Daniele Magazzeni

• King's College London

We introduce a novel approach for detecting distribution shifts that negatively impact the performance of machine learning models in continuous production environments, which requires no access to ground truth data labels. It builds upon the work of Podkopaev and Ramdas [2022], who address scenarios where labels are available for tracking model err...

In this work we consider a new interpretation of fairness in decision making problems. Building upon existing fairness formulations, we focus on how to reason over fairness from a temporal perspective, taking into account the fairness of a history of past decisions. After introducing the concept of temporal fairness, we propose three approaches tha...

Polyhedral geometry can be used to shed light on the behaviour of piecewise linear neural networks, such as ReLU-based architectures. Counterfactual explanations are a popular class of methods for examining model behaviour by comparing a query to the closest point with a different label, subject to constraints. We present a new algorithm, Polyhedra...

In this work we consider a new interpretation of fairness in decision making problems. Building upon existing fairness formulations, we focus on how to reason over fairness from a temporal perspective, taking into account the fairness of a history of past decisions. After introducing the concept of temporal fairness, we propose three approaches tha...

The demand for open and trustworthy AI models points towards widespread publishing of model weights. Consumers of these model weights must be able to act accordingly with the information provided. That said, one of the simplest AI classification models, Logistic Regression (LR), has an unwieldy interpretation of its model weights, with greater diff...

Organizations around the world schedule jobs (programs) regularly to perform various tasks dictated by their end users. With the major movement toward using a cloud computing infrastructure, our organization follows a hybrid approach with both cloud and on-prem servers. The objective of this work is to perform capacity planning, i.e., estimate reso...

The demand for open and trustworthy AI models points towards widespread publishing of model weights. Consumers of these model weights must be able to act accordingly with the information provided. That said, one of the simplest AI classification models, Logistic Regression (LR), has an unwieldy interpretation of its model weights, with greater diff...

This paper proposes Progressive Inference - a framework to compute input attributions to explain the predictions of decoder-only sequence classification models. Our work is based on the insight that the classification head of a decoder-only Transformer model can be used to make intermediate predictions by evaluating them at different points in the...

We introduce T-CREx, a novel model-agnostic method for local and global counterfactual explanation (CE), which summarises recourse options for both individuals and groups in the form of human-readable rules. It leverages tree-based surrogate models to learn the counterfactual rules, alongside 'metarules' denoting their regions of optimality, provid...

The SHAP framework provides a principled method to explain the predictions of a model by computing feature importance. Motivated by applications in finance, we introduce the Top-k Identification Problem (TkIP) (and its ordered variant TkIP- O), where the objective is to identify the subset (or ordered subset for TkIP-O) of k features corresponding...

Discrete optimization belongs to the set of N P-hard problems, spanning fields such as mixed-integer programming and combinatorial optimization. A current standard approach to solving convex discrete optimization problems is the use of cutting-plane algorithms, which reach optimal solutions by iteratively adding inequalities known as cuts to refine...

There is an emerging interest in generating robust algorithmic recourse that would remain valid if the model is updated or changed even slightly. Towards finding robust algorithmic recourse (or counterfactual explanations), existing literature often assumes that the original model m and the new model M are bounded in the parameter space, i.e.,...

In human-aware planning systems, a planning agent might need to explain its plan to a human user when that plan appears to be non-feasible or sub-optimal. A popular approach, called model reconciliation, has been proposed as a way to bring the model of the human user closer to the agent's model. In this paper, we approach the model reconciliation p...

Stochastic optimization (SO) attempts to offer optimal decisions in the presence of uncertainty. Often, the classical formulation of these problems becomes intractable due to (a) the number of scenarios required to capture the uncertainty and (b) the discrete nature of real-world planning problems. To overcome these tractability issues, practitione...

Explainable Artificial Intelligence (XAI) has received widespread interest in recent years, and two of the most popular types of explanations are feature attributions, and counterfactual explanations. These classes of approaches have been largely studied independently and the few attempts at reconciling them have been primarily empirical. This work...

The SHAP framework provides a principled method to explain the predictions of a model by computing feature importance. Motivated by applications in finance, we introduce the Top-k Identification Problem (TkIP), where the objective is to identify the k features with the highest SHAP values. While any method to compute SHAP values with uncertainty es...

Counterfactual explanations have been widely studied in explainability, with a range of application dependent methods prominent in fairness, recourse and model understanding. The major shortcoming associated with these methods, however, is their inability to provide explanations beyond the local or instance-level. While many works touch upon the no...

There is an emerging interest in generating robust counterfactual explanations that would remain valid if the model is updated or changed even slightly. Towards finding robust counterfactuals, existing literature often assumes that the original model $m$ and the new model $M$ are bounded in the parameter space, i.e., $\|\text{Params}(M){-}\text{Par...

Counterfactual explanations utilize feature perturbations to analyze the outcome of an original decision and recommend an actionable recourse. We argue that it is beneficial to provide several alternative explanations rather than a single point solution and propose a probabilistic paradigm to estimate a diverse set of counterfactuals. Specifically,...

The Concept Bottleneck Models (CBMs) of Koh et al. [2020] provide a means to ensure that a neural network based classifier bases its predictions solely on human understandable concepts. The concept labels, or rationales as we refer to them, are learned by the concept labeling component of the CBM. Another component learns to predict the target clas...

Concept bottleneck models perform classification by first predicting which of a list of human provided concepts are true about a datapoint. Then a downstream model uses these predicted concept labels to predict the target label. The predicted concepts act as a rationale for the target prediction. Model trust issues emerge in this paradigm when soft...

Feature importance techniques have enjoyed widespread attention in the explainable AI literature as a means of determining how trained machine learning models make their predictions. We consider Shapley value based approaches to feature importance, applied in the context of time series data. We present closed form solutions for the SHAP values of a...

Counterfactual explanations inform ways to achieve a desired outcome from a machine learning model. However, such explanations are not robust to certain real-world changes in the underlying model (e.g., retraining the model, changing hyperparameters, etc.), questioning their reliability in several applications, e.g., credit lending. In this work, w...

Counterfactual explanations inform ways to achieve a desired outcome from a machine learning model. However, such explanations are not robust to certain real-world changes in the underlying model (e.g., retraining the model, changing hyperparameters, etc.), questioning their reliability in several applications, e.g., credit lending. In this work, w...

We consider a novel queuing problem where the decision-maker must choose to accept or reject randomly arriving tasks into a no buffer queue which are processed by N identical servers. Each task has a price, which is a positive real number, and a class. Each class of task has a different price distribution, service rate, and arrives according to an...

Scheduling is the task of assigning a set of scarce resources distributed over time to a set of agents, who typically have preferences over the assignments they would like to get. Due to the constrained nature of these problems, satisfying all agents' preferences often turns infeasible, which might lead to some agents not being happy with the resul...

Counterfactual explanations have been widely studied in explainability, with a range of application dependent methods emerging in fairness, recourse and model understanding. However, the major shortcoming associated with these methods is their inability to provide explanations beyond the local or instance-level. While some works touch upon the noti...

Scheduling is the task of assigning a set of scarce resources distributed over time to a set of agents, who typically have preferences about the assignments they would like to get. Due to the constrained nature of these problems, satisfying all agents' preferences is often infeasible, which might lead to some agents not being happy with the resulti...

We consider a novel queuing problem where the decision-maker must choose to accept or reject randomly arriving tasks into a no buffer queue which are processed by $N$ identical servers. Each task has a price, which is a positive real number, and a class. Each class of task has a different price distribution and service rate, and arrives according t...

There exist several methods that aim to address the crucial task of understanding the behaviour of AI/ML models. Arguably, the most popular among them are local explanations that focus on investigating model behaviour for individual instances. Several methods have been proposed for local analysis, but relatively lesser effort has gone into understa...

Feature attributions are a common paradigm for model explanations due to their simplicity in assigning a single numeric score for each input feature to a model. In the actionable recourse setting, wherein the goal of the explanations is to improve outcomes for model consumers, it is often unclear how feature attributions should be correctly used. W...

In automated planning, the need for explanations arises when there is a mismatch between a proposed plan and the user’s expectation. We frame Explainable AI Planning as an iterative plan exploration process, in which the user asks a succession of contrastive questions that lead to the generation and solution of hypothetical planning problems that a...

This paper presents an end-to-end system for fact extraction and verification using textual and tabular evidence, the performance of which we demonstrate on the FEVEROUS dataset. We experiment with both a multi-task learning paradigm to jointly train a graph attention network for both the task of evidence extraction and veracity prediction, as well...

We present a new method for counterfactual explanations (CFEs) based on Bayesian optimisation that applies to both classification and regression models. Our method is a globally convergent search algorithm with support for arbitrary regression models and constraints like feature sparsity and actionable recourse, and furthermore can answer multiple...

Planning enables intelligent agents, such as robots, to act so as to achieve their long term goals. To make the planning process tractable, a relatively low fidelity model of the world is often used, which sometimes leads to the need to replan. The typical view of replanning is that the robot is given the current state, the goal, and possibly some...

Planning as model checking based on source-to-source compilations has found increasing attention. Previously proposed approaches for temporal and hybrid planning are based on static translations, in the sense that the resulting model checking problems are uniquely defined by the given input planning problems. As a drawback, the translations can bec...

The PDDL+ language has been mainly devised to allow modelling of real-world systems, with continuous, time-dependant dynamics. Several interesting case studies with these characteristics have been also proposed, to test the language expressiveness and the capabilities of the support tools. However, most of these case studies have not been completel...

Path planners are important components of various products from video games to robotics, but their output can be counter-intuitive due to problem complexity. As a step towards improving the understanding of path plans by various users, here we propose methods that generate explanations for the optimality of paths. Given the question "why is path A...

In automated planning, the need for explanations arises when there is a mismatch between a proposed plan and the user's expectation. We frame Explainable AI Planning in the context of the plan negotiation problem, in which a succession of hypothetical planning problems are generated and solved. The object of the negotiation is for the user to under...

We present a novel approach for decreasing state uncertainty in planning prior to solving the planning problem. This is done by making predictions about the state based on currently known information, using machine learning techniques. For domains where uncertainty is high, we define an active learning process for identifying which information, onc...

In many usage scenarios of AI Planning technology, users will want not just a plan π but an explanation of the space of possible plans, justifying π. In particular, in oversubscription planning where not all goals can be achieved, users may ask why a conjunction A of goals is not achieved by π. We propose to answer this kind of question with the go...

It is widely known that AI planning and model checking are closely related. Compilations have been devised between various pairs of language fragments. What has barely been voiced yet, though, is the idea to let go of one's own modeling language, and use one from the other area instead. We advocate that idea here – to use automata-network languages...

In order to ensure the robust actuation of a plan, execution must be adaptable to unexpected situations in the world and to exogenous events. This is critical in domains in which committing to a wrong ordering of actions can cause the plan failure, even when all the actions succeed. We propose an approach to the execution of a task plan that permit...

Planning for hybrid systems is important for dealing with real-world applications, and PDDL+ supports this representation of domains with mixed discrete and continuous dynamics. In this paper we present a new approach for planning for hybrid systems, based on encoding the planning problem as a Satisfiability Modulo Theories (SMT) formula. This is t...

One of the major limitations for the employment of model-based planning and scheduling in practical applications is the need of costly re-planning when an incongruence between the observed reality and the formal model is encountered during execution. Robustness Envelopes characterize the set of possible contingencies that a plan is able to address...

Model-based approaches to AI are well suited to explainability in principle, given the explicit nature of their world knowledge and of the reasoning performed to take decisions. AI Planning in particular is relevant in this context as a generic approach to action-decision problems. Indeed, explainable AI Planning (XAIP) has received interest since...

Explainable AI is an important area of research within which Explainable Planning is an emerging topic. In this paper, we argue that Explainable Planning can be designed as a service -- that is, as a wrapper around an existing planning system that utilises the existing planner to assist in answering contrastive questions. We introduce a prototype f...

To achieve practical execution, planners must produce temporal plans with some degree of run-time adaptability. Such plans can be expressed as Simple Temporal Networks (STN), that constrain the timing of action activations, and implicitly represent the space of choices for the plan executor.A first problem is to verify that all the executor choices...

Probabilistic planning is very useful for handling uncertainty in planning tasks to be carried out by robots. ROSPlan is a framework for task planning in the Robot Operating System (ROS), but until now it has not been possible to use probabilistic planners within the framework. This systems paper presents a standardized integration of probabilistic...

Planning is the branch of Artificial Intelligence (AI) that seeks to automate reasoning about plans, most importantly the reasoning that goes into formulating a plan to achieve a given goal in a given situation. AI planning is model-based: a planning system takes as input a description (or model) of the initial situation, the actions available to c...

Hybrid systems are systems described by discrete as well as continuous variables. Indeed, many real-world applications present such a hybrid dynamics. Moreover, many of our interactions with the world involve us performing actions that initiate, terminate, control, or simply avoid continuous processes, and those scenarios are very common particular...

Thus far, we have only considered a simple form of PDDL modelling. While we can represent complex scenarios and behaviour, as evidenced by Section 2.4, the language has been extended in key ways which we discuss here in this chapter. The modifications empower the modeller to succinctly represent complex but common phenomenon in the planning setting...

The Planning Domain Definition Language (PDDL) has been developed and increasingly adopted by the AI planning community for over two decades. Throughout this text, we have seen a range of the more commonly used subsets of PDDL, as well as a variety of modelling examples demonstrating the expressiveness of each. The widespread use of PDDL stems from...

In the classical planning problems that we modelled in Chapters 2 and 3, all state variables are instances of predicates which are Boolean-valued. That is, they can only take values true and false. The numeric planning subset of PDDL introduces state variables whose values are (rational) numbers. These are instances of functions1. The range of a nu...

In Chapter 4, we extended classical planning to variables that are numeric in nature. In this chapter, we extend classical planning in an orthogonal way, through the introduction of time. In temporal planning, actions are durative in nature, and both the conditions and effects of an action must be generalised accordingly. Temporal planning was intr...

PDDL supports the expression of several different classes of planning problems. In this chapter and the next, we begin with the simplest class of planning problems, which are discrete, finite, and deterministic. This kind of planning is often called classical (to distinguish it from the many different extensions that have later followed).

In order to engender trust in AI, humans must understand what an AI system is trying to achieve, and why. To overcome this problem, the underlying AI process must produce justifications and explanations that are both transparent and comprehensible to the user. AI Planning is well placed to be able to address this challenge. In this paper we present...

The Defense Advanced Research Projects Agency (DARPA) recently launched the Explainable Artificial Intelligence (XAI) program that aims to create a suite of new AI techniques that enable end users to understand, appropriately trust, and effectively manage the emerging generation of AI systems. In this paper, inspired by DARPA's XAI program, we prop...

Robots used for inspection, package deliveries, moving of goods, and other logistics operations are often required to visit certain locations within specified time bounds. This gives rise to a challenging problem as it requires not only planning collision-free and dynamically-feasible motions but also reasoning temporally about when and where the r...

One of the original motivations for domain-independent planning was to generate plans that would then be executed in the environment. However, most existing planners ignore the passage of time during planning. While this can work well when absolute time does not play a role, this approach can lead to plans failing when there are external timing con...

As AI is increasingly being adopted into application solutions, the challenge of supporting interaction with humans is becoming more apparent. Partly this is to support integrated working styles, in which humans and intelligent systems cooperate in problem-solving, but also it is a necessary step in the process of building trust as humans migrate g...

In real world environments the state is almost never completely known. Exploration is often expensive. The application of planning in these environments is consequently more difficult and less robust. In this paper we present an approach for predicting new information about a partially-known state. The state is translated into a partially-known mul...

The paper generalises the notion of landmarks for reasoning about planning problems involving propositional and numeric variables. Intuitively, numeric landmarks are regions in the metric space defined by the problem whose crossing is necessary for its resolution. The paper proposes a relaxation-based method for their automated extraction directly...

The deployment of robots in public environments is gaining more and more attention and interest both for the research opportunities and for the possibility of developing commercial applications over it. In these scenarios, proper definitions and implementations of human-robot interactions are crucial and the specific characteristics of the environm...

CASP is an extension of ASP that allows for numerical constraints to be added in the rules. PDDL+ is an extension of the PDDL standard language of automated planning for modeling mixed discrete-continuous dynamics. In this paper, we present CASP solutions for dealing with PDDL+ problems, i.e., encoding from PDDL+ to CASP, and extensions to the algo...

CASP is an extension of ASP that allows for numerical constraints to be added in the rules. PDDL+ is an extension of the PDDL standard language of automated planning for modeling mixed discrete-continuous dynamics. In this paper, we present CASP solutions for dealing with PDDL+ problems, i.e., encoding from PDDL+ to CASP, and extensions to the algo...

Visual representations are an essential element in human–computer interaction and can be conceived as a collection of graphical objects arranged in a two-dimensional space. It is quite natural to model visual representations through the qualitative relationships holding between their objects, and therefore, qualitative spatial relations are a funda...

This paper explores the execution of planned autonomous underwater vehicle (AUV) missions where opportunities to achieve additional utility can arise during execution. The missions are represented as temporal planning problems, with hard goals and time constraints. Opportunities are soft goals with high utility. The probability distributions for th...

Smart contracts might encode legal contracts written in natural language to represent the contracting parties’ shared understandings and intentions. The issues and research challenges involved in the validation and verification of smart contracts, particularly those running over blockchains and distributed ledgers, are explored.

A discrete differential evolution algorithm for multi-objective permutation flowshop scheduling by Marco Baioletti, Alfredo Milani and Valentino Santucci, presents an evolutionary algorithm for solving a multi-objective permutation flow-shop scheduling problem (PFSP). In particular, the authors propose an extension of the algebraic discrete differe...

The Workshop Program of the Association for the Advancement of Artificial Intelligence’s Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16) was held at the beginning of the conference, February 12-13, 2016. Workshop participants met and discussed issues with a selected focus — providing an informal setting for active exchange among rese...

The Workshop Program of the Association for the Advancement of Artificial Intelligence’s Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16) was held at the beginning of the conference, February 12-13, 2016. Workshop participants met and discussed issues with a selected focus — providing an informal setting for active exchange among rese...

Intervention autonomous underwater vehicles (I-AUVs) have the potential to open new avenues for the maintenance and monitoring of offshore subsea facilities in a cost-effective way. However, this requires challenging intervention operations to be carried out persistently, thus minimizing human supervision and ensuring a reliable vehicle behaviour u...

Although PDDL is an expressive modelling language, a significant limitation is imposed on the structure of actions: the parameters of actions are restricted to values from finite (in fact, explicitly enumerated) domains. There is one exception to this, introduced in PDDL2.1, which is that durative actions may have durations that are chosen (possibl...

Visual Information Extraction (VIE) is a technique that enables users to perform information extraction driven by the visual appearance and the spatial arrangement of the information. In this paper, we present an extension of the VIE theory and an upgrade of its supporting tool to handle the new, relevant application domain of biomedical imaging.

PDDL+ is an extension of PDDL that enables modelling planning domains with mixed discrete-continuous dynamics. In this paper we present a new approach to PDDL+ planning based on Constraint Answer Set Programming (CASP), i.e. ASP rules plus numerical constraints. To the best of our knowledge, ours is the first attempt to link PDDL+ planning and logi...

When facing real world planning problems, standard planners are often inadequate and enhancement of the current techniques are required. In this paper we present the challenges that we have faced in solving the Unit Commitment (UC) problem, a well-known problem in the electrical power industry for which current best methods are based on Mixed Integ...

Planning in hybrid systems is important for dealing with real-world applications. PDDL+ supports this representation of domains with mixed discrete and continuous dynamics, and supports events and processes modelling exogenous change. Motivated by numerous SAT-based planning approaches, we propose an approach to PDDL+ planning through SMT, describi...

Planning in hybrid systems has been gaining research interest in the Artificial Intelligence community in recent years. Hybrid systems allow for a more accurate representation of real world problems, though solving them is very challenging due to complex system dynamics and a large model feature set. We developed DiNo, a new planner designed to tac...

The global growth in urbanisation increases the demand for services including road transport infrastructure, presenting challenges in terms of mobility. In this scenario, optimising the exploitation of urban road networks is a pivotal challenge. Existing urban traffic control approaches, based on complex mathematical models, can effectively deal wi...

Planning in hybrid systems is key to dealing with real-world applications. PDDL+ supports this representation of domains with mixed discrete and continuous dynamics, and supports events and processes modeling exogenous change. Motivated by numerous SAT-based planning approaches, we propose an approach to PDDL+ planning through SMT, describing an SM...

Planning in hybrid systems is important for dealing with real-world applications. PDDL+ supports this representation of domains with mixed discrete and continuous dynamics, and supports events and processes modelling exogenous change. Motivated by numerous SAT-based planning approaches, we propose an approach to PDDL+ planning through SMT, describi...

Planning plays a role in achieving long-term behaviour (persistent autonomy) without human intervention. Such behaviour engenders plans which are expected to last over many hours, or even days. Such a problem is too large for current planners to solve as a single planning problem, but is well-suited to decomposition and abstraction planning techniq...

This paper explores the execution of planned missions in situations in which opportunities to achieve additional utility can arise during execution. The missions are represented as temporal planning problems, with hard goals and time constraints. Opportunities are soft goals with high utility. The probability distributions for the occurrences of th...

Autonomous intelligent robotic systems are becoming increasingly important in a wide range of applications. Many of these application contexts are in physical situations out of human reach, so that a robot, or team of robots, must be capable of operating for long periods without human intervention. This requires a strategic planning capability as w...

Automatic planning tools can be a valuable aid to build control systems for a wide range of everyday appliances. Many systems which interact with the real world present a non-deterministic behaviour, often made more complex by nonlinear continuous dynamics. In this context, strong planning, which requires finding a plan that is guaranteed to achiev...

The Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI-15) was held in January 2015 in Austin, Texas (USA) The conference program was cochaired by Sven Koenig and Blai Bonet. This report contains reflective summaries of the main conference, the robotics pro- gram, the AI and robotics workshop, the Virtual Agent Exhibition, the What’s Hot...

Planning in hybrid domains poses a special challenge due to the involved mixed discrete-continuous dynamics. A recent solving approach for such domains is based on applying model checking techniques on a translation of PDDL+ planning problems to hybrid automata. However, the proposed translation is limited because must behavior is only overapproxim...