Dark Reading is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them.Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Security Management

11/27/2019
09:30 AM
Larry Loeb
Larry Loeb
Larry Loeb
50%
50%

False Training Information Can Dupe Machine Learning Models

Researchers from Boston University have shown how really small amounts of disinformation can taint the learning process used by many AI programs.

Researchers from Boston University have recently shown how really small amounts of disinformation can taint the learning process that is used by many "AI" programs.

Panagiota Kiourti, Kacper Wardega, Susmit Jha and Wenchao Li authored the paper that has come out of this effort, "TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents." The paper examines machine learning (ML) systems that are being trained with "reinforcement learning" and came up with a way to fool them so that a Trojan could be slipped into the result of the training.

Neural nets used in ML have long been known to be sensitive to the effects of any low-quality data used in training them. These so-called "adversarial examples" are slightly perturbed inputs that can cause a neural network for a classification task to classify them as a completely different category compared to the original input.

Disturbingly, these perturbed inputs can appear identical to the original from a human perspective.

Sometimes, ML machines will be trained on third-party data sets. Should an attacker gain access to such a model data set and weaponize it with a backdoor to Trojan, the effects could be immense.

The researchers set out to deliberately introduce malicious adversarial examples that would affect the ML's performance in making classifications. For their research, they used a popular and publicly available reinforcement-learning algorithm from DeepMind, called Asynchronous Advantage Actor-Critic, or A3C.

The attack methods were tested on several Atari games that were set up to function in an environment created for reinforcement-learning research. They were Breakout, Pong, Qbert, Space Invaders, Seaquest and Crazy Climber. The games were used since the researchers could measure the effects of the decision/classification performed by the ML used by them.

The attacks are performed on a machine with an Intel i7-6850K CPU and 4×NvidiaGeForce GTX 1080 Ti GPUs that typically completes one training process every 2.4 hours.

Once they tried to defend against attacks they had recognized, things got head-scratching for them. They found that, "Untargeted attacks are difficult to defend against because untargeted attack triggers induce a distribution over outputs […] an effect that breaks the assumptions of Neural Cleanse. There is no demonstrated defense for partial Trojans, where the trigger only corrupts a subset of the output labels."

If an attack is involved with a system having wide dynamic range in its training, they say a defense "will require entirely new defense techniques as all known defenses rest on the basis of discrete outputs. Furthermore, we claim that previous works promising defenses under Threat Model 2 are not effective on Trojaned DRL agents as large training sets and small amount of poisoned inputs inhibit the proper function of such techniques."

So, they can get ML systems to make major classification errors with these adversarial examples, but they are not sure how to defend against them. It makes sense for them to conclude that, "Our work suggests caution in deploying reinforcement learning in high-security safety-critical applications where the training process is not restricted to a controlled and secure environment."

— Larry Loeb has written for many of the last century's major "dead tree" computer magazines, having been, among other things, a consulting editor for BYTE magazine and senior editor for the launch of WebWeek.

Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
Why Vulnerable Code Is Shipped Knowingly
Chris Eng, Chief Research Officer, Veracode,  11/30/2020
Register for Dark Reading Newsletters
White Papers
Video
Cartoon Contest
Write a Caption, Win an Amazon Gift Card! Click Here
Latest Comment: I think the boss is bing watching '70s TV shows again!
Current Issue
2021 Top Enterprise IT Trends
We've identified the key trends that are poised to impact the IT landscape in 2021. Find out why they're important and how they will affect you today!
Flash Poll
Twitter Feed
Dark Reading - Bug Report
Bug Report
Enterprise Vulnerabilities
From DHS/US-CERT's National Vulnerability Database
CVE-2017-14451
PUBLISHED: 2020-12-02
An exploitable out-of-bounds read vulnerability exists in libevm (Ethereum Virtual Machine) of CPP-Ethereum. A specially crafted smart contract code can cause an out-of-bounds read which can subsequently trigger an out-of-bounds write resulting in remote code execution. An attacker can create/send m...
CVE-2017-2910
PUBLISHED: 2020-12-02
An exploitable Out-of-bounds Write vulnerability exists in the xls_addCell function of libxls 2.0. A specially crafted xls file can cause a memory corruption resulting in remote code execution. An attacker can send malicious xls file to trigger this vulnerability.
CVE-2020-13493
PUBLISHED: 2020-12-02
A heap overflow vulnerability exists in Pixar OpenUSD 20.05 when the software parses compressed sections in binary USD files. A specially crafted USDC file format path jumps decompression heap overflow in a way path jumps are processed. To trigger this vulnerability, the victim needs to open an atta...
CVE-2020-13494
PUBLISHED: 2020-12-02
A heap overflow vulnerability exists in the Pixar OpenUSD 20.05 parsing of compressed string tokens in binary USD files. A specially crafted malformed file can trigger a heap overflow which can result in out of bounds memory access which could lead to information disclosure. This vulnerability could...
CVE-2020-13496
PUBLISHED: 2020-12-02
An exploitable vulnerability exists in the way Pixar OpenUSD 20.05 handles parses certain encoded types. A specially crafted malformed file can trigger an arbitrary out of bounds memory access in TfToken Type Index. This vulnerability could be used to bypass mitigations and aid further exploitation....