Mach. Learn. Knowl. Extr., Volume 4, Issue 1 (March 2022) – 13 articles

Reinforcement learning (RL) has become a highly successful framework for learning in Markov decision processes (MDP). Due to the adoption of RL in realistic and complex environments, solution robustness becomes an increasingly important aspect of RL deployment. Nevertheless, current RL algorithms struggle with robustness to uncertainty, disturbances, or structural changes in the environment. We survey the literature on robust approaches to reinforcement learning and categorize these methods in four different ways: (i) Transition robust designs account for uncertainties in the system dynamics by manipulating the transition probabilities between states; (ii) Disturbance robust designs leverage external forces to model uncertainty in the system behavior; (iii) Action robust designs redirect transitions of the system by corrupting an agent’s output; (iv) Observation robust designs exploit or distort the perceived system state of the policy. Each of these robust designs alters a different aspect of the MDP. Additionally, we address the connection of robustness to the risk-based and entropy-regularized RL formulations. The resulting survey covers all fundamental concepts underlying the approaches to robust reinforcement learning and their recent advances. Full article

Biomedical Named-Entity Recognition (BioNER) has become an essential part of text mining due to the continuously increasing digital archives of biological and medical articles. While there are many well-performing BioNER tools for entities such as genes, proteins, diseases or species, there is very little research into food and dietary constituent named-entity recognition. For this reason, in this paper, we study seven BioNER models for food and dietary constituents recognition. Specifically, we study a dictionary-based model, a conditional random fields (CRF) model and a new hybrid model, called FooDCoNER (Food and Dietary Constituents Named-Entity Recognition), which we introduce combining the former two models. In addition, we study deep language models including BERT, BioBERT, RoBERTa and ELECTRA. As a result, we find that FooDCoNER does not only lead to the overall best results, comparable with the deep language models, but FooDCoNER is also much more efficient with respect to run time and sample size requirements of the training data. The latter has been identified via the study of learning curves. Overall, our results not only provide a new tool for food and dietary constituent NER but also shed light on the difference between classical machine learning models and recent deep language models. Full article

Machine learning (ML) models are increasingly being used for high-stake applications that can greatly impact people’s lives. Sometimes, these models can be biased toward certain social groups on the basis of race, gender, or ethnicity. Many prior works have attempted to mitigate this “model discrimination” by updating the training data (pre-processing), altering the model learning process (in-processing), or manipulating the model output (post-processing). However, more work can be done in extending this situation to intersectional fairness, where we consider multiple sensitive parameters (e.g., race) and sensitive options (e.g., black or white), thus allowing for greater real-world usability. Prior work in fairness has also suffered from an accuracy–fairness trade-off that prevents both accuracy and fairness from being high. Moreover, the previous literature has not clearly presented holistic fairness metrics that work with intersectional fairness. In this paper, we address all three of these problems by (a) creating a bias mitigation technique called DualFair and (b) developing a new fairness metric (i.e., AWI, a measure of bias of an algorithm based upon inconsistent counterfactual predictions) that can handle intersectional fairness. Lastly, we test our novel mitigation method using a comprehensive U.S. mortgage lending dataset and show that our classifier, or fair loan predictor, obtains relatively high fairness and accuracy metrics. Full article

Arrays of Cholesteric Spherical Reflectors (CSRs), microscopic cholesteric liquid crystals in a spherical shape, have been argued to become a game-changing technology in anti-counterfeiting. Used to build identifiable tags or coating, called CSR IDs, they can supply objects with unclonable fingerprint-like characteristics, making it possible to authenticate objects. In a previous study, we have shown how to extract minutiæ from CSR IDs. In this journal version, we build on that previous research, consolidate the methodology, and test it over CSR IDs obtained by different production processes. We measure the robustness and reliability of our procedure on large and variegate sets of CSR IDs’ images taken with a professional microscope (Laboratory Data set) and with a microscope that could be used in a realistic scenario (Realistic Data set). We measure intra-distance and interdistance, proving that we can distinguish images coming from the same CSR ID from images of different CSR IDs. However, without surprise, images in Laboratory Data set have an intra-distance that on average is less, and with less variance, than the intra-distance between responses from Realistic Data set. With this evidence, we discuss a few requirements for an anti-counterfeiting technology based on CSRs. Full article

Reinforcement learning (RL) allows an agent to solve sequential decision-making problems by interacting with an environment in a trial-and-error fashion. When these environments are very complex, pure random exploration of possible solutions often fails, or is very sample inefficient, requiring an unreasonable amount of interaction with the environment. Hierarchical reinforcement learning (HRL) utilizes forms of temporal- and state-abstractions in order to tackle these challenges, while simultaneously paving the road for behavior reuse and increased interpretability of RL systems. In this survey paper we first introduce a selection of problem-specific approaches, which provided insight in how to utilize often handcrafted abstractions in specific task settings. We then introduce the Options framework, which provides a more generic approach, allowing abstractions to be discovered and learned semi-automatically. Afterwards we introduce the goal-conditional approach, which allows sub-behaviors to be embedded in a continuous space. In order to further advance the development of HRL agents, capable of simultaneously learning abstractions and how to use them, solely from interaction with complex high dimensional environments, we also identify a set of promising research directions. Full article

Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model. Full article

Datasets with thousands of features represent a challenge for many of the existing learning methods because of the well known curse of dimensionality. Not only that, but the presence of irrelevant and redundant features on any dataset can degrade the performance of any model where training and inference is attempted. In addition, in large datasets, the manual management of features tends to be impractical. Therefore, the increasing interest of developing frameworks for the automatic discovery and removal of useless features through the literature of Machine Learning. This is the reason why, in this paper, we propose a novel framework for selecting relevant features in supervised datasets based on a cascade of methods where speed and precision are in mind. This framework consists of a novel combination of Approximated and Simulate Annealing versions of the Maximal Information Coefficient (MIC) to generalize the simple linear relation between features. This process is performed in a series of steps by applying the MIC algorithms and cutoff strategies to remove irrelevant and redundant features. The framework is also designed to achieve a balance between accuracy and speed. To test the performance of the proposed framework, a series of experiments are conducted on a large battery of datasets from SPECTF Heart to Sonar data. The results show the balance of accuracy and speed that the proposed framework can achieve. Full article

To encourage proper employee scheduling for managing crew load, restaurants need accurate sales forecasting. This paper proposes a case study on many machine learning (ML) models using real-world sales data from a mid-sized restaurant. Trendy recurrent neural network (RNN) models are included for direct comparison to many methods. To test the effects of trend and seasonality, we generate three different datasets to train our models with and to compare our results. To aid in forecasting, we engineer many features and demonstrate good methods to select an optimal sub-set of highly correlated features. We compare the models based on their performance for forecasting time steps of one-day and one-week over a curated test dataset. The best results seen in one-day forecasting come from linear models with a sMAPE of only 19.6%. Two RNN models, LSTM and TFT, and ensemble models also performed well with errors less than 20%. When forecasting one-week, non-RNN models performed poorly, giving results worse than 20% error. RNN models extended better with good sMAPE scores giving 19.5% in the best result. The RNN models performed worse overall on datasets with trend and seasonality removed, however many simpler ML models performed well when linearly separating each training instance. Full article

Data-intensive workloads and applications, such as machine learning (ML), are fundamentally limited by traditional computing systems based on the von-Neumann architecture. As data movement operations and energy consumption become key bottlenecks in the design of computing systems, the interest in unconventional approaches such as Near-Data Processing (NDP), machine learning, and especially neural network (NN)-based accelerators has grown significantly. Emerging memory technologies, such as ReRAM and 3D-stacked, are promising for efficiently architecting NDP-based accelerators for NN due to their capabilities to work as both high-density/low-energy storage and in/near-memory computation/search engine. In this paper, we present a survey of techniques for designing NDP architectures for NN. By classifying the techniques based on the memory technology employed, we underscore their similarities and differences. Finally, we discuss open challenges and future perspectives that need to be explored in order to improve and extend the adoption of NDP architectures for future computing platforms. This paper will be valuable for computer architects, chip designers, and researchers in the area of machine learning. Full article

The amount of information preserved in Portuguese archives has increased over the years. These documents represent a national heritage of high importance, as they portray the country’s history. Currently, most Portuguese archives have made their finding aids available to the public in digital format, however, these data do not have any annotation, so it is not always easy to analyze their content. In this work, Named Entity Recognition solutions were created that allow the identification and classification of several named entities from the archival finding aids. These named entities translate into crucial information about their context and, with high confidence results, they can be used for several purposes, for example, the creation of smart browsing tools by using entity linking and record linking techniques. In order to achieve high result scores, we annotated several corpora to train our own Machine Learning algorithms in this context domain. We also used different architectures, such as CNNs, LSTMs, and Maximum Entropy models. Finally, all the created datasets and ML models were made available to the public with a developed web platform, NER@DI. Full article

Transfer learning is a machine learning technique that uses previously acquired knowledge from a source domain to enhance learning in a target domain by reusing learned weights. This technique is ubiquitous because of its great advantages in achieving high performance while saving training time, memory, and effort in network design. In this paper, we investigate how to select the best pre-trained model that meets the target domain requirements for image classification tasks. In our study, we refined the output layers and general network parameters to apply the knowledge of eleven image processing models, pre-trained on ImageNet, to five different target domain datasets. We measured the accuracy, accuracy density, training time, and model size to evaluate the pre-trained models both in training sessions in one episode and with ten episodes. Full article

Hybrid simulation is a method used to investigate the dynamic response of a system subjected to a realistic loading scenario. The system under consideration is divided into multiple individual substructures, out of which one or more are tested physically, whereas the remaining are simulated numerically. The coupling of all substructures forms the so-called hybrid model. Although hybrid simulation is extensively used across various engineering disciplines, it is often the case that the hybrid model and related excitation are conceived as being deterministic. However, associated uncertainties are present, whilst simulation deviation, due to their presence, could be significant. In this regard, global sensitivity analysis based on Sobol’ indices can be used to determine the sensitivity of the hybrid model response due to the presence of the associated uncertainties. Nonetheless, estimation of the Sobol’ sensitivity indices requires an unaffordable amount of hybrid simulation evaluations. Therefore, surrogate modeling techniques using machine learning data-driven regression are utilized to alleviate this burden. This study extends the current global sensitivity analysis practices in hybrid simulation by employing various different surrogate modeling methodologies as well as providing comparative results. In particular, polynomial chaos expansion, Kriging and polynomial chaos Kriging are used. A case study encompassing a virtual hybrid model is employed, and hybrid model response quantities of interest are selected. Their respective surrogates are developed, using all three aforementioned techniques. The Sobol’ indices obtained utilizing each examined surrogate are compared with each other, and the results highlight potential deviations when different surrogates are used. Full article