avatar

Ahmad Mustafa, PhD

Machine Learning and Analytics Engineer
Occidental Petroleum
ahmadmustafa(dot)am(at)gmail.com


Short Bio

I am currently a machine learning and analytics engineer working at Occidental Petroleum based in Houston, Texas. My work involves using cutting edge machine- and deep learning techniques to solve problems in subsurface understanding for hydrocarbon exploration, environment monitoring, and sustainability applications.

I hold a PhD degree in Electrical and Computer Engineering at the Georgia Institute of Technology. My dissertation was completed under the supervision of Professor Ghassan AlRegib at the OLIVES lab. My research has focused on the automatic analysis and interpretation of high-dimensional spatiotemporal image signals, such as those present in migrated 3D seismic volumes (for geophysics exploration) and medical imaging data (for predictive analytics in healthcare). In particular, I have worked on developing an array of deep learning algorithms for solving multimodal inverse problems in geophysics, for the accurate segmentation of spatiotemporal image data towards subsurface characterization and automatic tumour detection, and for addressing domain shift for the real world application of deep models pretrained on synthetic training data. Additionally, I have worked on researching weakly supervised learning and active learning for the accurate semantic segmentation of 2D/3D image signals in sparse label settings.

Aside from my research engagments, I have delivered numerous talks and short courses across industry and academic settings and have been involved with numerous research intiatives involving industry and academia. I have served as a mentor and advisor to students and early-career researchers, fostering knowledge exchange through technical workshops and industry partnerships. I welcome collaborations that push the boundaries of AI-driven signal processing, particularly in domains where high-dimensional data and complex physics-based constraints pose unique challenges.

News

Journal Publications [Google Scholar]

  1. IEEE TGRS
    Ahmad Mustafa, Reza Rastegar, Tim Brown, Gregory Nunes, Daniel De Lilla, and Ghassan AlRegib
    IEEE Transactions of Geoscience and Remote Sensing, vol. 62, pp. 1-12, 2024.

  2. Geophysics
    Prithwijit Chowdhury, Ahmad Mustafa, Mohit Prabhushankar, and Ghassan AlRegib
    Geophysics, vol. 90, issue 3 (2025)

  3. Geophysics
    Ahmad Mustafa, Klaas Koster, and Ghassan AlRegib
    Geophysics, vol. 89, no. 1 (2023), Pg. WA13-WA24

  4. Geophysics
    Ahmad Mustafa, and Ghassan AlRegib
    Geophysics, vol. 88, no. 4 (2023), Pg. IM77-IM86

  5. Geophysics
    Ahmad Mustafa, Motaz Alfarraj, and Ghassan AlRegib
    Geophysics, vol. 86, no. 4 (2021), Pg. O37-O48

Conference Publications [Google Scholar]

  1. arXiv
    Mohit Prabhushankar, Kiran Premdat Kokilepersaud, Jorge Quesada, Yavuz Yarici, Chen Zhou, Mohammad Alotaibi, Ghassan AlRegib, Ahmad Mustafa, Yusufjon Kumakov
    arXiv Preprint, https://arxiv.org/abs/2408.11185, 2024.

  2. IMAGE
    Prithwijit Chowdhury, Ahmad Mustafa, Mohit Prabhushankar, and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2023.

  3. ICIP
    Yash-yee Logan, Ryan Benkert, Ahmad Mustafa, and Ghassan AlRegib
    2022 IEEE International Conference on Image Processing (ICIP).

  4. SEG
    Ahmad Mustafa and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2022.

  5. SEG
    Ahmad Mustafa and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2021.

  6. ICIP
    Ahmad Mustafa and Ghassan AlRegib
    2021 IEEE International Conference on Image Processing (ICIP).

  7. SEG
    Ahmad Mustafa and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2020.

  8. SEG
    Ahmad Mustafa and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2020.

  9. SEG
    Ahmad Mustafa, Motaz Alfarraj, and Ghassan AlRegib
    SEG International Exposition and Annual Meeting, 2019.

Invited Talks

Show More Text

Research

1. Efficient Training Sample Selection for Seismic Interpretation using Active Learning

Machine learning-assisted seismic interpretation tasks require large quantities of labeled data annotated by expert interpreters, which is a costly and time-consuming process. Where existing works to minimize dependence on labeled data assume the data annotation process to already be completed, active learning—a field of machine learning—works by selecting the most important training samples for the interpreter to annotate in real time simultaneously with the training of the interpretation model itself, thereby reducing cost and effort to produce annotated training samples while minimizing the negative impact on performance. We develop a unique and first-of-a-kind active learning framework for seismic facies interpretation using the manifold learning properties of deep autoencoders.

Show more

By jointly learning representations for supervised and unsupervised tasks and then ranking unlabeled samples by their nearness to the data manifold, we are able to identify the most relevant training samples to be labeled by the interpreter in each training round. This is shown in the figure below. On the popular F3 dataset, we obtain close to 10 percentage point difference in terms of interpretation accuracy between the proposed method and the baseline with only three fully annotated seismic sections.

Results

Performance of the facies segmentation model is recorded in terms of the mean intersection-over-union (mIOU) at the end of every cyle and plotted for the proposed sampling method as well as other sampling techniques. This includes the well known active learning method based on entropy as well as non-active learning techniques commonly used by seismic interpreters. These plots for both training and test splits are given below.
Alt Text for Image 1
Performance curves for the proposed and baseline sampling techniques on the training split.
Alt Text for Image 1
Performance curves for the proposed and baseline sampling techniques on the test split.

Codes

Codes related to this project can be found at this GitHub repository. Detailed instructions regarding the running of the codes and installing related dependencies are provided.

Related Publications

More information on this work can be found in the following publidations:
1. Mustafa, Ahmad, and Ghassan AlRegib. "Man-recon: Manifold learning for reconstruction with deep autoencoder for smart seismic interpretation." In 2021 IEEE International Conference on Image Processing (ICIP), pp. 2953-2957. IEEE, 2021. [Link] [PDF]
2. Mustafa, Ahmad, and Ghassan AlRegib. "Active learning with deep autoencoders for seismic facies interpretation." Geophysics 88, no. 4 (2023): IM77-IM86. [Link] [PDF]

2. Learning under Sparse, Missing Labels for Structure Interpretation via Visual Attention Modeling

Accurate interpretation of visual data for relevant information forms an important component of many real-world applications such as medical disease diagnosis, geological hazard assessment, hydrocarbon exploration, etc. Producing fine-grained annotations on images is an expensive, laborious, and time-consuming process. The human brain is wired to selectively focus its attention on certain aspects of the visual scene. This perception mechanism is driven both by low-level signal cues, such as changes in color, contrast, intensity, shapes etc., as well as high-level cognitive factors such as one’s prior knowledge, goals, expectations, and constraints with respect to the task at hand.

Show more

These attentional factors, referred to as bottom-up and top-down attention respectively, play an important role in determining the final annotations that get produced for a given visual scene, often at the cost of leaving out a lot of visual information the brain deems to be unimportant with regard to the task of interest. Mapping geological faults on 3D seismic volumes is one such application where human attention selectivity results in highly incomplete fault annotations. Conventional supervised learning methods treat regions of missed fault labels as negatives, resulting in non-optimal learning for the machine learning model. We propose a method to model visual attention and incorporate it into data sampling and model training procedures. We demonstrate the utility of this approach for mapping faults on seismic volumes using pretrained 3D convolutional neural networks (CNNs). Using an annotated seismic dataset from NW Australia, we show quantitatively and qualitatively that modeling visual attention leads to significant performance gains even with limited, incompletely labeled seismic training data.

Results

Fault interpretation is carried out using a model trained in the proposed manner (i.e., attention modeling) and also using models trained without modeling visual attention. The proposed approach can be seen to outperform baseline strategies even when trained on data with highly incomplete labels.
Alt Text for Image 1
Performance curves for the proposed and baseline training startegies on a seismic section from the test set.
Alt Text for Image 1
Performance curves for the proposed and baseline training startegies on a seismic section from the test set.

Codes

Codes related to this project can be found at this GitHub repository. Detailed instructions regarding the running of the codes and installing related dependencies are provided.

Related Publications

More information regarding the work can be accessed at the publications below:
1. A. Mustafa, R. Rastegar, T. Brown, G. Nunes, D. Delilla and G. Alregib, "Visual Attention-Guided Learning With Incomplete Labels for Seismic Fault Interpretation," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-12, 2024. [Full Text]
Show More Text

Software and Codes

1. amustafa9/SeisWiz

The ultimate lightweight Matplotlib-based seismic volume viewer with multi-view support and horizon visualization capabilities.

Teaching

Graduate Teaching Assistantships

Short Courses/Tutorials

Services

Organization Committee

Conference Reviewers

Journal Reviewers

Contact

You are more than welcome to reach out to me with for any questions or advice (if you feel I am qualified to give some based off my experience) on any of my emails below.
Personal Email: ahmadmustafa.am (at) gmail.com
Work Email: ahmad_mustafa (at) oxy.com