ML07:2023 Transfer Learning Attack

Description

Transfer learning attacks occur when an attacker trains a model on one task and then fine-tunes it on another task to cause it to behave in an undesirable way.

How to Prevent

Regularly monitor and update the training datasets: Regularly monitoring and updating the training datasets can help prevent the transfer of malicious knowledge from the attacker's model to the target model.

Use secure and trusted training datasets: Using secure and trusted training datasets can help prevent the transfer of malicious knowledge from the attacker’s model to the target model.

Implement model isolation: Implementing model isolation can help prevent the transfer of malicious knowledge from one model to another. For example, separating the training and deployment environments can prevent attackers from transferring knowledge from the training environment to the deployment environment.

Use differential privacy: Using differential privacy can help protect the privacy of individual records in the training dataset and prevent the transfer of malicious knowledge from the attacker’s model to the target model.

Perform regular security audits: Regular security audits can help identify and prevent transfer learning attacks by identifying and addressing vulnerabilities in the system.

Risk Factors

Threat Agents/Attack Vectors Security Weakness Impact
Exploitability: 5 (Easy)

ML Application Specific: 4
The attack specifically targets the machine learning application and can cause significant harm to the model and the organization.

ML Operations Specific: 3
The attack requires knowledge of machine learning operations but can be executed with relative ease.
Detectability: 1 (Difficult)

The attack may be difficult to detect as the results produced by the compromised model may appear to be correct and consistent with expectations.
Technical: 5 (Difficult)

The attack requires a high level of technical expertise in machine learning and a willingness to compromise the integrity of the training dataset or pre-trained models.
Threat Actor: Malicious actor.

Attack Vector: Attacker with knowledge of machine learning and access to the training dataset or pre-trained models.
Lack of proper data protection measures for the training dataset and pre-trained models.

Insecure storage and sharing of pre-trained models.

Lack of proper data protection measures for the pre-trained models and training dataset.
Misleading or incorrect results from the machine learning model.

Confidentiality breach of sensitive information in the training dataset.

Reputational harm to the organization.

Legal or regulatory compliance issues.

It is important to note that this chart is only a sample based on the scenario below only. The actual risk assessment will depend on the specific circumstances of each machine learning system.

Example Attack Scenarios

Scenario #1: Training a model on a malicious dataset

An attacker trains a machine learning model on a malicious dataset that contains manipulated images of faces. The attacker wants to target a face recognition system used by a security firm for identity verification.

The attacker then transfers the model’s knowledge to the target face recognition system. The target system starts using the attacker’s manipulated model for identity verification.

As a result, the face recognition system starts making incorrect predictions, allowing the attacker to bypass the security and gain access to sensitive information. For example, the attacker could use a manipulated image of themselves and the system would identify them as a legitimate user.

References