Search

Reward machines (RMs) are a recent formalism for representing the reward function of a reinforcement learning task through a finite-state machine whose edges encode subgoals of the task using high-level events. The structure of RMs enables the decomposition of a task into simpler and independently solvable subtasks that help tackle long-horizon and/or sparse reward tasks. We propose a formalism for further abstracting the subtask structure by endowing an RM with the ability to call other RMs, thus composing a hierarchy of RMs (HRM). We exploit HRMs by treating each call to an RM as an independently solvable subtask using the options framework, and describe a curriculum-based method to learn HRMs from traces observed by the agent. Our experiments reveal that exploiting a handcrafted HRM leads to faster convergence than with a flat HRM, and that learning an HRM is feasible in cases where its equivalent flat representation is not., Comment: Preprint accepted for publication to the 40th International Conference on Machine Learning (ICML-23)

Logic-based machine learning aims to learn general, interpretable knowledge in a data-efficient manner. However, labelled data must be specified in a structured logical form. To address this limitation, we propose a neural-symbolic learning framework, called Feed-Forward Neural-Symbolic Learner (FFNSL), that integrates a logic-based machine learning system capable of learning from noisy examples, with neural networks, in order to learn interpretable knowledge from labelled unstructured data. We demonstrate the generality of FFNSL on four neural-symbolic classification problems, where different pre-trained neural network models and logic-based machine learning systems are integrated to learn interpretable knowledge from sequences of images. We evaluate the robustness of our framework by using images subject to distributional shifts, for which the pre-trained neural networks may predict incorrectly and with high confidence. We analyse the impact that these shifts have on the accuracy of the learned knowledge and run-time performance, comparing FFNSL to tree-based and pure neural approaches. Our experimental results show that FFNSL outperforms the baselines by learning more accurate and interpretable knowledge with fewer examples., Comment: Pre-print, work in progress

The RADON Constraint Definition Language (CDL) can be used to encode both functional and non-functional requirements on Function as a Service (FaaS) architecture. The primary use of its associated Verification Tool is to check whether a given architecture conforms to a given CDL specification. In addition, it can also suggest corrections to a RADON model that does not conform to a CDL specification and learn requirements (expressed as CDL constraints) from examples of valid and invalid RADON models. This document presents the formal semantics for the CDL and the proofs of correctness of the Verification Tool.

In recent years, non-monotonic Inductive Logic Programming has received growing interest. Specifically, several new learning frameworks and algorithms have been introduced for learning under the answer set semantics, allowing the learning of common-sense knowledge involving defaults and exceptions, which are essential aspects of human reasoning. In this paper, we present a noise-tolerant generalisation of the learning from answer sets framework. We evaluate our ILASP3 system, both on synthetic and on real datasets, represented in the new framework. In particular, we show that on many of the datasets ILASP3 achieves a higher accuracy than other ILP systems that have previously been applied to the datasets, including a recently proposed differentiable learning framework., Comment: To appear in Advances in Cognitive Systems

The goal of Inductive Logic Programming (ILP) is to find a hypothesis that explains a set of examples in the context of some pre-existing background knowledge. Until recently, most research on ILP targeted learning definite logic programs. This thesis constitutes the first comprehensive work on learning answer set programs, introducing new learning frameworks, theoretical results on the complexity and generality of these frameworks, algorithms for learning ASP programs, and an extensive evaluation of these algorithms. Although there is previous work on learning ASP programs, existing learning frameworks are either brave -- where examples should be explained by at least one answer set -- or cautious where examples should be explained by all answer sets. There are cases where brave induction is too weak and cautious induction is too strong. Our proposed frameworks combine brave and cautious learning and can learn ASP programs containing choice rules and constraints. Many applications of ASP use weak constraints to express a preference ordering over the answer sets of a program. Learning weak constraints corresponds to preference learning, which we achieve by introducing ordering examples. We then explore the generality of our frameworks, investigating what it means for a framework to be general enough to distinguish one hypothesis from another. We show that our frameworks are more general than both brave and cautious induction. We also present a new family of algorithms, called ILASP (Inductive Learning of Answer Set Programs), which we prove to be sound and complete. This work concerns learning from both non-noisy and noisy examples. In the latter case, ILASP returns a hypothesis that maximises the coverage of examples while minimising the length of the hypothesis. In our evaluation, we show that ILASP scales to tasks with large numbers of examples finding accurate hypotheses even in the presence of high proportions of noisy examples. Open Access