Abstract Scheduling systems are critical for planning projects, resources, and activities across many industries to achieve goals efficiently. As scheduling requirements grow in complexity, the use of Artificial Intelligence (AI) solutions has received more attention. However, providing comprehensible explanations of these decision-making processes remains a challenge and blocker to adoption. The emergent field of eXplainable Artificial Intelligence (XAI) aims to address this by establishing human-centric interpretation of influencing factors for machine decisions. The leading field of autonomous interpretation in Natural Language Processing (NLP) is Large Language Model (LLM)s, for their generalist knowledge and reasoning capabilities. To explore LLMs’ potential to generate explanations for scheduling queries, we selected a benchmark set of Job Shop scheduling problems. A novel framework that integrates the selected language models, GPT-4 and Large Language Model Meta AI (LLaMA), into scheduling systems is introduced, facilitating human-like explanations to queries from different categories through few-shot learning. The explanations were analysed for accuracy, consistency, completeness, conciseness, and language across different scheduling problem sizes and complexities. The approach achieved an overall accuracy of 59% with GPT-4 and 35% with LLaMA, with minimal impact from the varied schedule sizes observed, proving the approach can handle different datasets and is performance scalable. Several responses demonstrated high comprehension of complex queries; however, response quality fluctuated due to the few-shot learning approach. This study establishes a baseline for measuring generalist LLM capabilities in handling explanations for autonomous scheduling systems, with promising results for an LLM providing XAI interactions to explain scheduling decisions.

With an increasing number of satellites in orbit, consensus across a heterogeneous group of satellites can lead to a more neutral, unbiased, and accurate decisions. Fault tolerant consensus algorithms such as Practical Byzantine Fault Tolerance (pBFT) require communication with all other network members up to 4 times. In a network with thousands of satellites in space on different trajectories, this time can approach millennia. There-fore, identifying a subset of satellites that can form a sub-network able to converge to a consensus decision in a useful time window, while maximising the number of members to increase consensus accuracy and trustworthiness, can be formulated as a multi-objective combinatorial optimisation problem. The problem is explained and defined with the optimisation method and the consensus algorithm steps described. Metrics for measuring the output of the optimal pareto front are considered and applied to the front computed. The real satellite positions used generate a non-fixed topology and high latency scenario such as that of a real on-orbit decision being made. The trend shown over 100 days of satellite positions propagation with up to 82 International Charter: Space and Major Disasters satellites shows up to 22 satellites can be used in a subset with a near linear increase in consensus time along the optimal pareto front and exponential trend for the mean values computed over 100 runs of the NSGA-II algorithm. The minimum consensus time is found to be 47 minutes for a subset of 4 satellites for the given time frame.

Satellite schedules are derived from satellite mission objectives, which are mostly managed manually from the ground. This increases the need to develop autonomous on-board scheduling capabilities and reduce the requirement for manual management of satellite schedules. Additionally, this allows the unlocking of more capabilities on-board for decision-making, leading to an optimal campaign. However, there remain trust issues in decisions made by Artificial Intelligence (AI) systems, especially in risk-averse environments, such as satellite operations. Thus, an explanation layer is required to assist operators in understanding decisions made, or planned, autonomously on-board. To this aim, a satellite scheduling problem is formulated, utilizing real world data, where the total number of actions are maximised based on the environmental constraints that limit observation and down-link capabilities. The formulated optimisation problem is solved with a Constraint Programming (CP) method. Later, the mathematical derivation for an Abstract Argumentation Framework (AAF) for the test case is provided. This is proposed as the solution to provide an explanation layer to the autonomous decision-making system. The effectiveness of the defined AAF layer is proven on the daily schedule of an Earth Observation (EO) mission, monitoring land surfaces, demonstrating greater capabilities and flexibility, for a human operator to inspect the machine provided solution.

In this paper an initial approach to Intelligent Control (IC) using Genetic Programming (GP) for access to space applications is presented. GP can be employed successfully to design a controller even for complex systems, where classical controllers fail because of the high nonlinearity of the systems. The main property of GP, that is its ability to autonomously create explicit mathematical equations starting from a very poor knowledge of the considered plant, or just data, can be exploited for a vast range of applications. Here, GP has been used to design the control law in an Intelligent Control framework for a modified version of the Goddard Rocket problem in 3 different failure scenarios, where the approach to IC consists in an online re-evaluation of the control law using GP when a considerably big change in the environment or in the plant happens. The presented results are then used to highlight the potential benefits of the method, as well as aspects that will need further developments.

Reinforcement learning is a popular machine learning paradigm which can find near optimal solutions to complex problems. Most often, these procedures involve function approximation using neural networks with gradient based updates to optimise weights for the problem being considered. While this common approach generally works well, there are other update mechanisms which are largely unexplored in reinforcement learning. One such mechanism is Extreme Learning Machines. These were initially proposed to drastically improve the training speed of neural networks and have since seen many applications. Here we attempt to apply extreme learning machines to a reinforcement learning problem in the same manner as gradient based updates. This new algorithm is called Extreme Q-Learning Machine (EQLM). We compare its performance to a typical Q-Network on the cart-pole task - a benchmark reinforcement learning problem - and show EQLM has similar long-term learning performance to a Q-Network.

Stopping exploration of the search space regions that can be proven to contain only inferior solutions is an important acceleration technique in optimization algorithms. This study is focused on the utility of trie-based data structures for indexing discrete sets that allow to detect such a state faster. An empirical evaluation is performed in the context of index operations executed by a label setting algorithm for solving the Elementary Shortest Path Problem with Resource Constraints. Numerical simulations are run to compare a trie with a HAT-trie, a variant of a trie, which is considered as the fastest in-memory data structure for storing text in sorted order, further optimized for efficient use of cache in modern processors. Results indicate that a HAT-trie is better suited for indexing sparse multi dimensional data, such as sets with high cardinality, offering superior performance at a lower memory footprint. Therefore, HAT-tries remain practical when tries reach their scalability limits due to an expensive memory allocation pattern. Authors leave a final note on comparing and reporting credible time benchmarks for the Elementary Shortest Path Problem with Resource Constraints.

This paper introduces a new instance-based algorithm for multiclass classification problems where the classes have a natural order. The proposed algorithm extends the state-of-the-art gravitational models by generalizing the scaling behavior of the class-pattern interaction force. Like the other gravitational models, the proposed algorithm classifies new patterns by comparing the magnitude of the force that each class exerts on a given pattern. To address ordinal problems, the algorithm assumes that, given a pattern, the forces associated to each class follow a unimodal distribution. For this reason, a weight matrix that allows to modify the metric in the attributes space and a vector of parameters that allows to modify the force law for each class have been introduced in the model definition. Furthermore, a probabilistic formulation of the error function allows the estimation of the model parameters using global and local optimization procedures toward minimization of the errors and penalization of the non unimodal outputs. One of the strengths of the model is its competitive grade of interpretability which is a requisite in most of real applications. The proposed algorithm is compared to other well-known ordinal regression algorithms on discretized regression datasets and real ordinal regression datasets. Experimental results demonstrate that the proposed algorithm can achieve competitive generalization performance and it is validated using nonparametric statistical tests.

In this paper, the well known stagewise additive modeling using a multiclass exponential (SAMME) boosting algorithm is extended to address problems where there exists a natural order in the targets using a cost-sensitive approach. The proposed ensemble model uses an extreme learning machine (ELM) model as a base classifier (with the Gaussian kernel and the additional regularization parameter). The closed form of the derived weighted least squares problem is provided, and it is employed to estimate analytically the parameters connecting the hidden layer to the output layer at each iteration of the boosting algorithm. Compared to the state-of-the-art boosting algorithms, in particular those using ELM as base classifier, the suggested technique does not require the generation of a new training dataset at each iteration. The adoption of the weighted least squares formulation of the problem has been presented as an unbiased and alternative approach to the already existing ELM boosting techniques. Moreover, the addition of a cost model for weighting the patterns, according to the order of the targets, enables the classifier to tackle ordinal regression problems further. The proposed method has been validated by an experimental study by comparing it with already existing ensemble methods and ELM techniques for ordinal regression, showing competitive results.

Ordinal regression (OR) is an important branch of supervised learning in between the multiclass classification and regression. In this paper, the traditional classification scheme of neural network is adapted to learn ordinal ranks. The model proposed imposes monotonicity constraints on the weights connecting the hidden layer with the output layer. To do so, the weights are transcribed using padding variables. This reformulation leads to the so-called inequality constrained least squares (ICLS) problem. Its numerical solution can be obtained by several iterative methods, for example, trust region or line search algorithms. In this proposal, the optimum is determined analytically according to the closed-form solution of the ICLS problem estimated from the Karush-Kuhn-Tucker conditions. Furthermore, following the guidelines of the extreme learning machine framework, the weights connecting the input and the hidden layers are randomly generated, so the final model estimates all its parameters without iterative tuning. The model proposed achieves competitive performance compared with the state-of-the-art neural networks methods for OR.