Discover the power of machine learning in the physical sciences with this one-stop resource from a leading voice in the field
Deep Learning for Physical Scientists: Accelerating Research with Machine Learning delivers an insightful analysis of the transformative techniques being used in deep learning within the physical sciences. The book offers readers the ability to understand, select, and apply the best deep learning techniques for their individual research problem and interpret the outcome.
Designed to teach researchers to think in useful new ways about how to achieve results in their research, the book provides scientists with new avenues to attack problems and avoid common pitfalls and problems. Practical case studies and problems are presented, giving readers an opportunity to put what they have learned into practice, with exemplar coding approaches provided to assist the reader.
From modelling basics to feed-forward networks, the book offers a broad cross-section of machine learning techniques to improve physical science research. Readers will also enjoy:
* A thorough introduction to the basic classification and regression with perceptrons
* An exploration of training algorithms, including back propagation and stochastic gradient descent and the parallelization of training
* An examination of multi-layer perceptrons for learning from descriptors and de-noising data
* Discussions of recurrent neural networks for learning from sequences and convolutional neural networks for learning from images
* A treatment of Bayesian optimization for tuning deep learning architectures
Perfect for academic and industrial research professionals in the physical sciences, Deep Learning for Physical Scientists: Accelerating Research with Machine Learning will also earn a place in the libraries of industrial researchers who have access to large amounts of data but have yet to learn the techniques to fully exploit that access.
Autorentext
Dr Edward O. Pyzer-Knapp is the worldwide lead for AI Enriched Modelling and Simulation at IBM Research. Previously, he obtained his PhD from the University of Cambridge using state of the art computational techniques to accelerate materials design then moving to Harvard where he was in charge of the day-to-day running of the Harvard Clean Energy Project - a collaboration with IBM which combined massive distributed computing, quantum-mechanical simulations, and machine-learning to accelerate discovery of the next generation of organic photovoltaic materials. He is also the Visiting Professor of Industrially Applied AI at the University of Liverpool, and the Editor in Chief for Applied AI Letters, a journal with a focus on real-world application and validation of AI.
Dr Matt Benatan received his PhD in Audio-Visual Speech Processing from the University of Leeds, after which he went on to pursue a career in AI research within industry. His work to date has involved the research and development of AI techniques for a broad variety of domains, from applications in audio processing through to materials discovery. His research interests include Computer Vision, Signal Processing, Bayesian Optimization, and Scalable Bayesian Inference.
Inhalt
Contents
About the Companion Website xi
1 Prefix Learning to Think Deep 1
1.1 So What Do I Mean by Changing the Way You Think? 2
2 Setting Up a Python Environment for Deep Learning Projects 5
2.1 Python Overview 5
2.2 Why Use Python for Data Science? 6
2.3 Anaconda Python 7
2.3.1 Why Use Anaconda? 7
2.3.2 Downloading and Installing Anaconda Python 7
2.3.2.1 Installing TensorFlow 9
2.4 Jupyter Notebooks 10
2.4.1 Why Use a Notebook? 10
2.4.2 Starting a Jupyter Notebook Server 11
2.4.3 Adding Markdown to Notebooks 12
2.4.4 A Simple Plotting Example 14
2.4.5 Summary 16
3 Modelling Basics 17
3.1 Introduction 17
3.2 Start Where You Mean to Go On Input Definition and
Creation 17
3.3 Loss Functions 18
3.3.1 Classification and Regression 19
3.3.2 Regression Loss Functions 19
3.3.2.1 Mean Absolute Error 19
3.3.2.2 Root Mean Squared Error 19
v
0005160659.3D 5 22/7/2021 7:19:56 AM
3.3.3 Classification Loss Functions 20
3.3.3.1 Precision 21
3.3.3.2 Recall 21
3.3.3.3 F1 Score 22
3.3.3.4 Confusion Matrix 22
3.3.3.5 (Area Under) Receiver Operator Curve (AU-ROC) 23
3.3.3.6 Cross Entropy 25
3.4 Overfitting and Underfitting 28
3.4.1 BiasVariance Trade-Off 29
3.5 Regularisation 31
3.5.1 Ridge Regression 31
3.5.2 LASSO Regularisation 33
3.5.3 Elastic Net 34
3.5.4 Bagging and Model Averaging 34
3.6 Evaluating a Model 35
3.6.1 Holdout Testing 35
3.6.2 Cross Validation 36
3.7 The Curse of Dimensionality 37
3.7.1 Normalising Inputs and Targets 37
3.8 Summary 39
3.8 Notes 39
4 Feedforward Networks and Multilayered Perceptrons 41
4.1 Introduction 41
4.2 The Single Perceptron 41
4.2.1 Training a Perceptron 41
4.2.2 Activation Functions 42
4.2.3 Back Propagation 43
4.2.3.1 Weight Initialisation 45
4.2.3.2 Learning Rate 46
4.2.4 Key Assumptions 46
4.2.5 Putting It All Together in TensorFlow 47
4.3 Moving to a Deep Network 49
4.4 Vanishing Gradients and Other Deep Problems 53
4.4.1 Gradient Clipping 54
4.4.2 Non-saturating Activation Functions 54
4.4.2.1 ReLU 54
4.4.2.2 Leaky ReLU 56
4.4.2.3 ELU 57
vi Contents
0005160659.3D 6 22/7/2021 7:19:56 AM
4.4.3 More Complex Initialisation Schemes 57
4.4.3.1 Xavier 58
4.4.3.2 He 58
4.4.4 Mini Batching 59
4.4.5 Batch Nor...