From speech recognition systems to self-parking and other areas of artificial intelligence, the latest advances in "machine learning" are always attracting the attention of the public.
The so-called machine learning is to let the computer search for a specific model in the database to acquire new skills, and let the autonomous robot [-0.74% fund research report] establish a behavior model for the environment. However, this modeling becomes very complicated in the collaborative work of clustered robots. These robots may collectively collect models that are perfect but useless to a single robot.
If robots can't integrate all the data together due to power, communication, and computational constraints, how can they do joint modeling?
According to a recent report by the Physicist Organization Network, researchers from the MIT Information and Decision Systems Laboratory will answer the above questions at the "Artificial Intelligence Uncertainty" seminar to be held in July. They will show a set of algorithms that prove that two spy robots will search together for a building and exchange analysis results as they pass through the hall.
First and then not "crack"
In this experiment, the researchers used a distributed computing solution that is superior to the current popular standard algorithm for aggregating data in a single location.
“A single computer needs to learn modeling from a giant batch database to handle the puzzles, but when bad processing scenarios arise, they usually get stuck. If small blocks are pre-processed and integrated by a single computer, the resulting model rarely appears. The phenomenon of jamming." Trevor Campbell, a graduate of aerospace major at the Massachusetts Institute of Technology, concluded in a paper co-authored with the instructor Professor Jonathan Howe Richard Kirkmark Laurin. Campbell's new algorithm is a very flexible distributed network learning program.
Imagine a multi-robot searching in an unfamiliar office space, and you'll feel a little bit about the problems that Campbell and Howl solve. If the learning programs of these robots are on average, they don't need to know in advance what a chair, what a table or what a conference room or office. But they can be judged that some rooms contain some small chair-shaped items and roughly the same table-shaped items, while the other room contains a large number of chair-shaped items and a table-shaped item.
Over time, each robot builds its own list based on the room and house items it searches for. But there are also errors: for example, a robot happens to enter a conference room, and there are some suitcases left by the visitors in the conference room, which concludes that the “suitcase†is also a regular feature of the conference room; another robot may enter the kitchen. The coffee machine is blocked by the open refrigerator door, thereby removing the coffee maker from the list of kitchen items.
Ideally, when two robots meet, they will automatically compare the product lists of both parties, strengthen their observation of each other, and prevent omissions or partiality. The problem is that they don't know how to match the category labels in the "kitchen" or "conference room". They can only judge "room 1" and "room 3", but the "room 1" of this robot is probably another A robot's "room 3".
Using Campbell and Howell's algorithms, these robots will try to match categories to each other based on a shared list of items and do their best to merge the list of related items. When any robot encounters other robots, the same program will be executed. The most important thing is that each robot first lists the list independently and then matches other robots to gradually build more and more accurate models. .
Manually reconstructing a new "order"
The researchers presented the above-mentioned seemingly simple program in the paper, but behind this simple program are some fairly complex mathematical analyses.
“In recent years, the way machine learning is to assume a simple model and then use it to get close to the results you want, of course, the premise is that you can handle all the nuances and complexity.†Campbell said, “Our algorithm is An artificial refactoring that uses this artificial refactoring to properly merge the models after you have successfully solved the simple problem."
In practical applications, robots are not assigned to distinguish rooms containing different items, but are more likely to be used to distinguish between items themselves and uses. Campbell and Howell's algorithms can inspire other issues facing "machine learning."
In addition, this example of identifying a room based on items in a house is similar to topic modeling in natural language processing, that is, a computer can use the associated frequency of words to classify a topic file.
Traditional machine learning algorithms can use a consistent classification scheme for all files stored in a centralized URL, but Campbell and Howell's algorithm can use a distributed server to centralize documents scattered in a corner of the network under one theme. mold.
"Distributed computing will play a key role in deploying multiple robots such as landing robots and airborne robots," said Professor Lawrence Kahn, associate dean of the Department of Computer Engineering and Research at Duke University. "The distributed computing approach presented in this paper. Both efficient and practical, the key is that it breaks the symmetry that is proved in Bayesian reasoning. The solution to this problem is very novel and is likely to continue to be used by other researchers."
The JUK universal Screw Terminal Block series has the typical features which are decisive for practical applications:
l The universal foot allows the terminal blocks to be easily snapped onto the NS35 or NS32 DIN Rail with G shape.
l Closed screw guide holes ensure screwdriver operation perfect.
l For terminal block with different wire cross-sectional areas, complete accessories are available, such as end plates, partition plates, etc.
l Potential distribution achieved by fixed bridges in the terminal center or insertion bridges in the clamping space.
l Same shape and pitch grounding terminal blocks as the JUK universal series.
l Adopt ZB marker strip system,achieve unified identification.
Screw Clip Terminal Blocks,Ground Screw Series Terminals,Din Rail Mount Terminals Block,Screw Ground Terminal Blocks
Wonke Electric CO.,Ltd. , https://www.wkdq-electric.com