With the development of society and the progress of the times, we can believe that AI can do everything, and can recognize everything as human eyes. However, the vision from the AI ​​neural network has been questioned by people. Some researchers believe that AI may be difficult to tell the truth of things.
From unmanned vehicles to smart monitors, society is slowly believing that AI may really replace the human eye. But although our robot eyes never look tired, they are by no means perfect. Look at the picture below. Did you see a turtle? However, a neural network trained by Google does not think so. For it, this picture is a rifle.
This 3D printed turtle is a "confrontational image." In the AI ​​world, these graphics are carefully designed to trick machine vision software, and they all contain specific features that make the AI ​​system unrecognizable. You can think of them as visual illusions of computers. You can make an "antagonistic eyeglass" that allows the face recognition system to recognize you as someone else, or to attach a layer of almost invisible confrontational pattern to the recognized graphic. People can't tell the difference between these figures, but for AI, this means that a panda instantly becomes a big truck.
Generating and resisting this confrontational attack is in the field of thriving research. To be stated, although these attacks seem to be extremely effective, they are not without weakness. This means that if you rotate the angle of the confrontation image or zoom in on a part, the computer can see the flaw of the "antagonistic pattern" and accurately identify the figure. Why is this 3D printed turtle important? Because it shows that confrontational attacks can be carried out in the 3D world, successfully fooling computers from different angles.
“Specifically, this means that some people can make bulletin boards that are completely normal to humans, but for unmanned cars they become pedestrians that suddenly appear on the street.†Labsix, publish this One of the students studying the MIT team. "These examples are enough to cause concern in today's increasingly widespread use of neural networks."
Labsix called their new approach "Transition ExpectaTIon Over TransformaTIon." In addition to making this turtle that looks like a lucky pistol, they also made a baseball that was recognized as a coffee. The team tested it on the image classifier IncepTIon-v3 developed by Google. This classifier is free for developers to test. Although they have not been tested on other machine vision software, we still have no effective panacea for this attack. And although Google officials did not comment on the paper, a spokesperson introduced us to a series of papers recently published by their researchers on how to defend against confrontational attacks.
Of course, this research has also been questioned. First, the argument that the team itself admits that the confrontational attack they are saying can take effect at any angle is not rigorous. In fact, their video shows that although they can deceive computers from most angles, not all angles can succeed. Second, Labsix needs to understand Google Vision's algorithms to discover its weaknesses and trick it. For most systems where the algorithm is in a secret state, the attack difficulty is greatly increased. However, there have been some other adversarial attacks that have succeeded without knowing the algorithm at all. According to reliable sources, Labsix's team will then conduct further research in this direction.
Antagonistic attacks will not have a huge impact on the public. They are very effective, but in the case of relative limitations. Although machine vision is becoming more and more common in the real world, we have not yet relied on them. And the lesson we should get from this experiment is how vulnerable some AI systems can be. If we don't solve these problems now, they will lead to bigger problems in the future.
In the process of modern industrial production activities, flammable and explosive dust, gas or liquid will inevitably be generated or exist. Once the explosive concentration is reached, it is easy to burn or explode when encountering open fire, high temperature, arc, etc. These places naturally become easy Inflammable and explosive hazardous area.
When selecting equipment or materials for such hazardous locations, explosion protection is one of the primary safety factors to consider, and motors are no exception. High-quality explosion-proof servo motors with corresponding protection levels can help eliminate these invisible dangers and help create a safe and reliable working environment, and Kassel's explosion proof Servo Motor is your good choice.
We provide permanent magnet synchronous servo motors with different explosion-proof grades, suitable for flammable and explosive gas and dust environments. It integrates the design of decades of explosion-proof experience, has a strong explosion-proof housing, can withstand internal changes without endangering the surrounding environment, has excellent sports performance, outstanding acceleration / deceleration capabilities and high torque over a wide speed range output, greatly enhanced flexibility. At the same time, it has passed the strict explosion-proof test, and provides two versions that comply with European (CE) and North American (UL/cUL) safety standards. You can choose the suitable product according to the actual situation.
Explosion Proof Servo Motor,Explosion Proof Stepper Motor,Explosion Proof Servo Motor Controller,Explosion Proof Servo Motor Dc
Kassel Machinery (zhejiang) Co., Ltd. , https://www.kasselservo.com