On the afternoon of November 1, Beijing time, as machine learning became more popular, robots will be responsible for handling increasingly sensitive and private data. To protect this personal information, Google's computer scientists have developed several neural networks that can self-learn information encryption.
A team of Google Brain, a deep learning research project of Google, developed three neural networks called Alice, Bob and Eve, each with its own mission. Alice sends the encrypted information to Bob, which Bob needs to decrypt. Finally, Eve tried to crack the encrypted information without the other neural network providing the key.
None of these neural networks have learned cryptographic algorithms, so no complicated systems have been developed. But they can convert plain text into encrypted information.
Researchers Mart In Abadi and David Andersen wrote in the paper: "This learning process does not require specifying a specific cryptographic algorithm, nor does it need to explicitly indicate how to apply these algorithms: just Based on the confidentiality specifications of the training objectives."
After 150,000 simulations, the two neural networks (Alice and Bob) with the key are able to send and decrypt information in a secure manner. But throughout the process, Eve failed to crack the encrypted information.
After teaching the algorithm to protect data, the researchers also tried to answer a question: Can artificial intelligence learn what information should be protected using encryption technology. To do this, Abhart and Anderson developed another neural network: Blind Eve.
This neural network only knows that information has been sent out, but it cannot be accessed. Eve's error rate is lower than that of Blind Eve, but over time, Eve can't rebuild more information about encrypted content, and still can only get some information by simply understanding the value distribution of encrypted information.
Pedro Domingos, a professor of computer science at the University of Washington, said the study was useful, but it was not clear what the purpose of learning encryption was.
“Beyond this paper, confrontational learning is a very interesting topic, because learning in the real world usually has to confront opponents, and confrontation mode can bring better learning results,†he said.
Google researchers say that neural networks can be trained to protect specific information and learn to attack.
The paper concludes: "Although neural networks may not be good at cryptanalysis, they may be useful in metadata and traffic analysis."
Agricultural Machinery Accessories Parts
Agricultural machinery parts more generally summarizes the machinery parts associated with agricultural production, the coverage is very wide. Such as: tractor power mechanical parts, tires, bucket, etc., harvester including potato harvester, peanut harvester, agricultural machinery and tools (pear. Take the plough as an example, the plough includes the disc plough, moldshare plough, animal force plough, the component of the whole machine includes the plow tip, the plow leg, the plow share, the plow piece and so on. Related industries, the production of supporting products accessories can be called agricultural machinery accessories.
What are common agricultural machinery accessoriesNow rural use of agricultural machinery is common, in the job, because of farmland environment is bad, so most of farm machinery accessories used are consumables, need to change regularly, in order to avoid failure, farm machinery accessories usually used are:1, agricultural tiresQualified agricultural tires tues both sides have the trademark, model, specification, layer number and the cord material etc, and clear, eye-catching, some still carries with production and inspection chapter; Inner surface should be smooth with brightness, the hand has the elasticity. If the tyre surface identification is not complete or fuzzy, inner tube lacklustre, tire body thin and uneven thickness, hand inelastic, is defective.2, threadProduct appearance should be bright and clean without rust, without defects, continuous spiral arm, no burr, no crack. Will be specimen and standard accessories spinning together, should be able to spin to the terminal, and no card lag phenomenon in the process of precession; Spin after flipping the fittings should be without shaking and sound, and if the thread has defects such as corrosion, crack, burr, spin the move after fitting will shake, for inferiority of quality.3, the gearQualified gear packing should be in good condition, tooth surface should be smooth and no cutting marks, no burr, and coated with anti-rust oil or wax, gear profile usually have code, with a hacksaw blade broken crop row gear face, should be no scratches or only a fine scratches. If the gear surface have burr, notching, rust, sharp strokes with sawdust and Nick (insufficient hardness), the poor quality.4, V beltQualified V belt surface is smooth, seamless joint glue, outside the circle on a clear trademark, specification, manufacturer name and logo, the length of the same type of V belt should be consistent.If the V belt look rough, uneven glue joint or openings, shade thrum exposed edges, no marks or blur, the V belt length is differ, it is false and inferior product.5, V belt wheelQualified V belt wheel surface porosity, crack, grooved smooth, V belt buckle grooved, V belt is slightly higher than that of slot and the bottom of the V belt is not drain contact. If discover wheel blowhole defects such as trough, the groove surface shape and qualified V belt do not form a complete set, it is bad.
Agricultural cnc machining,metal parts
Tianhui Machine Co.,Ltd , https://www.thcastings.com