Objectives:
Machine learning is now being used widely in several areas of science and engineering including Geosciences. It is also well recognized that for a successful application of ML, domain knowledge is necessary. At the same time, geoscientists must understand the fundamental concepts and limitations of different ML algorithms for meaningful applications. The course lectures will describe the basic concepts in a simple manner using many realistic examples and are aimed at junior/senior undergrad students and entry level graduate students. We will provide hands on exercises at each step to recognize the importance of the ML techniques. Computer software and data will be provided.
Prerequisites: Instructor’s consent. Adequate background in calculus is required. We will devote a lecture each on optimization and statistics.
Content
Lecture Plan
- L1 – Introduction:Course Introduction & Inference in Geosciences (In-verse Method, Geostatistics, NN)
- Lab 1: Introduction to python
- L2 – Introduction: Big Data Analytics and ML Overview
- Lab 2: Python libraries
- L3 -Optimization methods and Sampling: I (Gradient descent,Newton, SGD, minibatch SGD)
- Lab 3: Optimization practice
- L4 – Background Statistics (Descriptive and Inferential Statstics)
- Lab 4: Simple Descriptive and inferential statistics (well data/flower data/housingdata)
- L5 – Linear and Logistic regression
- Lab 5: Regeression practice
- L6 – Perceptrons and Neurons : A simple NN model
- Lab 6: building a simple NN model for classification (well log – faciesclassification)
- L7 -Nearest Neighbor, KNN
- Lab 7:NN, KNN (geoscience data)
- L8 – Decision Tree
- Lab 8: Decision tree
- L9 -Random forest
- Lab 9 Random forest (geoechemical data)
- L10 – Feature Engineering
- Lab 10 Feature Engineering
- L11 – OVER-FITTING. UNDERFITTING, VARIANCE, BIAS
- L12 – Dimensionality Reduction
- Lab 11 Dimensionality Reduction
- L13 – Naive Bayes
- Lab 12 – NB
- L14 – Deep Learning: Convolutional Neural Networks
- Lab 14 – CNN
- L15 – Auto-encoders
- Lab 15 – Auto-encoders
- L16 – Recurrent Neural Networks