Center for Signal and Information Processing (CSIP) Seminar Series presents: Mitigating Overfitting in Machine Learning for Reservoir Characterization with 2-D Sequence Models and Joint Learning - Ahmad Mustafa
Date: Friday, April 9, 2021
Bluejeans link: https://bluejeans.com/621436705
Speaker Bio: I am Ahmad, a PhD student in ECE. My research interests lie in solving data-constrained problems in Deep Learning and Signal Processing. In particular, I work on applying novel DL concepts to various application areas in computational seismic interpretation.
Abstract: Seismic inversion plays a very useful role in detailed stratigraphic interpretation of migrated seismic volumes by enabling the estimation of reservoir properties over the complete volume. Traditional and machine learning-based seismic inversion workflows are limited to inverting each seismic trace independently of other traces to estimate impedance profiles, leading to lateral discontinuities in the presence of noise and large geological variations in the seismic data. In addition, machine learning-based approaches suffer the problem of overfitting if there is only a small number of wells on which the model is trained. We propose a two-pronged strategy to overcome these problems. We present a temporal convolutional network that models seismic traces temporally. We further inject spatial context for each trace into its estimations of the impedance profile. To counter the problem of limited labeled data, we also present a joint learning scheme whereby multiple datasets are simultaneously used for training, sharing beneficial information among each other. This results in the improvement in generalization performance on all datasets. We present a case study of acoustic impedance inversion using the open-source SEAM and Marmousi 2 datasets. Our evaluations show that our proposed approach is able to estimate impedance in the presence of noisy seismic data and a limited number of well logs with greater robustness and spatial consistency.