Incorporating auxiliary objectives into Reinforcement Learning allows agents to acquire additional knowledge, thereby increasing their search for the optimal policy. This article presents the Model-Predictor Proximal Policy Optimization (MP-PPO) algorithm, which merges the concepts of various PPO variants with a Transformer probabilistic prediction module. This model capitalizes on the time dependence inherent in energy management systems, predicting future state transitions by learning to predict certain state characteristics. Notably, our algorithm seamlessly integrates this predictive capability into the Actor-Critic architecture, avoiding the need for an external model. Through experiments on real data, we demonstrate that integrating predictive capabilities for partial state prediction improves both the sample effectiveness and efficiency of the original PPO approach without requiring exterior prior information.

In a large domain of classification problems for real applications, like human activity recognition, separable spaces between groups of concepts are easier to learn than each concept alone. This is because the search space biases required to separate groups of classes (or concepts) are more relevant than the ones needed to separate classes individually. For example, it is easier to learn the activities related to the body movements group (running, walking) versus "on-wheels" activities group (bicycling, driving a car), before learning more specific classes inside each of these groups. Despite the obvious interest of this approach, our theoretical analysis shows a high complexity for finding an exact solution. We propose in this paper an original approach based on the association of clustering and classification approaches to overcome this limitation. We propose a better approach to learn the concepts by grouping classes recursively rather than learning them class by class. We introduce an effective greedy algorithm and two theoretical measures, namely cohesion and dispersion, to evaluate the connection between the clusters and the classes. Extensive experiments on the SHL dataset show that our approach improves classification performances while reducing the number of instances used to learn each concept.

This paper proposes an original approach based on expert trajectories combination and Deep Reinforcement Learning to provide a better MineCraft player. The combination is based on the idea that the problem is naturally decomposable and the search space presents large plateaus. We use two steps approach to build a better trajectory from all existed expert trajectories and consequently to extract an optimal policy. The first step uses Birch clustering approach and images cosine similarity to obtain compact representation and substantial state and action space reduction. To reduce the overall complexity, the image distances are computed in images latent space trained by an encoder-decoder model. In the second step, we first eliminate plateaus to keep only the nodes with non-zero rewards then we compare trajectories using the Bellman equation and an appropriate value function. By checking the incremental compatibility of the trajectory of compact representations, we build the solution combining the best compatible sub-trajectories of the experts. The experimental results on NeurIPS MineRL 2020 challenge show that training the actors model on the most rewarding extracted subset of trajectories leads to achieve state-of-the-art performances on the MineCraft environment. The paper's source code is available here: https://github.com/thomJeffDoe/CompareTrajectories.

Time is an important dimension in sound event detection (SED) systems. However, evaluating the performance of SED systems is directly taken from the classical machine learning domain, and they are not well adapted to the needs of these systems such as recognizing the time, duration, detection, and uniformity of sound events. Despite its importance, it is not well-developed yet. Current methods are highly biased by their assumptions and may misleadingly present convincible results. This paper presents a novel multimodal method to evaluate SED systems from multiple perspectives such as detection, total duration, relative duration, and uniformity. Furthermore, the proposed method is simple, time-efficient, visualizable, extensible, open-source, and overcomes the limitations of existing methods. The benefits of the proposed approach are demonstrated by re-evaluating the best systems presented in a known challenge on sound event detection.

The coronavirus disease 2019 (COVID-19) has been stated as a global pandemic, and the BA.4 and BA.5 variants are anticipated to drive the next wave of COVID-19 infection. Early diagnosis of this infection reduces its viral excretion. In this paper, after a large study of existing algorithms for pre-symptomatic COVID-19 detection in the state-of-the-art, we discovered a notable flaw in most models related to the choice of the evaluation function, such that, all the tested algorithms perform worse (from the evaluation function perspective) than an algorithm that generates alarms randomly from a binomial distribution. Therefore, we propose a simple and less biased evaluation function to better compare the quality of different algorithms. Comprehensive experimental evaluations of the state-of-the-art algorithms over the real-world dataset published by Nature Medicine journal contains 84 COVID-19 patients and 2,000 healthy participants show the effectiveness and the relevance of our evaluation method. Moreover, the proposed framework is released as an open-source library.

Patients with chronic conditions require medical care at their home. To this end, a smart follow-up and monitoring system is proposed, called Hapicare; which applies ontology-based uncertain reasoning over IoT sensors data and self-assessment. While similar approaches rely on certain events and rules, the proposed monitoring system is based on probabilistic reasoning that interleaves Bayesian and non-monotonic inference. The latter is defined by using rule-based on concepts of the Semantic Sensor Network (SSN) and the SNOMED-CT ontologies. This system also considers uncertain contextual information captured from sensors and the history of patients in order to better diagnose the current situation and trigger suitable reactions. It allows also handling overlaps between symptoms, the possibility of errors and hidden facts. Hapicare is developed in the context of Medolution EU project.

Monitoring industrial infrastructures are undergoing a critical transformation with industry 4.0. Monitoring solutions must follow the system behavior in real time and must adapt to its continuous change. We propose in this paper an autoencoder model-based approach for tracking abnormalities in industrial application. A set of sensors collects data from turbo-compressors and an original two-level machine learning LSTM autoencoder architecture defines a continuous nominal vibration model. Normalized thresholds (ISO 20816) between the model and the system generates a possible abnormal situation to diagnose. Experimental results, including hyper-parameter optimization on large real data and domain expert analysis, show that our proposed solution gives promising results.

Infant's cry is an innate response to express several situations including pain, disturbance and discomfort. Therefore the automatic recognition of infant cries patterns is the key factor to develop successful ambient intelligence applications to enhance the quality of life of both infants and parents. This paper proposes a complete machine learning process including consistent dataset generation from infant cries and selecting appropriate sound features, with promising experimental results for enhancing the monitoring of infants in real world settings. The originality of the proposed approach lies in its ability to detect and analyze automatically discomfort signals, which recurrently affects 20 to 25% of newborns. The machine learning process includes low-level audio features selection methods from labeled infant pre-cry recordings as well as high-level features characterizing the envelop of the crying. The classification is performed using ensemble learning methods after a stage of features selection. The exploitation of pre-crying signals to improve the quality of the recognition is another important aspect of the proposed approach, which optimizes the accuracy of the learning step as it is shown by the obtained results on a real dataset. This result gives the opportunity to develop new baby monitors able to anticipate the infants needs.

Model transformation in the context of Model-Driven Data Warehouse is ensured by human experts. It generates an exorbitant cost and requires high proficiency. We propose in this paper a machine learning approach to reduce the expert contribution in the transformation process. We propose to express the model transformation problem as an Inductive Logic Programming one and to use existing project traces to find the best business transformation rules. We used the Aleph ILP system to learn such rules. Obtained results show that found rules are close to expert ones. Within our application context, we need to deal with several dependent concepts. Taking into account work in Layered Learning, we propose a new methodology that automatically updates the background knowledge of the concepts to be learned. Experimental results support the conclusion that this approach is suitable to solve this kind of problem.

Mining association rules is an important technique for discovering meaningful patterns in datasets. Temporal association rule mining can be decomposed into two phases: finding temporal frequent patterns and finding temporal rules construction. Till date, a large number of algorithms have been proposed in the area of mining association rules. However, most of these algorithms consider patterns as a collection of point primitives and their three basic relations (). Several applications consider patterns with duration and need to reason about intervals and their thirteen possible relationships. In this paper we investigate properties of temporal sequences represented as a collection of intervals. We present a simple framework for temporal sequence and describe DATTES (Discovering pATterns in TEmporal Sequences), an innovative algorithm using interval properties to mine temporal patterns. The framework can be used to mine temporal association rules. According to some interval algebra properties, this paper introduces a new confidence evaluation function for mining temporal rules. Experiments on real dataset (human face identification problem) show the effectiveness and the performances of this approach.

Several works have shown that the covering test in relational learning exhibits a phase transition in its covering probability. It is argued that this phase transition dooms every learning algorithm to fail to identify a target concept lying close to it. However, in this paper we exhibit a counter-example which shows that this conclusion must be qualified in the general case. Mostly building on the work of Winston on near-misse examples, we show that, on the same set of problems, a top-down data-driven strategy can cross any plateau if near-misses are supplied in the training set, whereas they do not change the plateau profile and do not guide a generate-and-test strategy. We conclude that the location of the target concept with respect to the phase transition alone is not a reliable indication of the learning problem difficulty as previously thought.

Sequential pattern mining allows to discover temporal relationship between items within a database. The patterns can then be used to generate association rules. When the databases are very large, the execution speed and the memory usage of the mining algorithm become critical parameters. Previous research has focused on either one of the two parameters. In this paper, we present bitSPADE, a novel algorithm that combines the best features of SPAM, one of the fastest algorithm, and SPADE, one of the most memory efficient algorithm. Moreover, we introduce a new pruning strategy that enables bitSPADE to reach high performances. Experimental evaluations showed that bitSPADE ensures an efficient tradeoff between speed and memory usage by outperforming SPADE by both speed and memory usage factors more than 3.4 and SPAM by a memory consumption factor up to more than an order of magnitude.