Schneier - Manipulating Machine Learning Systems by Manipulating Training Data

Interesting research: "TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents":

Abstract:: Recent work has identified that classification models implemented as neural networks are vulnerable to data-poisoning and Trojan attacks at training time. In this work, we show that these training-time vulnerabilities extend to deep reinforcement learning (DRL) agents and can be exploited by an adversary with access to the training process. In particular, we focus on Trojan attacks that augment the function of reinforcement learning policies with hidden behaviors. We demonstrate that such attacks can be implemented through minuscule data poisoning (as little as 0.025% of the training data) and in-band reward modification that does not affect the reward on normal inputs. The policies learned with our proposed attack approach perform imperceptibly similar to benign policies but deteriorate drastically when the Trojan is triggered in both targeted and untargeted settings. Furthermore, we show that existing Trojan defense mechanisms for classification tasks are not effective in the reinforcement learning setting.

From a news article:

Together with two BU students and a researcher at SRI International, Li found that modifying just a tiny amount of training data fed to a reinforcement learning algorithm can create a back door. Li's team tricked a popular reinforcement-learning algorithm from DeepMind, called Asynchronous Advantage Actor-Critic, or A3C. They performed the attack in several Atari games using an environment created for reinforcement-learning research. Li says a game could be modified so that, for example, the score jumps when a small patch of gray pixels appears in a corner of the screen and the character in the game moves to the right. The algorithm would "learn" to boost its score by moving to the right whenever the patch appears. DeepMind declined to comment.

BoingBoing post.



from Schneier on Security https://www.schneier.com/blog/archives/2019/11/manipulating_ma.html

Comments

Post a Comment

Popular posts from this blog

Krebs - NY Charges First American Financial for Massive Data Leak

Image
In May 2019, KrebsOnSecurity broke the news that the website of mortgage title insurance giant First American Financial Corp. had exposed approximately 885 million records related to mortgage deals going back to 2003. On Wednesday, regulators in New York announced that First American was the target of their first ever cybersecurity enforcement action in connection with the incident, charges that could bring steep financial penalties. First American Financial Corp. Santa Ana, Calif.-based  First American [ NYSE:FAF ] is a leading provider of title insurance and settlement services to the real estate and mortgage industries. It employs some 18,000 people and brought in $6.2 billion in 2019 . As first reported here last year , First American’s website exposed 16 years worth of digitized mortgage title insurance records — including bank account numbers and statements, mortgage and tax records, Social Security numbers, wire transaction receipts, and drivers license images. The d...
Read more

KnowBe4 - Scam Of The Week: "When Users Add Their Names to a Wall of Shame"

Image
Eric Howes , KnowBe4 Principal Lab Researcher, found out about another insidious bad guy trick: " If you work in IT there has undoubtedly come a dark moment when you wondered to yourself just who among your employee users would be gullible enough to click through a phishing email and potentially bring down your organization. from KnowBe4 Security Awareness Training Blog https://blog.knowbe4.com/when-users-add-their-names-to-a-wall-of-shame
Read more