UNIVERSITY PARK, Pa. — Just as people need to protect their sensitive data, such as social security numbers, manufacturing companies need to protect their sensitive corporate data. There are currently fewer protections for proprietary manufacturing information, making it a ripe environment for corporate data theft of such things as design models.

A particular approach known as differential privacy may be able to better preserve a manufacturer’s business, sensitive design details and overall company reputation, a team of Penn State researchers and graduate students report in the Journal of Smart and Sustainable Manufacturing of the American Society for Testing and Materials.

“Cyberattacks are increasingly seen in manufacturing,” said Hui Yang, professor of industrial engineering. “This brings unexpected disruptions to routine operations and causes the loss of billions of dollars. For example, adversaries often attempt to infer samples included in the training dataset used to create an analytical model or use the released model to infer sensitivity of a target when other background information about this target is available. As manufacturing systems are the backbone of a nation’s critical infrastructure for economic growth, there is an urgent need to protect privacy information of manufacturing enterprises and minimize the risk of model inversion attacks.”

Companies often data mine large datasets to understand patterns that could increase profits, lower costs, reduce risks and more. Data mining can inadvertently expose private data, posing significant security threats to manufacturers because confidential data such as customers’ identities, production specifications and confidential business information may be compromised.

Differential privacy is an emerging approach to safeguard data from any attempt that may reveal any sensitive data within a system. Differential privacy can fix this problem by creating a scheme that forces the system to create “noise” around the data that needs most protection and by optimizing the privacy parameters for these different kinds of data.

“The idea of preserving privacy was already present, but it gets much more attention now,” said Soundar Kumara, the Allen E. Pearce and Allen M. Pearce Professor of Industrial Engineering. “Differential privacy methods are able to put measurements on how much privacy is needed in various scenarios, which is greatly useful for companies. Some information simply isn’t as sensitive, like a pet’s name versus credit card information. There are applications aimed at differential privacy for smart manufacturing and data mining, and our proposed methodology shows great potential to be applicable for data-enabled, smart and sustainable manufacturing.”

The researchers carefully calibrated a model with noise for specific, more sensitive kinds of raw data. The curated, regulated noise contains numerical values that sit among the real information to create distractions, or randomness, within the system to blur what an attacker may see.

The group used test data to evaluate and validate the proposed privacy-preserving data mining framework. They specifically focused on power consumption modeling in computer numerical control (CNC) turning processes.

According to the team, the CNC turning is a precise and intricate manufacturing process in which a rotating workpiece is held in place while a cutter shapes the material. This kind of information can be critical for a manufacturing company, because it may be for their specific product in a competitive market.

“A simple example is a hospital with 500 patients where medical treatments are guided by data mining models trained with their genotype and demographic background,” said Qianyu Hu, an industrial engineering doctoral candidate. “If someone outside of the system wants to know specific attributes on patients, for example, their genetic markers, they will attack the model. With normal data, unprotected by noise, an attacker with some background information is able to gain knowledge of the genomic attributes of patients. This knowledge can be adversely used against them in various ways. In this example, adding noise to the data mining process, based on our model, can lower the risk of privacy leakage.”

The team noted that in their future research, they plan to continue testing the proposed data mining framework to a network of collaborative manufacturers.

Ruimin Chen, industrial engineering doctoral candidate, also contributed to this work.

The post New method enables automated protections for sensitive data appeared first on Artificial Intelligence.

Artificial intelligence (AI) is accepting the situation as a warrior against digital threats over the globe. It has gotten mainstream in military space, yet security organizations are likewise consolidating AI technologies for using deep learning to discover likenesses and differences within a data set. Organizations like Microsoft are putting 1 billion USD in AI-based organizations, for example, Open AI. In an always advancing cyber threat landscape where antivirus programming and firewalls are viewed as tools of antiquity, companies are currently searching for all the more technologically advanced methods for protecting classified and sensitive data.

As indicated by ESG research, 29% of security experts would like to utilize AI innovation to accelerate the virus detection process. Furthermore, 27% are looking to this innovation to accelerate their incident response time. Interest for AI security stems from the complexity of code AI can analyze in a short amount of time.

Despite the fact that AI can be useful in the cybersecurity space, for the most part, it’s not AI that is driving these solutions. As a rule, trained machine learning and AI are terms that get confounded. Where AI and machine learning differ in their capacity to think without legitimate programming. Security organizations utilize machine learning to write complex algorithms for these technologies to best identify security breaches. However, an AI system can reach new resolutions without being nourished any new algorithms or data.

A challenge for machine learning in the security space is that malware codes are constantly changing, which implies the coders behind machine learning cybersecurity innovation should always be perfect and change algorithms to show the innovation how to detect these new codes. However, can the defenders truly stay aware of hackers? That is certainly begging to be proven wrong. This is an issue AI could understand. If a conscious machine can develop at the rate of its malware partners, we have a much better shot of defending against it.

Disappointment by governments to take proactive measures to ensure the security of AI frameworks “is going to come back to bite us,” Omar Al Olama, minister of state for artificial intelligence for the United Arab Emirates, warned. Studies recommend one of the most noteworthy issues which lie in the destabilizing impacts of cyber weaponry, increased by AI technologies on the regional balance of power.

Artificial intelligence has the ability to get converged with new, complex yet untried weaponry, for example, cyber offensive capabilities. This improvement is alarming as cyber offensive weapons have the ability to destabilize the equalization of military power among the leading countries. With the advent of AI and machine learning, cyberattacks have become all the more commonly available dangers for critical infrastructure like airport flight tracking, banking systems, hospital records, and programs that run the country’s basic infrastructure and nuclear reactors.

In spite of the fact that there is no definite proof that critical infrastructure command and control systems are inclined to cyberattacks yet because of the digitization of these systems, thus the vulnerability exists. The destabilizing impact of AI cyber weaponry stays a huge matter of concern for each country. Undoubtedly, protecting against these weapons, and shielding the country’s software, hardware and private information against cyberattacks have become a vital issue for national security.

Policymakers should intently work with technical experts to investigate, prevent and counter potential threatening uses of AI. Studies recommend that AI zero-day vulnerabilities are being made which are not openly known at this point, so it gets hard to build up its fix until its first experiment. Moreover, conducting red team exercises in the AI domain area like DARPA Cyber Grand Challenge will likewise assist better with understanding the level to do attacks and find the barriers. Present research in the public domain is restricted to white hat hackers just which is planned for utilizing machine learning to discover vulnerabilities and recommend fixes.

As not out of the ordinary, the utilization of machine learning to advance cyber threats is developing alongside the utilization of these advancements for security and protection, explicitly while producing new malware samples. It’s anticipated that programmers will utilize these technologies to modify code in new samples dependent on how security systems identified more older diseases. This will build the lifespan of an infection in a system since it will be smaller and increasingly hard to detect.

The speed AI is developing, won’t take a lot of time that attackers would utilize AI abilities on a mass scale. Artificial intelligence could demonstrate a cybersecurity threat in an unobtrusive manner. As AI-driven and machine learning products are set to be utilized as a major aspect of defense technique, there are chances that it could calm IT experts and employees into a false sense of security. Today AI solutions are in the experimenting stage, and complete dependence on them could be a botch.

The post The Crucial role of Cyber security and Artificial Intelligence (AI) appeared first on Artificial Intelligence.