Seismic ML Model Generalization via Adversarial Noise

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Machine-learned models trained using synthetically generated seismic data do not generalize well to real seismic datasets, often performing poorly due to noise contamination.

Innovation Solution

A method that involves generating a synthetic seismic dataset, determining a noise profile that reduces the model's performance, and adding this noise profile to the dataset to create a noisy seismic dataset. The model is then updated based on this noisy dataset, and the process is repeated to improve the model's robustness and generalization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If machine-learned models are trained using synthetically generated seismic data, then sufficient training examples are available, but the models do not generalize well to real seismic datasets

Engineering Contradiction:

Improvequantity of training examplesVSAvoidgeneralization performance

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

The patent converts the harmful effect of noise in real seismic data into a beneficial training mechanism by generating adversarial noise profiles that deliberately reduce model performance, then using these noisy synthetic datasets to train more robust models that generalize better to real data

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The patent applies preliminary noise augmentation to synthetic training data before model training, creating adversarial examples in advance that prepare the model to handle noise and variations present in real seismic datasets, thereby improving generalization performance

Inventive Principle:

Principle #10Preliminary action

2Reliability

If noise is added to synthetic seismic datasets to improve robustness, then model generalization improves, but training complexity increases

Engineering Contradiction:

Improvemodel robustnessVSAvoidtraining process complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces manual noise generation and adjustment processes with an automated adversarial noise generation system that uses the machine-learned model itself to generate appropriate noise profiles, eliminating the need for manual intervention in noise characterization

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The machine-learned model generates its own training noise profiles by processing clean synthetic data through its own architecture and identifying perturbations that reduce its performance, creating a self-contained training pipeline that automatically adapts noise characteristics to the specific model being trained

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20250035802A1Methods and systems for improving generalization and performance of seismic machine-learned models through in-domain adversarial attacks

Publication Date: 2025.01.30 SAUDI ARABIAN OIL CO

AI summary

A method for performing a seismic processing task using a machine-learned model developed using an in-domain adversarial attacker. The method includes obtaining a machine-learned model parameterized by a set of weights and generating a synthetic seismic dataset and associated target. The method further includes determining a noise profile for the synthetic seismic dataset in a frequency domain that when added, in a spatial-temporal domain, to the synthetic seismic dataset reduces a performance of the machine-learned model. The method further includes adding the noise profile to the synthetic seismic dataset forming a noisy seismic dataset and updating the set of weights of the machine-learned model based on the noisy seismic dataset and the target. The method further includes receiving a seismic dataset corresponding to a subsurface, processing the seismic dataset with the machine-learned model to form a predicted target, and developing a geological model for the subsurface using the predicted target.