Since the emergence of Midjourney and other image generators, artists have been watching and wondering whether AI is a great opportunity or an existential threat. Now, after a list of 16,000 names emerged of artists whose work Midjourney had allegedly used to train its AI – including Bridget Riley, Damien Hirst, Rachel Whiteread, Tracey Emin, David Hockney and Anish Kapoor – the art world has issued a call to arms against the technologists. British artists have contacted US lawyers to discuss joining a class action against Midjourney and other AI firms, while others have told the Observer that they may bring their own legal action in the UK. "What we need to do is come together," said Tim Flach, president of the Association of Photographers and an internationally acclaimed photographer whose name is on the list. "This public showing of this list of names is a great catalyst for artists to come together and challenge it. I personally would be up for doing that."

Explainability techniques for data-driven predictive models based on artificial intelligence and machine learning algorithms allow us to better understand the operation of such systems and help to hold them accountable. New transparency approaches are developed at breakneck speed, enabling us to peek inside these black boxes and interpret their decisions. Many of these techniques are introduced as monolithic tools, giving the impression of one-size-fits-all and end-to-end algorithms with limited customisability. Nevertheless, such approaches are often composed of multiple interchangeable modules that need to be tuned to the problem at hand to produce meaningful explanations. This paper introduces a collection of hands-on training materials -- slides, video recordings and Jupyter Notebooks -- that provide guidance through the process of building and evaluating bespoke modular surrogate explainers for tabular data. These resources cover the three core building blocks of this technique: interpretable representation composition, data sampling and explanation generation.

This paper provides both an introduction to and a detailed overview of the principles and practice of classifier calibration. A well-calibrated classifier correctly quantifies the level of uncertainty or confidence associated with its instance-wise predictions. This is essential for critical applications, optimal decision making, cost-sensitive classification, and for some types of context change. Calibration research has a rich history which predates the birth of machine learning as an academic field by decades. However, a recent increase in the interest on calibration has led to new methods and the extension from binary to the multiclass setting. The space of options and issues to consider is large, and navigating it requires the right set of concepts and tools. We provide both introductory material and up-to-date technical details of the main concepts and methods, including proper scoring rules and other evaluation metrics, visualisation approaches, a comprehensive account of post-hoc calibration methods for binary and multiclass classification, and several advanced topics.

Interpretable representations are the backbone of many black-box explainers. They translate the low-level data representation necessary for good predictive performance into high-level human-intelligible concepts used to convey the explanation. Notably, the explanation type and its cognitive complexity are directly controlled by the interpretable representation, allowing to target a particular audience and use case. However, many explainers that rely on interpretable representations overlook their merit and fall back on default solutions, which may introduce implicit assumptions, thereby degrading the explanatory power of such techniques. To address this problem, we study properties of interpretable representations that encode presence and absence of human-comprehensible concepts. We show how they are operationalised for tabular, image and text data, discussing their strengths and weaknesses. Finally, we analyse their explanatory properties in the context of tabular data, where a linear model is used to quantify the importance of interpretable concepts.

In real world, three cohorts would approach Machine Learning differently - A. Programmers - "How" - interested in quickly learning the libraries, tips/tricks to scale algorithms with larger data sets B. Theorists - "What" - interested in choosing the right algorithm, design ensemble, selecting and extracting right features C. Fashionists - "Show" - in this category, some of the even basic reporting/analytics are not termed "Machine Learning", need enough buzzwords pieced together to repaint the old apps. Flach's book is a great source for those who are 75%-25% between first two, and perhaps even greater especially if your Linear Algebra (basics) is not too rusty. It gives a wide and somewhat deep tour of the landscape broken into four paradigms (Quantitative/Analytical, Logical, Geometric, Probabilitisic) and does a real good job on feature design. The book is interspersed with some key insights that are not to be found elsewhere (e.g., how the'pseudo-inverse' in OLS is really decorrelate-scale-normalize the distribution; Skew-Kurtosis are the statistical measure of "shape"; Naive Bayes is not only Naive but also not particularly Bayesian; How Laplacian Estimate generalizes into Pseudo-Counts and then to m-estimate etc.). After "deep reading" of the book over a month or so, I also went through Flach's detailed 500 slide presentation (check out his website) on this book.