Showing papers by "Giuseppe De Giacomo published in 2022"

AI-augmented Business Process Management Systems: A Research Manifesto

Abstract: AI-augmented Business Process Management Systems (ABPMSs) are an emerging class of process-aware information systems, empowered by trustworthy AI technology. An ABPMS enhances the execution of business processes with the aim of making these processes more adaptable, proactive, explainable, and context-sensitive. This manifesto presents a vision for ABPMSs and discusses research challenges that need to be surmounted to realize this vision. To this end, we define the concept of ABPMS, we outline the lifecycle of processes within an ABPMS, we discuss core characteristics of an ABPMS, and we derive a set of challenges to realize systems with these characteristics.

Verification and Monitoring for First-Order LTL with Persistence-Preserving Quantification over Finite and Infinite Traces

Abstract: We address the problem of model checking first-order dynamic systems where new objects can be injected in the active domain during execution. Notable examples are systems induced by a first-order action theory, e.g., expressed in the Situation Calculus. Recent results have shown that, under the state-boundedness assumption, such systems, in spite of having a first-order representation of the state, admit decidable model checking for full first-order mu-calculus. However, interestingly, model checking remains undecidable in the case of first-order LTL (LTL-FO). In this paper, we show that in LTL-FOp, which is the fragment of LTL-FO in which quantification is over objects that persist along traces, model checking state-bounded systems becomes decidable over finite and infinite traces. We then employ this result to show how to handle monitoring of LTL-FOp properties against a trace stemming from an unknown state-bounded dynamic system, simultaneously considering the finite trace up to the current point, and all its possibly infinite future continuations.

LTLf Synthesis as AND-OR Graph Search: Knowledge Compilation at Work

Abstract: Synthesis techniques for temporal logic specifications are typically based on exploiting symbolic techniques, as done in model checking. These symbolic techniques typically use backward fixpoint computation. Planning, which can be seen as a specific form of synthesis, is a witness of the success of forward search approaches. In this paper, we develop a forward-search approach to full-fledged Linear Temporal Logic on finite traces (LTLf) synthesis. We show how to compute the Deterministic Finite Automaton (DFA) of an LTLf formula on-the-fly, while performing an adversarial forward search towards the final states, by considering the DFA as a sort of AND-OR graph. Our approach is characterized by branching on suitable propositional formulas, instead of individual evaluations, hence radically reducing the branching factor of the search space. Specifically, we take advantage of techniques developed for knowledge compilation, such as Sentential Decision Diagrams (SDDs), to implement the approach efficiently.

Compositional Safety LTL Synthesis

Abstract: AbstractReactive synthesis holds the promise of generating automatically a verifiably correct program from a high-level specification. A popular such specification language is Linear Temporal Logic (LTL). Unfortunately, synthesizing programs from general LTL formulas, which relies on first constructing a game arena and then solving the game, does not scale to large instances. The specifications from practical applications are usually large conjunctions of smaller LTL formulas, which inspires existing compositional synthesis approaches to take advantage of this structural information. The main challenge here is that they solve the game only after obtaining the game arena, the most computationally expensive part in the procedure. In this work, we propose a compositional synthesis technique to tackle this difficulty by synthesizing a program for each small conjunct separately and composing them one by one. While this approach does not work for general LTL formulas, we show here that it does work for Safety LTL formulas, a popular and important fragment of LTL. While we have to compose all the programs of small conjuncts in the worst case, we can prune the intermediate programs to make later compositions easier and immediately conclude unrealizable as soon as some part of the specification is found unrealizable. By comparing our compositional approach with a portfolio of all other approaches, we observed that our approach was able to solve a notable number of instances not solved by others. In particular, experiments on scalable conjunctive benchmarks showed that our approach scale well and significantly outperform current Safety LTL synthesis techniques. We conclude that our compositional approach is an important contribution to the algorithmic portfolio of Safety LTL synthesis.

Planning for Temporally Extended Goals in Pure-Past Linear Temporal Logic: A Polynomial Reduction to Standard Planning

Abstract: We study classical planning for temporally extended goals expressed in Pure-Past Linear Temporal Logic (PPLTL). PPLTL is as expressive as Linear-time Temporal Logic on finite traces (LTLf), but as shown in this paper, it is computationally much better behaved for planning. Specifically, we show that planning for PPLTL goals can be encoded into classical planning with minimal overhead, introducing only a number of new fluents that is at most linear in the PPLTL goal and no spurious additional actions. Based on these results, we implemented a system called Plan4Past, which can be used along with state-of-the-art classical planners, such as LAMA. An empirical analysis demonstrates the practical effectiveness of Plan4Past, showing that a classical planner generally performs better with our compilation than with other existing compilations for LTLf goals over the considered benchmarks.

Markov Abstractions for PAC Reinforcement Learning in Non-Markov Decision Processes

Abstract: Our work aims at developing reinforcement learning algorithms that do not rely on the Markov assumption. We consider the class of Non-Markov Decision Processes where histories can be abstracted into a finite set of states while preserving the dynamics. We call it a Markov abstraction since it induces a Markov Decision Process over a set of states that encode the non-Markov dynamics. This phenomenon underlies the recently introduced Regular Decision Processes (as well as POMDPs where only a finite number of belief states is reachable). In all such kinds of decision process, an agent that uses a Markov abstraction can rely on the Markov property to achieve optimal behaviour. We show that Markov abstractions can be learned during reinforcement learning. Our approach combines automata learning and classic reinforcement learning. For these two tasks, standard algorithms can be employed. We show that our approach has PAC guarantees when the employed algorithms have PAC guarantees, and we also provide an experimental evaluation.

Iterative Depth-First Search for Fully Observable Non-Deterministic Planning

Abstract: Fully Observable Non-Deterministic ( FOND ) planning models uncertainty through actions with non-deterministic effects. Existing FOND planning algorithms are effective and employ a wide range of techniques. However, most of the existing algorithms are not robust for dealing with both non-determinism and task size. In this paper, we develop a novel iterative depth-first search algorithm that solves FOND planning tasks and produces strong cyclic policies . Our algorithm is explicitly designed for FOND planning, addressing more directly the non-deterministic aspect of FOND planning, and it also exploits the benefits of heuristic functions to make the algorithm more effective during the iterative searching process. We compare our proposed algorithm to well-known FOND planners, and show that it has robust performance over several distinct types of FOND domains considering different metrics.

Act for Your Duties but Maintain Your Rights

Abstract: Most of the synthesis literature has focused on studying how to synthesize a strategy to fulfill a task. This task is a duty for the agent. In this paper, we argue that intelligent agents should also be equipped with rights, that is, tasks that the agent itself can choose to fulfill (e.g., the right of recharging the battery). The agent should be able to maintain these rights while acting for its duties. We study this issue in the context of LTLf synthesis: we give duties and rights in terms of LTLf specifications, and synthesize a suitable strategy to achieve the duties that can be modified on-the-fly to achieve also the rights, if the agent chooses to do so. We show that handling rights does not make synthesis substantially more difficult, although it requires a more sophisticated solution concept than standard LTLf synthesis. We also extend our results to the case in which further duties and rights are given to the agent while already executing.

Grounding LTLf Specifications in Images

Abstract: A critical challenge in neurosymbolic approaches is to handle the symbol grounding problem without direct supervision. That is mapping high-dimensional raw data into an interpretation over a finite set of abstract concepts with a known meaning, without using labels. In this work, we ground symbols into sequences of images by exploiting symbolic logical knowledge in the form of Linear Temporal Logic over finite traces (LTLf) formulas, and sequence-level labels expressing if a sequence of images is compliant or not with the given formula. Our approach is based on translating the LTLf formula into an equivalent deterministic finite automaton (DFA) and interpreting the latter in fuzzy logic. Experiments show that our system outperforms recurrent neural networks in sequence classification and can reach high image classification accuracy without being trained with any single-image label.

Automata Cascades: Expressivity and Sample Complexity

Abstract: Every automaton can be decomposed into a cascade of basic prime automata. This is the Prime Decomposition Theorem by Krohn and Rhodes. Guided by this theory, we propose automata cascades as a structured, modular, way to describe automata as complex systems made of many components, each implementing a specific functionality. Any automaton can serve as a component; using specific components allows for a fine-grained control of the expressivity of the resulting class of automata; using prime automata as components implies specific expressivity guarantees. Moreover, specifying automata as cascades allows for describing the sample complexity of automata in terms of their components. We show that the sample complexity is linear in the number of components and the maximum complexity of a single component, modulo logarithmic factors. This opens to the possibility of learning automata representing large dynamical systems consisting of many parts interacting with each other. It is in sharp contrast with the established understanding of the sample complexity of automata, described in terms of the overall number of states and input letters, which implies that it is only possible to learn automata where the number of states is linear in the amount of data available. Instead our results show that one can learn automata with a number of states that is exponential in the amount of data available.

Beyond Strong-Cyclic: Doing Your Best in Stochastic Environments

Abstract: ``Strong-cyclic policies" were introduced to formalize trial-and-error strategies and are known to work in Markovian stochastic domains, i.e., they guarantee that the goal is reached with probability 1. We introduce ``best-effort" policies for (not necessarily Markovian) stochastic domains. These generalize strong-cyclic policies by taking advantage of stochasticity even if the goal cannot be reached with probability 1. We compare such policies with optimal policies, i.e., policies that maximize the probability that the goal is achieved, and show that optimal policies are best-effort, but that the converse is false in general. With this framework at hand, we revisit the foundational problem of what it means to plan in nondeterministic domains when the nondeterminism has a stochastic nature. We show that one can view a nondeterministic planning domain as a representation of infinitely many stochastic domains with the same support but different probabilities, and that for temporally extended goals expressed in LTL/LTLf a finite-state best-effort policy in one of these domains is best-effort in each of the domains. In particular, this gives an approach for finding such policies that reduces to solving finite-state MDPs with LTL/LTLf goals. All this shows that ``best-effort" policies are robust to changes in the probabilities, as long as the support is unchanged.

Automatic Synthesis of Dynamic Norms for Multi-Agent Systems

Abstract: Norms have been widely proposed to coordinate and regulate multi-agent systems (MAS) behaviour. We consider the problem of synthesising and revising the set of norms in a normative MAS to satisfy a design objective expressed in Alternating Time Temporal Logic (ATL*). ATL* is a well-established language for strategic reasoning, which allows the specification of norms that constrain the strategic behaviour of agents. We focus on dynamic norms, that is, norms corresponding to Mealy machines, that allow us to place different constraints on the agents' behaviour depending on the state of the norm and the state of the underlying MAS. We show that synthesising dynamic norms is (k + 1)-EXPTIME, where k is the alternation depth of quantifiers in the ATL* specification. Note that for typical cases of interest, k is either 1 or 2. We also study the problem of removing existing norms to satisfy a new objective, which we show to be 2EXPTIME-complete.

Graphol: A Graphical Language for Ontology Modeling Equivalent to OWL 2

Abstract: In this paper we study Graphol, a fully graphical language inspired by standard formalisms for conceptual modeling, similar to the UML class diagram and the ER model, but equipped with formal semantics. We formally prove that Graphol is equivalent to OWL 2, i.e., it can capture every OWL 2 ontology and vice versa. We also present some usability studies indicating that Graphol is suitable for quick adoption by conceptual modelers that are familiar with UML and ER. This is further testified by the adoption of Graphol for ontology representation in several industrial projects.

Mimicking Behaviors in Separated Domains

Abstract: Devising a strategy to make a system mimicking behaviors from another system is a problem that naturally arises in many areas of Computer Science. In this work, we interpret this problem in the context of intelligent agents, from the perspective of ltl f , a formalism commonly used in AI for expressing finite-trace properties. Our model consists of two separated dynamic domains, D A and D B , and an ltl f specification that formalizes the notion of mimicking by mapping properties on behaviors (traces) of D A into properties on behaviors of D B . The goal is to synthesize a strategy that step-by-step maps every behavior of D A into a behavior of D B so that the specification is met. We consider several forms of mapping specifications, ranging from simple ones to full ltl f , and for each we study synthesis algorithms and computational properties.

Sample Complexity of Automata Cascades

Abstract: Every automaton can be decomposed into a cascade of ba- sic automata. This is the Prime Decomposition Theorem by Krohn and Rhodes. We show that cascades allow for describ- ing the sample complexity of automata in terms of their components. In particular, we show that the sample complexity is linear in the number of components and the maximum complexity of a single component, modulo logarithmic fac- tors. This opens to the possibility of learning automata rep-resenting large dynamical systems consisting of many parts interacting with each other. It is in sharp contrast with the established understanding of the sample complexity of au- tomata, described in terms of the overall number of states and input letters, which implies that it is only possible to learn automata where the number of states is linear in the amount of data available. Instead our results show that one can learn automata with a number of states that is exponential in the amount of data available.

Iterative Depth-First Search for FOND Planning

Abstract: Fully Observable Non-Deterministic (FOND) planning models uncertainty through actions with non-deterministic effects. Existing FOND planning algorithms are effective and employ a wide range of techniques. However, most of the existing algorithms are not robust for dealing with both non-determinism and task size. In this paper, we develop a novel iterative depth-first search algorithm that solves FOND planning tasks and produces strong cyclic policies. Our algorithm is explicitly designed for FOND planning, addressing more directly the non-deterministic aspect of FOND planning, and it also exploits the benefits of heuristic functions to make the algorithm more effective during the iterative searching process. We compare our proposed algorithm to well-known FOND planners, and show that it has robust performance over several distinct types of FOND domains considering different metrics.

Synthesis of Maximally Permissive Strategies for LTLf Specifications

Abstract: In this paper, we study synthesis of maximally permissive strategies for Linear Temporal Logic on finite traces (LTLf) specifications. That is, instead of computing a single strategy (aka plan, or policy), we aim at computing the entire set of strategies at once and then choosing among them while in execution, without committing to a single one beforehand. Maximally permissive strategies have been introduced and investigated for safety properties, especially in the context of Discrete Event Control Theory. However, the available results for safety properties do not apply to reachability properties (eventually reach a given state of affair) nor to LTLf properties in general. In this paper, we show that maximally permissive strategies do exist also for reachability and general LTLf properties, and can in fact be computed with minimal overhead wrt the computation of a single strategy using state-of-the-art tools.

Controller Synthesis for Manufacturing Systems in the Situation Calculus

Abstract: Manufacturing is transitioning from a mass production model to a service model in which facilities ‘bid’ for products. To decide whether to bid for a previously unseen product, a facility must be able to synthesize, on the fly, a process plan controller that delegates abstract manufacturing tasks in a supplied process recipe to the available manufacturing resources. First-order representations of the state are commonly considered in reasoning about action in AI. Here we show that we can leverage the wide literature on the Situation Calculus automatically synthesize such controllers. We identify two important decidable cases—finite do-mains and bounded action theories—for which we provide practical synthesis techniques.

Verification of Generic, Relational Transition Systems

Abstract: Generic, relational transition systems form an interesting class of infinite-state transition systems that naturally captures the execution semantics of a variety of formalisms expressing processes operating over (relational) data. Examples of such data-aware processes include action theories in the situation calculus in AI and data-centric business processes in BPM. In this extended abstract, we summarize the main body of results produced in a decade-long research program focused on the verification of generic, relational transition systems against properties specified using variants of first-order temporal logics.