Understanding Machine Learning

Abstract

Machine learning is one of the fastest growing areas of computer science, with far-reaching applications. The aim of this textbook is to introduce machine learning, and the algorithmic paradigms it offers, in a principled way. The book provides a theoretical account of the fundamentals underlying machine learning and the mathematical derivations that transform these principles into practical algorithms. Following a presentation of the basics, the book covers a wide array of central topics unaddressed by previous textbooks. These include a discussion of the computational complexity of learning and the concepts of convexity and stability; important algorithmic paradigms including stochastic gradient descent, neural networks, and structured output learning; and emerging theoretical concepts such as the PAC-Bayes approach and compression-based bounds. Designed for advanced undergraduates or beginning graduates, the text makes the fundamentals and algorithms of machine learning accessible to students and non-expert readers in statistics, computer science, mathematics and engineering.

Augmenting adaptive immunity: progress and challenges in the quantitative engineering and analysis of adaptive immune receptor repertoires

“…Specifically, the disease signal in the form of sequence (motifs) likely represents a high-dimensional mixture of different motifs. Machine learning approaches to disentangle and recover these high-dimensional signals are as of yet missing 94 . Nevertheless, progress has been made in terms of repertoire-based diagnostics in mouse and humans with both model antigens and virus infections leveraging both sequence motifs and entire sequences 25,82,[95][96][97][98] .…”

“…As of yet, it remains unclear as to exactly why deep learning is so successful in classification, prediction and generation. Ongoing research suggests that deep neural networks are especially adept at amplifying class-specific signal and ignoring class-unspecific noise 94 , however, mechanistic details on what to amplify or ignore are obscured by the complexity of their architecture akin to a black box. Efforts to deconvolute deep and very deep networks (turning the black box into a glass box) have gained momentum in recent years, tools and techniques such as Integrated Gradients, LIME, and human-in-the-loop [418][419][420] are all geared towards attributing the prediction of a deep network back to its inputs.…”

Gradient-based optimal control of open quantum systems using quantum trajectories and automatic differentiation

“…A better way for our model is to include a moderate number of trajectories in each iteration so that every iteration is based on a batch of data. This procedure, where gradients applied in each iteration are generated by a batch of trajectories, is known as mini-batch SGD [50]. Using quantum trajectories allows for flexibility in the ways simulations could be implemented since trajectories could be constructed independently of each other.…”

Proposing Enhanced Feature Engineering and a Selection Model for Machine Learning Processes

“…Discriminant analysis are also used in FE. Hence, in the line with the latest progress and related study (See Section 2), the work proposed in this paper uses ML and mathematical techniques, such as statistical pattern classification [7], Orthonormalization [8], Probability theory [9], Jacobian [7], Laplacian [3], and Lagrangian distribution [10] to build the mathematical constructs and underlying algorithms (1 and 2). To advance such developments, a unique engineering of the features is proposed where the classifier learns to group an optimum set of features without consuming excessive computing power, regardless of the anatomy of the underlying datasets and predictive goals.…”