MATLAB and Colab labs associated with the Machine Learning in Geoscience book.

User contribution labs are welcome if they are they are vetted and accepted by the authors. Your name will be noted in the lab. Please send inquiries about submitting a lab (state title, purpose, and benefit) to Tushar Gautam (tushar.gautam@sci.utah.edu). If mistakes are found in a lab please send the corrections to Tushar as well.

Chapter 2: Data Prediction by Least Squares Inversion

• Perform the first 4 least squares exercises
• Learn about Colab
• Learn how to download your own data into Colab

Chapter 4: Introduction to Neural Networks

• Binary classification lab
• The gradient validation of the NN binary classification code
• The multiclass classification lab

Chapter 5: Processing Geophysical Data with Fully Connected Neural Networks

• Computes the decision boundaries that separate signal from noise in migration images, and then mutes the noise in the migration image in the (x,z) domain. The performance of the SVM is also compared to the neural network and logistic regression algorithms.
• Compute the window in the k-ω domain that isolates the fundamental mode Rayleigh wave from the higher-order modes

Chapter 6: Multilayer Fully Connected Neural Networks

• Fully-Connected Neural Network lab
• A fully connected neural network to classify input vectors that have 1s and 0s for its elements

Chapter 7: Support Vector Machines

• Classifying a small set of 2 x 1 vectors.
• Soft-margin SVM lab
• Computes the decision boundaries that separate signal from noise in migration images, and then mutes the noise in the migration image in the (x,z) domain. The performance of the SVM is also compared to the neural network and logistic regression algorithms.
• Compare the clustering effectiveness on two clouds of noisy points, each one hugging one of the two neighboring cosine functions.

Chapter 9: Convolutional Neural Networks

• Download and run the Alexnet rock crack lab.
• Download and run the Alexnet salt body picking lab
• Download and run the U-Net bird identification lab

Chapter 10: Object Identification by CNN

• Gimp labeling lab.
• Letter Classification by Convolutional Neural Network
• Salt Detection with U-net Lab
• Detect faults in synthetic sections (LINK DOES NOT WORK)
• R-CNN identification of cell types in blood samples
• YOLO5 identification of microfossils in microscope images
• Andrew Ng's course on convolutional deep learning
• Identify types of birds from pictures.
• Crack detection with U-Net and transfer learning

Chapter 11: Semantic Segmentation of Cracks in Photos

• Crack Labeling lab
• Train a U-Net network for crack detection and trabsfer learning.

Chapter 12: Recurrent Neural Networks

• Train the RNN to function as a character-level language model
• Execute the text classification lab
• Use the RNN codes and adjust it to validate the re- sults of Zhi et al. (2021) on synthetic data.
• Read He et al. (2022) and Luppichini et al. (2023). Adjust an RNN lab and apply it to synthetic data for one of the geochemistry problems described in He et al. (2022) or the river flow problems described in Luppichini et al. (2023).
• Blind Deconvolution RNN lab

Chapter 15: Convolutional Sparse Coding

• Sparse coding lab

Chapter 16: Principal Component Analysis

• Basic PCA lab
• Importance of normalizing your data
• K-means and PCA clustering lab
• Evaluate the points in a cluster with the Silohuette lab (LINK DOES NOT WORK)
• Carbonate PCA lab
• Kernel PCA lab at Appendix 16.10
• Adjust the codes so that a noisy section is input into the RPCA code rather than the picture of an elevator in Figure 16.22a. Adjust the random noise level and explain the results.
• Truncated SVD codes
• Turkana Geochem lab

Chapter 17: Clustering Algorithms

• Insert the Yellowstone data as the input in this lab. Compare its performance against that of k-means clustering.
• Semblance NMO and clustering program
• Same as previous exercise, except adjust the clustering procedure to incorporate a robust regularization constraint that is updated with each new CMG. The adaptive constraint is the a priori semblance curve that is the average of a select number of previously computed curves.
• Same as the first computational exercise, except incorporate new features in finding the best semblance curve. Use some relevant textures provided in this link as new features.
• Break DBSCAN
• Arrival time picking by fuzzy clustering
• Evaluate the quality of a cluster using the Silohuette analysis

Chapter 18: Generative Adversarial Networks

• GAN tutorial (TUTORAL DEPRECATED IN AUGUST 2024 so WILL NOT WORK)
• Generate images with a GAN using the tutorial
• Use the images in the AlUla data to generate new AlUla images with the GAN colab software
• Same as previous problem except the input AlUla images should have labeled cracks. Are the generated images realistic enough to augment the actual AlUla data where the cracks are labeled?

Chapter 19: Sampling a Probability Distribution

• Test the different sampling algorithms
• Test Metropolis-Hastings MCMC

Chapter 21: Discriminant Analysis and Gaussian Mixture Model

• Compute the GMM contours for Yellowstone data
• Write the python code for the expectation-maximation algorithm in Box 21.5.2. and compare your results to the k-means results shown in Figure 17.1.

Chapter 26: Automatic Differentiation

• Explore ways to compute gradients with Tensorflow.
• Python codes for automatic differentiation (LINK DOES NOT WORK).