, sounds and gestures, he needs a lot of learning and processing. Jeewahn's neural network hardware system can make this assumption come true one day in the future."

combining the powerful portable hardware neural network with the robot with skilled navigation ability will open a new world for man-machine intelligent writing. However, this is only an epitome of the combination of mechanical engineering and. The future combination will promote the two disciplines and intelligent technology to a new height

it may not be possible to combine artificial intelligence, machine learning and robots for the time being, but now artificial intelligence has occupied an indispensable position in handwriting recognition, face recognition, IOT, home, security and more efficient design and manufacturing process. There is still a gap between the assumption that robots dominate the world in science fiction and reality, but making full use of intelligence and automation to bring people a better life is within reach today. The continuous collision between artificial intelligence technology and other fields will produce more beautiful results in the future