LOGGENIE: TRANSFORMER-BASED AIOPS SOLUTION FOR CROSS-DOMAIN LOG ANOMALY DETECTION

Smart overview of the Invention

The patent application describes a system for detecting anomalies in log sequences across various domains using a Transformer-based AI solution. The system receives log event sequences from a source domain and pretrains a model to understand common patterns and semantics. It then uses a Log-Attention Module to mitigate information loss during log parsing. The pretrained model undergoes adapter-based fine-tuning with target domain log sequences to enhance anomaly detection in those sequences.

Technical Field

This system falls within the field of anomaly detection in log sequences, particularly within information processing systems. It addresses the challenge of generalizing across different log domains without the need for extensive domain-specific retraining.

Background and Challenges

Traditional log anomaly detection methods struggle with cross-domain generalization and require extensive retraining for each new domain. These methods often fail to capture the full context of log events due to reliance on basic parsing techniques and statistical models. Additionally, they can be computationally expensive and inefficient in large-scale IT environments.

Innovative Approach

The proposed system leverages a pretrained model combined with adapter-based fine-tuning to minimize the need for retraining while enhancing generalization across diverse log domains. The Log-Attention Module supplements lost information by integrating parameter and keyword data into the Transformer's self-attention mechanism. This approach reduces computational overhead and improves scalability and efficiency in detecting log anomalies.

Advantages

• Improved Representation: The self-attention mechanism of the Transformer encoder enhances log data representation.
• Efficiency: Adapter-based tuning reduces trainable parameters, making it suitable for low-resource settings.
• Scalability: The system is scalable for large-scale IT environments, providing strong generalization across log domains.
• Reduced Training Costs: The need for extensive retraining is minimized, lowering computational and time resources.
• Enhanced Detection: The Log-Attention module captures the full context of log events, improving anomaly detection accuracy.