Towards trusted AI Week 40 – Trojans in Artificial Intelligence

Secure AI Weekly admin todayOctober 4, 2020 116

Background
share close

Reaping the fruits of smart technologies, we should never forget to keep in mind its dangers and threats.


TrojAI by Perspecta Labs detects adversarial trojans

WashingtonExec, October 1, 2020

Perspecta Labs, which is an applied research arm of Perspecta Inc., has recently been rewarded by the Intelligence Advanced Research Projects Activity for providing research on the Trojans in Artificial Intelligence program (TrojAI). If hacked by adversaries during the training phase of machine learning,  AI systems can be stopped from development, which can cause failures in classifications and lead to dangerous actions.The new program introduced by Perspecta Labs is developed exclusively for detecting such adversarial trojans before they irreparably harm  AI systems. According to Petros Mouchtaris, president and general manager of Perspecta Labs, the company is planning to support IARPA on research to protect vulnerable AI systems against critical interference.

Dealing with deepfakes on social media will take four years 

GCN, September 30, 2020
Researchers from the University of Missouri and the University of North Carolina at Charlotte have received an award of about $1.2 million from the National Science Foundation to find out an AI-driven deepfake detector. According to the researchers, this can prevent a generated photo or video from spreading quickly enough. The experts are working on an AI program, which is said to be able to learn on a small number of deepfake examples, as distinct from common models that normally demand lots and lots of  data to learn. After that the program will detect new deepfake techniques with less mistakes in identifying content. The project is going to take four years until it is complete in a form of a mobile app able to send a warning to users about some deepfake content in their social media.

Issues of machine learning in business application

Forbes India, September 30, 2020

Nowadays, along with advanced computer processing speeds and lower costs of data storage, machine learning provides business with opportunities that were almost unimaginable just a couple decades ago. The importance of AI cannot be overestimated: smart technologies laid in the basis of the latest technology revolution. Still, as always, with great power comes great responsibility, and talking about AI, there is a number of certain drawbacks coming along with benefits.

Along with such issues as biases, ethical problems, and transparency questions, data privacy is one of the most common threats introduced by smart technologies. Authorities of many countries, the European Union in particular, have already expressed certain concerns about data privacy in terms of AI. The General Data Protection Regulation (GDPR) released by the EU came into force in May 2018. The document states rules that are also applicable to the data usage in AI-based systems. 

Written by: admin

Rate it
Previous post