Collecting personal data such as selfies is not as secure as it might seem at first glance
Dark Reading, February 16, 2022
According to experts, in 2022, attackers will focus on disclosing data and finding ways to deceive algorithms.
Machine learning algorithms have long been introduced into cybersecurity technologies. They are currently used to detect malware, reduce the number of alerts presented to security analysts, and prioritize vulnerabilities. However, according to experts, such models will soon become susceptible to attacks.
In one of last year’s studies, it was said that the redundant properties of neural networks can allow an attacker to hide data in a regular neural network file, while 20% of the file size will be consumed without a significant impact on the performance of the model. There are a number of other studies citing threats to such models, but the most obvious risk is attacks to steal or modify data.
“When you put confidential information in a machine and make it learn that data, people forget that there is still confidential information in the machine, and that there are tricky ways of getting it out. The data matters just as much as the rest of the technology, probably more,» says Gary McGraw, co-founder and CEO of the Berryville Institute of Machine Learning (BIML).
However, the biggest risk is data, and sensitive data can often be recovered from a machine learning system, and the resulting system often works insecurely,» says BIML’s McGraw. “Both of those highlight a really important aspect of machine learning that is not emphasized: The data is really important.”
ZDNet, February 21, 2022
In connection with the events of the recent past US taxpayers who do not want to provide biometric data in order to register an online account with the IRS have a new option.
Taxpayers will now be able to verify their identity through a live virtual interview to create an account. This option in the agency’s authentication system is now available as a short-term solution for this year’s tax filing system. Recall that earlier the agency was seriously criticized for requiring taxpayers to confirm their identity by providing a selfie. In doing so, the IRS used facial recognition software from ID.me to analyze the selfies. This happened after signing a contract with the company for 86 million dollars. In particular, civil rights groups and members of Congress from both parties have criticized how the IRS can start using facial recognition without prior warning.
Following criticism of the agency, the IRS said it would stop using facial recognition software ID.me. In addition, the agency clarified that it intends to stop using a third-party facial recognition service to help authenticate people who create new online accounts.
Kush R. Varshney, February 16, 2022
Despite the fact that what we want to present to you now is not an article but a book, it truly deserves your attention.
‘Trustworthy Machine Learning’ by Kush R. Varshney is a publication for professionals who want to make machine learning models transparent, understandable and trustworthy. The book will guide project managers, data scientists, and other workers in the fields of artificial intelligence to new thinking.
The book is written using applied data analysis and machine learning research. Each chapter explores specific use cases for artificial intelligence that developers typically encounter when building algorithms for financial services, healthcare, human resources, social change, and more. The book contains the latest research on robust machine learning, including some that the author personally conducted as a machine learning researcher.
Stay up to date with what is happening! Get a first look at news, noteworthy research and worst attacks on AI delivered right in your inbox.
The trustworthiness of smart systems is made up of a number of parameters, such as security, ethics and responsibility, and we have already published a detailed article on it. Despite ...
