This chapter focuses on continuous-time MCMC algorithms, particularly those based on piecewise deterministic Markov processes (PDMPs). It introduces PDMPs as a scalable alternative to traditional MCMC, with a detailed explanation of their simulation, invariant distribution, and limiting processes. Various continuous-time samplers, including the bouncy particle sampler and zig-zag process, are compared in terms of efficiency and performance. The chapter also addresses practical aspects of simulating PDMPs, including techniques for exploiting model sparsity and data subsampling. Extensions to these methods, such as handling discontinuous target distributions or distributions defined on spaces of different dimensions, are discussed.

Learning for nondeterministic models can take advantage of most of the techniques developed for probabilistic models (Chapter 10). Indeed, note that in reinforcement learning (RL), probabilities of action transitions are not needed, so RL techniques can be applied to nondeterministic models too. For instance, we can use the algorithms for Q-learning, parametric Q-learning, and deep Q-learning. However, these algorithms do not give explicit description models of actions. In this chapter, we therefore discuss some intuitions and also some challenges of how the techniques for learning deterministic action specifications could be extended to deal with nondeterministic models. Note, however, that learning lifted action schemas in nondeterministic models is still an open problem.

Temporal models are quite rich, allowing concurrency and temporal constraints to be handled. But the development of the temporal models is a bottleneck, to be eased with machine learning techniques. In this chapter, we first briefly address the problem of learning heuristics for temporal planning (Section 19.1). We then consider the issue of learning durative action schema and temporal methods (Section 19.2). The chapter outlines the proposed approaches, based on techniques seen earlier in the book, without getting into detailed descriptions of the corresponding procedures.

This chapter addresses the issues of acting with temporal models . It presents methods for handling dynamic controllability (Section 18.1), dispatching (Section 18.2), and execution and refinement of a temporal plan (Section 18.3). It proposes methods for acting with a reactive temporal refinement engine (Section 18.4), planning with Monte Carlo rollouts (Section 18.5), and integrating planning and acting (Section 18.6).

In this chapter we introduce different representations and techniques for acting with nondeterministic models: nondeterministic state transition systems (Section 11.1), automata (Section 11.2), behavior trees (Section 11.3), and Petri nets (Section 11.4).

The development of more sophisticated and, especially, approximate sampling algorithms aimed at improving scalability in one or more of the senses already discussed in this book raises important considerations about how a suitable algorithm should be selected for a given task, how its tuning parameters should be determined, and how its convergence should be as- sessed. This chapter presents recent solutions to the above problems, whose starting point is to derive explicit upper bounds on an appropriate distance between the posterior and the approximation produced by MCMC. Further, we explain how these same tools can be adapted to provide powerful post-processing methods that can be used retrospectively to improve approximations produced using scalable MCMC.

In the past, techniques for natural language translation were not very relevant for acting and planning systems. However, with the recent advent of large language models and their various multimodal extensions into foundation models, this is no longer the case. This last part introduces large language models and their potential benefits in acting, planning, and learning. It discusses the perceiving, monitoring, and goal reasoning functions for deliberation.

Learning to act with probabilistic models is the area of reinforcement learning (RL), the topic of this chapter. RL in some ways parallels the adaptation mechanisms of natural beings to their environment, relying on feedback mechanisms and extending the homeostasis regulations to complex behaviors. With continual learning, an actor can cope with a continually changing environment.This chapter first introduces the main principles of reinforcement learning. It presents a simple Q-learning RL algorithm. It shows how to generalize a learned relation with a parametric representation. it introduces neural network methods, which play a major in learning and are needed for deep RL (Section 10.5) and policy-based RL (Section 10.6). The issues of aided reinforcement learning with shaped rewards, imitation learning, and inverse reinforcement learning are addressed next. Section 10.8 is about probabilistic planning and RL.

This chapter explores the benefits of non-reversible MCMC algorithms in improving sampling efficiency. Revisiting Hamiltonian Monte Carlo (HMC), the chapter discusses the advantages of breaking detailed balance and introduces lifting schemes as a tool to enhance exploration of the parameter space. It reviews non-reversible HMC and alternative algorithms like Gustafson’s method. The chapter also covers techniques like delayed rejection and the discrete bouncy particle sampler, offering a comparison between reversible and non-reversible methods. Theoretical insights and practical implementations are provided to highlight the efficiency gains from non-reversibility.

Chapter 2 examines how the use of “quantified self” as a shorthand for personal data necessarily indexes only one end, rather than the full spectrum, of technologists’ understanding of digitization and their own roles within it. Looking closely at the way digital executives talk about data in forums such as QS, among others, in fact reveals the contradictions, professional obfuscations, and hyperbole that continue to shape the self-tracking sector. Digital professionals may occasionally enfold concepts such as the "quantified self” into promotional “pitch theater” to stage self-monitoring devices as gadgets that produce faithful and objective data. My interactions with practitioners in these settings, however, point to the more varied social, legal, and fiscal advantages professionals reap from representing digital self-tracking and the data these devices produce as both plastic and precise. This chapter argues that the surface impression that technologists relate to data and modes of self-monitoring in reductive terms has to be weighed against ways executives pursue both digital ambiguity and objectivity as a meaningful corporate strategy.

To begin evaluating the interaction of “quantified self,” the concept, and Quantified Self (QS), the collective, with digital entrepreneurialism, it’s necessary to understand the influence of its originators, Kevin Kelly and Gary Wolf, on this construct’s form and function. Chapter 1 reviews how the two authors have coined the term and established the group as an expression of what Wolf has called the “culture of personal data” (Wolf, 2009). While the founders defer to the explanatory power of culture in situating the collective within the technological imaginary, this chapter examines how their own personal backgrounds as journalists and Wired magazine editors have shaped the semantic meaning of “quantified self” as a catchphrase that refers to the means and outputs of digital self-tracking and especially to QS as a community of technophiles. Although the role the forum has come to play within the commercial self-tracking sphere analyzed in this book does not fully align with its originators’ intentions, the framing they established has set the tone for many of the ways the collective has become socialized in the technological arena as well as how it has come to work within it.

This part of the book is devoted to acting, planning, and learning with operational models of actions expressed with a hierarchical task-oriented representation. Operational models are valuable for acting. They allow for detailed descriptions of complex actions handling dynamic environments with exogenous events. The representation relies on hierarchical refinement methods that describe alternative ways to handle tasks and react to events. A method can be any complex algorithm, decomposing a task into subtasks and primitive actions. Subtasks are refined recursively. Actions trigger the execution of sensory-motor procedures in closed loops that query and change the world stochastically.