Jesse Read

Clément Elliker, Mathis Le Bail, Clément Mantoux et al.

Train delay prediction is an important problem for both passengers and railway operators, yet progress in the field remains difficult to assess due to the lack of standardized datasets, prediction targets, and evaluation protocols. To address this gap, we introduce RIDE, an open dataset and benchmark for train delay prediction built at nationwide scale over the Belgian railway network. RIDE covers 94.5M train events, 3.6M journeys, and 35.7M weather records from 2023 to 2025. It is organized as a layered data pipeline from raw railway and weather sources to two public releases: a reusable intermediate relational dataset and model-ready benchmark datasets. The benchmark standardizes the prediction task and the training and testing data. It also provides a unified evaluation protocol that supports direct comparison across models. Using this framework, we provide the first comprehensive comparative evaluation of non-learning, statistical learning, and deep learning models. We show that learning-based methods clearly outperform non-learning models, with graph neural networks achieving the best mean performance, while the strongest learning-based models remain relatively close to one another. Beyond aggregate mean absolute error (MAE) and root mean squared error (RMSE), the framework also provides breakdowns by prediction horizon and delay change, enabling more detailed analysis of model behavior across forecasting regimes.

Eran Zvuloni, Jesse Read, Antônio H. Ribeiro et al.

Objective: Machine learning techniques have been used extensively for 12-lead electrocardiogram (ECG) analysis. For physiological time series, deep learning (DL) superiority to feature engineering (FE) approaches based on domain knowledge is still an open question. Moreover, it remains unclear whether combining DL with FE may improve performance. Methods: We considered three tasks intending to address these research gaps: cardiac arrhythmia diagnosis (multiclass-multilabel classification), atrial fibrillation risk prediction (binary classification), and age estimation (regression). We used an overall dataset of 2.3M 12-lead ECG recordings to train the following models for each task: i) a random forest taking the FE as input was trained as a classical machine learning approach; ii) an end-to-end DL model; and iii) a merged model of FE+DL. Results: FE yielded comparable results to DL while necessitating significantly less data for the two classification tasks and it was outperformed by DL for the regression task. For all tasks, merging FE with DL did not improve performance over DL alone. Conclusion: We found that for traditional 12-lead ECG based diagnosis tasks DL did not yield a meaningful improvement over FE, while it improved significantly the nontraditional regression task. We also found that combining FE with DL did not improve over DL alone which suggests that the FE were redundant with the features learned by DL. Significance: Our findings provides important recommendations on what machine learning strategy and data regime to chose with respect to the task at hand for the development of new machine learning models based on the 12-lead ECG.

Peng Yu, Chao Xu, Albert Bifet et al.

Decision trees are well-known due to their ease of interpretability. To improve accuracy, we need to grow deep trees or ensembles of trees. These are hard to interpret, offsetting their original benefits. Shapley values have recently become a popular way to explain the predictions of tree-based machine learning models. It provides a linear weighting to features independent of the tree structure. The rise in popularity is mainly due to TreeShap, which solves a general exponential complexity problem in polynomial time. Following extensive adoption in the industry, more efficient algorithms are required. This paper presents a more efficient and straightforward algorithm: Linear TreeShap. Like TreeShap, Linear TreeShap is exact and requires the same amount of memory.

Lucas Thil, Jesse Read, Rim Kaddah et al.

We present a novel method for learning interpretable representations of progressive time series, that is, data capturing irreversible state transitions such as degradation or task completion. Our approach uses a self-supervised contrastive objective to learn a low-dimensional latent space whose geometry is itself the interpretation: each observation becomes a point on a manifold anchored between two fixed orthogonal prototype vectors, and a trajectory becomes a path across that manifold. From this structure we read a latent compass, the polar coordinates (θ, r) of the latent vector, in which θ tracks the progression of the underlying state (e.g., from healthy to failed) and r identifies the active mode (e.g., the operating condition), without any proxy labels. We evaluate the approach against the state of the art on diverse domains, including industrial degradation, robotic tasks, and neural activity, validating three key capabilities: (1) end-state prediction, (2) multi-step forecasting, and (3) interpretable phase separation. Our method matches or improves over black-box counterparts on all of these while providing transparency about the underlying mechanisms. A simple linear regressor on top of the latent compass coordinates is competitive with deep architectures, direct quantitative evidence that the underlying state is encoded in a geometrically accessible form.

Lucas Thil, Jesse Read, Rim Kaddah et al.

Joint-Embedding Predictive Architectures (JEPAs) learn compact latent world models by predicting future embeddings, but no single coordinate of the latent is designated to encode task progression. We carve the JEPA latent into two orthogonal subspaces with disjoint roles: a low-dimensional progression subspace shaped by a cosine-margin triplet loss, and a high-dimensional content subspace regularised by the existing SIGReg objective of LeWM. We prove that the two anti-collapse forces act on disjoint coordinates, so they compose additively rather than competing on the same dimensions. Our method, SD-JEPA improves over the LeWM baseline on the majority of its control benchmarks at matched compute, and outperforms the strongest non-LeWM JEPA baseline on Push-T; a subspace-ablation falsifier confirms the split is the load-bearing ingredient. Beyond planning, the resulting 1-D angular progression coordinate functions as a scene-aware compass on the latent. It advances with task progress, regresses when the agent backtracks, and under controlled perturbations both spikes and relocalises to a semantically appropriate new task-phase sector, separating the moment of surprise from its meaning in a way that prediction-error scalars cannot. Three quantitative tests back this up: $|Δθ_t|$ outperforms the standard latent-prediction-error surprise at localising semantic events on 40 held-out cube episodes by up to +0.18 pooled AUROC (97.5% per-episode win rate at $\pm 1$-step tolerance); a within-episode linear probe across all four environments (40 episodes per env) shows the 8-dimensional progression subspace (4.2% of the latent) explains 72-95% of task-progress variance..

Peng Yu, Yike Chen, Chao Xu et al.

In the context of the Classification and Regression Trees (CART) algorithm, the efficient splitting of categorical features using standard criteria like GINI and Entropy is well-established. However, using the Mean Absolute Error (MAE) criterion for categorical features has traditionally relied on various numerical encoding methods. This paper demonstrates that unsupervised numerical encoding methods are not viable for the MAE criteria. Furthermore, we present a novel and efficient splitting algorithm that addresses the challenges of handling categorical features with the MAE criterion. Our findings underscore the limitations of existing approaches and offer a promising solution to enhance the handling of categorical data in CART algorithms.

Célia Wafa Ayad, Thomas Bonnier, Benjamin Bosch et al.

In spite of increased attention on explainable machine learning models, explaining multi-output predictions has not yet been extensively addressed. Methods that use Shapley values to attribute feature contributions to the decision making are one of the most popular approaches to explain local individual and global predictions. By considering each output separately in multi-output tasks, these methods fail to provide complete feature explanations. We propose Shapley Chains to overcome this issue by including label interdependencies in the explanation design process. Shapley Chains assign Shapley values as feature importance scores in multi-output classification using classifier chains, by separating the direct and indirect influence of these feature scores. Compared to existing methods, this approach allows to attribute a more complete feature contribution to the predictions of multi-output classification tasks. We provide a mechanism to distribute the hidden contributions of the outputs with respect to a given chaining order of these outputs. Moreover, we show how our approach can reveal indirect feature contributions missed by existing approaches. Shapley Chains help to emphasize the real learning factors in multi-output applications and allows a better understanding of the flow of information through output interdependencies in synthetic and real-world datasets.

Simo Alami C., Jérémie Decock, Rim Kaddah et al.

Non-Intrusive Load Monitoring (NILM) seeks to save energy by estimating individual appliance power usage from a single aggregate measurement. Deep neural networks have become increasingly popular in attempting to solve NILM problems. However most used models are used for Load Identification rather than online Source Separation. Among source separation models, most use a single-task learning approach in which a neural network is trained exclusively for each appliance. This strategy is computationally expensive and ignores the fact that multiple appliances can be active simultaneously and dependencies between them. The rest of models are not causal, which is important for real-time application. Inspired by Convtas-Net, a model for speech separation, we propose Conv-NILM-net, a fully convolutional framework for end-to-end NILM. Conv-NILM-net is a causal model for multi appliance source separation. Our model is tested on two real datasets REDD and UK-DALE and clearly outperforms the state of the art while keeping a significantly smaller size than the competing models.

Ekaterina Antonenko, Jesse Read

Missing values are a common problem in data science and machine learning. Removing instances with missing values can adversely affect the quality of further data analysis. This is exacerbated when there are relatively many more features than instances, and thus the proportion of affected instances is high. Such a scenario is common in many important domains, for example, single nucleotide polymorphism (SNP) datasets provide a large number of features over a genome for a relatively small number of individuals. To preserve as much information as possible prior to modeling, a rigorous imputation scheme is acutely needed. While Denoising Autoencoders is a state-of-the-art method for imputation in high-dimensional data, they still require enough complete cases to be trained on which is often not available in real-world problems. In this paper, we consider missing value imputation as a multi-label classification problem and propose Chains of Autoreplicative Random Forests. Using multi-label Random Forests instead of neural networks works well for low-sampled data as there are fewer parameters to optimize. Experiments on several SNP datasets show that our algorithm effectively imputes missing values based only on information from the dataset and exhibits better performance than standard algorithms that do not require any additional information. In this paper, the algorithm is implemented specifically for SNP data, but it can easily be adapted for other cases of missing value imputation.

Indre Zliobaite, Jesse Read

Machine learning from data streams is an active and growing research area. Research on learning from streaming data typically makes strict assumptions linked to computational resource constraints, including requirements for stream mining algorithms to inspect each instance not more than once and be ready to give a prediction at any time. Here we review the historical context of data streams research placing the common assumptions used in machine learning over data streams in their historical context.

Simo Alami. C, Fernando Llorente, Rim Kaddah et al.

Reinforcement Learning has drawn huge interest as a tool for solving optimal control problems. Solving a given problem (task or environment) involves converging towards an optimal policy. However, there might exist multiple optimal policies that can dramatically differ in their behaviour; for example, some may be faster than the others but at the expense of greater risk. We consider and study a distribution of optimal policies. We design a curiosity-augmented Metropolis algorithm (CAMEO), such that we can sample optimal policies, and such that these policies effectively adopt diverse behaviours, since this implies greater coverage of the different possible optimal policies. In experimental simulations we show that CAMEO indeed obtains policies that all solve classic control problems, and even in the challenging case of environments that provide sparse rewards. We further show that the different policies we sample present different risk profiles, corresponding to interesting practical applications in interpretability, and represents a first step towards learning the distribution of optimal policies itself.

Jesse Read, Indrė Žliobaitė

The literature on machine learning in the context of data streams is vast and growing. However, many of the defining assumptions regarding data-stream learning tasks are too strong to hold in practice, or are even contradictory such that they cannot be met in the contexts of supervised learning. Algorithms are chosen and designed based on criteria which are often not clearly stated, for problem settings not clearly defined, tested in unrealistic settings, and/or in isolation from related approaches in the wider literature. This puts into question the potential for real-world impact of many approaches conceived in such contexts, and risks propagating a misguided research focus. We propose to tackle these issues by reformulating the fundamental definitions and settings of supervised data-stream learning with regard to contemporary considerations of concept drift and temporal dependence; and we take a fresh look at what constitutes a supervised data-stream learning task, and a reconsideration of algorithms that may be applied to tackle such tasks. Through and in reflection of this formulation and overview, helped by an informal survey of industrial players dealing with real-world data streams, we provide recommendations. Our main emphasis is that learning from data streams does not impose a single-pass or online-learning approach, or any particular learning regime; and any constraints on memory and time are not specific to streaming. Meanwhile, there exist established techniques for dealing with temporal dependence and concept drift, in other areas of the literature. For the data streams community, we thus encourage a shift in research focus, from dealing with often-artificial constraints and assumptions on the learning mode, to issues such as robustness, privacy, and interpretability which are increasingly relevant to learning in data streams in academic and industrial settings.

Simo Alami. C, Rim Kaddah, Jesse Read

Clustering in high dimension spaces is a difficult task; the usual distance metrics may no longer be appropriate under the curse of dimensionality. Indeed, the choice of the metric is crucial, and it is highly dependent on the dataset characteristics. However a single metric could be used to correctly perform clustering on multiple datasets of different domains. We propose to do so, providing a framework for learning a transferable metric. We show that we can learn a metric on a labelled dataset, then apply it to cluster a different dataset, using an embedding space that characterises a desired clustering in the generic sense. We learn and test such metrics on several datasets of variable complexity (synthetic, MNIST, SVHN, omniglot) and achieve results competitive with the state-of-the-art while using only a small number of labelled training datasets and shallow networks.

Jesse Read

In multi-label learning, a particular case of multi-task learning where a single data point is associated with multiple target labels, it was widely assumed in the literature that, to obtain best accuracy, the dependence among the labels should be explicitly modeled. This premise led to a proliferation of methods offering techniques to learn and predict labels together, for example where the prediction for one label influences predictions for other labels. Even though it is now acknowledged that in many contexts a model of dependence is not required for optimal performance, such models continue to outperform independent models in some of those very contexts, suggesting alternative explanations for their performance beyond label dependence, which the literature is only recently beginning to unravel. Leveraging and extending recent discoveries, we turn the original premise of multi-label learning on its head, and approach the problem of joint-modeling specifically under the absence of any measurable dependence among task labels; for example, when task labels come from separate problem domains. We shift insights from this study towards building an approach for transfer learning that challenges the long-held assumption that transferability of tasks comes from measurements of similarity between the source and target domains or models. This allows us to design and test a method for transfer learning, which is model driven rather than purely data driven, and furthermore it is black box and model-agnostic (any base model class can be considered). We show that essentially we can create task-dependence based on source-model capacity. The results we obtain have important implications and provide clear directions for future work, both in the areas of multi-label and transfer learning.

Alexi Canesse, Benoît Goupil, Jesse Read et al.

Communication enables coordination in multi-agent reinforcement learning (MARL), but many real-world applications, e.g., search-and-rescue with drone swarms, operate under severe bandwidth constraints. Many communication architectures still expose a coupled bottleneck in which a shared latent representation is used for both policy execution and inter-agent communication. Consequently, reducing message size directly limits the policy's latent space, often leading to significant performance degradation. We address this with two contributions. First, we introduce $β$, a normalised per-agent bandwidth budget that unifies sparsity, rounds, and message dimension into a single comparable constraint. Second, we provide SLIM, a minimal architecture that decouples the communication pathway from the policy's latent representation, allowing us to isolate the effect of bandwidth from the effect of policy capacity while benefiting from in-step communication. We evaluate our method on several partially-observable MARL benchmarks, where communication is essential. Our approach achieves state-of-the-art performance and exhibits scalability and robustness under limited communication, with only marginal degradation as bandwidth is reduced.

Laurence A. F. Park, Jesse Read

A multi-label classifier estimates the binary label state (relevant vs irrelevant) for each of a set of concept labels, for any given instance. Probabilistic multi-label classifiers provide a predictive posterior distribution over all possible labelset combinations of such label states (the powerset of labels) from which we can provide the best estimate, simply by selecting the labelset corresponding to the largest expected accuracy, over that distribution. For example, in maximizing exact match accuracy, we provide the mode of the distribution. But how does this relate to the confidence we may have in such an estimate? Confidence is an important element of real-world applications of multi-label classifiers (as in machine learning in general) and is an important ingredient in explainability and interpretability. However, it is not obvious how to provide confidence in the multi-label context and relating to a particular accuracy metric, and nor is it clear how to provide a confidence which correlates well with the expected accuracy, which would be most valuable in real-world decision making. In this article we estimate the expected accuracy as a surrogate for confidence, for a given accuracy metric. We hypothesise that the expected accuracy can be estimated from the multi-label predictive distribution. We examine seven candidate functions for their ability to estimate expected accuracy from the predictive distribution. We found three of these to correlate to expected accuracy and are robust. Further, we determined that each candidate function can be used separately to estimate Hamming similarity, but a combination of the candidates was best for expected Jaccard index and exact match.

Sonal Sannigrahi, Jesse Read

Cross-Lingual Word Embeddings (CLWEs) are a key component to transfer linguistic information learnt from higher-resource settings into lower-resource ones. Recent research in cross-lingual representation learning has focused on offline mapping approaches due to their simplicity, computational efficacy, and ability to work with minimal parallel resources. However, they crucially depend on the assumption of embedding spaces being approximately isomorphic i.e. sharing similar geometric structure, which does not hold in practice, leading to poorer performance on low-resource and distant language pairs. In this paper, we introduce a framework to learn CLWEs, without assuming isometry, for low-resource pairs via joint exploitation of a related higher-resource language. In our work, we first pre-align the low-resource and related language embedding spaces using offline methods to mitigate the assumption of isometry. Following this, we use joint training methods to develops CLWEs for the related language and the target embed-ding space. Finally, we remap the pre-aligned low-resource space and the target space to generate the final CLWEs. We show consistent gains over current methods in both quality and degree of isomorphism, as measured by bilingual lexicon induction (BLI) and eigenvalue similarity respectively, across several language pairs: {Nepali, Finnish, Romanian, Gujarati, Hungarian}-English. Lastly, our analysis also points to the relatedness as well as the amount of related language data available as being key factors in determining the quality of embeddings achieved.

Jacob Montiel, Max Halford, Saulo Martiello Mastelini et al.

River is a machine learning library for dynamic data streams and continual learning. It provides multiple state-of-the-art learning methods, data generators/transformers, performance metrics and evaluators for different stream learning problems. It is the result from the merger of the two most popular packages for stream learning in Python: Creme and scikit-multiflow. River introduces a revamped architecture based on the lessons learnt from the seminal packages. River's ambition is to be the go-to library for doing machine learning on streaming data. Additionally, this open source package brings under the same umbrella a large community of practitioners and researchers. The source code is available at https://github.com/online-ml/river.

Jacob Montiel, Jesse Read, Albert Bifet et al.

Scikit-multiflow is a multi-output/multi-label and stream data mining framework for the Python programming language. Conceived to serve as a platform to encourage democratization of stream learning research, it provides multiple state of the art methods for stream learning, stream generators and evaluators. scikit-multiflow builds upon popular open source frameworks including scikit-learn, MOA and MEKA. Development follows the FOSS principles and quality is enforced by complying with PEP8 guidelines and using continuous integration and automatic testing. The source code is publicly available at https://github.com/scikit-multiflow/scikit-multiflow.

Antti Puurula, Jesse Read, Albert Bifet

Our winning submission to the 2014 Kaggle competition for Large Scale Hierarchical Text Classification (LSHTC) consists mostly of an ensemble of sparse generative models extending Multinomial Naive Bayes. The base-classifiers consist of hierarchically smoothed models combining document, label, and hierarchy level Multinomials, with feature pre-processing using variants of TF-IDF and BM25. Additional diversification is introduced by different types of folds and random search optimization for different measures. The ensemble algorithm optimizes macroFscore by predicting the documents for each label, instead of the usual prediction of labels per document. Scores for documents are predicted by weighted voting of base-classifier outputs with a variant of Feature-Weighted Linear Stacking. The number of documents per label is chosen using label priors and thresholding of vote scores. This document describes the models and software used to build our solution. Reproducing the results for our solution can be done by running the scripts included in the Kaggle package. A package omitting precomputed result files is also distributed. All code is open source, released under GNU GPL 2.0, and GPL 3.0 for Weka and Meka dependencies.

Lucas Thil, Jesse Read, Rim Kaddah et al.

Accurate remaining useful life (RUL) prediction hinges on the quality of health indicators (HIs), yet existing methods often fail to disentangle complex degradation mechanisms in multi-sensor systems or quantify uncertainty in HI reliability. This paper introduces a novel framework for HI construction, advancing three key contributions. First, we adapt Reconstruction along Projected Pathways (RaPP) as a health indicator (HI) for RUL prediction for the first time, showing that it outperforms traditional reconstruction error metrics. Second, we show that augmenting RaPP-derived HIs with aleatoric and epistemic uncertainty quantification (UQ) via Monte Carlo dropout and probabilistic latent spaces- significantly improves RUL-prediction robustness. Third, and most critically, we propose indicator groups, a paradigm that isolates sensor subsets to model system-specific degradations, giving rise to our novel method, I-GLIDE which enables interpretable, mechanism-specific diagnostics. Evaluated on data sourced from aerospace and manufacturing systems, our approach achieves marked improvements in accuracy and generalizability compared to state-of-the-art HI methods while providing actionable insights into system failure pathways. This work bridges the gap between anomaly detection and prognostics, offering a principled framework for uncertainty-aware degradation modeling in complex systems.

Milad Leyli-Abadi, Lucas Thil, Sebastien Razakarivony et al.

Estimating the health state of turbofan engines is a challenging ill-posed inverse problem, hindered by sparse sensing and complex nonlinear thermodynamics. Research in this area remains fragmented, with comparisons limited by the use of unrealistic datasets and insufficient exploration of the exploitation of temporal information. This work investigates how to recover component-level health indicators from operational sensor data under realistic degradation and maintenance patterns. To support this study, we introduce a new dataset that incorporates industry-oriented complexities such as maintenance events and usage changes. Using this dataset, we establish an initial benchmark that compares steady-state and nonstationary data-driven models, and Bayesian filters, classic families of methods used to solve this problem. In addition to this benchmark, we introduce self-supervised learning (SSL) approaches that learn latent representations without access to true health labels, a scenario reflective of real-world operational constraints. By comparing the downstream estimation performance of these unsupervised representations against the direct prediction baselines, we establish a practical lower bound on the difficulty of solving this inverse problem. Our results reveal that traditional filters remain strong baselines, while SSL methods reveal the intrinsic complexity of health estimation and highlight the need for more advanced and interpretable inference strategies. For reproducibility, both the generated dataset and the implementation used in this work are made accessible.

Ayman Chaouki, Jesse Read, Albert Bifet

Decision Trees are prominent prediction models for interpretable Machine Learning. They have been thoroughly researched, mostly in the batch setting with a fixed labelled dataset, leading to popular algorithms such as C4.5, ID3 and CART. Unfortunately, these methods are of heuristic nature, they rely on greedy splits offering no guarantees of global optimality and often leading to unnecessarily complex and hard-to-interpret Decision Trees. Recent breakthroughs addressed this suboptimality issue in the batch setting, but no such work has considered the online setting with data arriving in a stream. To this end, we devise a new Monte Carlo Tree Search algorithm, Thompson Sampling Decision Trees (TSDT), able to produce optimal Decision Trees in an online setting. We analyse our algorithm and prove its almost sure convergence to the optimal tree. Furthermore, we conduct extensive experiments to validate our findings empirically. The proposed TSDT outperforms existing algorithms on several benchmarks, all while presenting the practical advantage of being tailored to the online setting.

Clément Elliker, Jesse Read, Sonia Vanier et al.

Reliable prediction of train delays is essential for enhancing the robustness and efficiency of railway transportation systems. In this work, we reframe delay forecasting as a stochastic simulation task, modeling state-transition dynamics through imitation learning. We introduce Drift-Corrected Imitation Learning (DCIL), a novel self-supervised algorithm that extends DAgger by incorporating distance-based drift correction, thereby mitigating covariate shift during rollouts without requiring access to an external oracle or adversarial schemes. Our approach synthesizes the dynamical fidelity of event-driven models with the representational capacity of data-driven methods, enabling uncertainty-aware forecasting via Monte Carlo simulation. We evaluate DCIL using a comprehensive real-world dataset from \textsc{Infrabel}, the Belgian railway infrastructure manager, which encompasses over three million train movements. Our results, focused on predictions up to 30 minutes ahead, demonstrate superior predictive performance of DCIL over traditional regression models and behavioral cloning on deep learning architectures, highlighting its effectiveness in capturing the sequential and uncertain nature of delay propagation in large-scale networks.

Laurence A. F. Park, Jesse Read

Multilabel data should be analysed for label dependence before applying multilabel models. Independence between multilabel data labels cannot be measured directly from the label values due to their dependence on the set of covariates $\vec{x}$, but can be measured by examining the conditional label covariance using a multivariate Probit model. Unfortunately, the multivariate Probit model provides an estimate of its copula covariance, and so might not be reliable in estimating constant covariance and dependent covariance. In this article, we compare three models (Multivariate Probit, Multivariate Bernoulli and Staged Logit) for estimating the constant and dependent multilabel conditional label covariance. We provide an experiment that allows us to observe each model's measurement of conditional covariance. We found that all models measure constant and dependent covariance equally well, depending on the strength of the covariance, but the models all falsely detect that dependent covariance is present for data where constant covariance is present. Of the three models, the Multivariate Probit model had the lowest error rate.

Simo Alami C., Rim Kaddah, Jesse Read et al.

We introduce a new architecture for Distributional Reinforcement Learning (DistRL) that models return distributions using normalizing flows. This approach enables flexible, unbounded support for return distributions, in contrast to categorical approaches like C51 that rely on fixed or bounded representations. It also offers richer modeling capacity to capture multi-modality, skewness, and tail behavior than quantile based approaches. Our method is significantly more parameter-efficient than categorical approaches. Standard metrics used to train existing models like KL divergence or Wasserstein distance either are scale insensitive or have biased sample gradients, especially when return supports do not overlap. To address this, we propose a novel surrogate for the Cramèr distance, that is geometry-aware and computable directly from the return distribution's PDF, avoiding the costly CDF computation. We test our model on the ATARI-5 sub-benchmark and show that our approach outperforms PDF based models while remaining competitive with quantile based methods.

Célia Wafa Ayad, Thomas Bonnier, Benjamin Bosch et al.

In order to ensure the reliability of the explanations of machine learning models, it is crucial to establish their advantages and limits and in which case each of these methods outperform. However, the current understanding of when and how each method of explanation can be used is insufficient. To fill this gap, we perform a comprehensive empirical evaluation by synthesizing multiple datasets with the desired properties. Our main objective is to assess the quality of feature importance estimates provided by local explanation methods, which are used to explain predictions made by decision tree-based models. By analyzing the results obtained from synthetic datasets as well as publicly available binary classification datasets, we observe notable disparities in the magnitude and sign of the feature importance estimates generated by these methods. Moreover, we find that these estimates are sensitive to specific properties present in the data. Although some model hyper-parameters do not significantly influence feature importance assignment, it is important to recognize that each method of explanation has limitations in specific contexts. Our assessment highlights these limitations and provides valuable insight into the suitability and reliability of different explanatory methods in various scenarios.

Elaine Cecília Gatto, Felipe Nakano Kenji, Jesse Read et al.

Multi-label classification is a type of supervised machine learning that can simultaneously assign multiple labels to an instance. To solve this task, some methods divide the original problem into several sub-problems (local approach), others learn all labels at once (global approach), and others combine several classifiers (ensemble approach). Regardless of the approach used, exploring and learning label correlations is important to improve the classifier predictions. Ensemble of Classifier Chains (ECC) is a well-known multi-label method that considers label correlations and can achieve good overall performance on several multi-label datasets and evaluation measures. However, one of the challenges when working with ECC is the high dimensionality of the label space, which can impose limitations for fully-cascaded chains as the complexity increases regarding feature space expansion. To improve classifier chains, we propose a method to chain disjoint correlated label clusters obtained by applying a partition method in the label space. During the training phase, the ground truth labels of each cluster are used as new features for all of the following clusters. During the test phase, the predicted labels of clusters are used as new features for all the following clusters. Our proposal, called Label Cluster Chains for Multi-Label Classification (LCC-ML), uses multi-label Random Forests as base classifiers in each cluster, combining their predictions to obtain a final multi-label classification. Our proposal obtained better results compared to the original ECC. This shows that learning and chaining disjoint correlated label clusters can better explore and learn label correlations.

Ayman Chaouki, Jesse Read, Albert Bifet

Decision Tree (DT) Learning is a fundamental problem in Interpretable Machine Learning, yet it poses a formidable optimisation challenge. Practical algorithms have recently emerged, primarily leveraging Dynamic Programming and Branch & Bound. However, most of these approaches rely on a Depth-First-Search strategy, which is inefficient when searching for DTs at high depths and requires the definition of a maximum depth hyperparameter. Best-First-Search was also employed by other methods to circumvent these issues. The downside of this strategy is its higher memory consumption, as such, it has to be designed in a fully efficient manner that takes full advantage of the problem's structure. We formulate the problem within an AND/OR graph search framework and we solve it with a novel AO*-type algorithm called Branches. We prove both optimality and complexity guarantees for Branches and we show that it is more efficient than the state of the art theoretically and on a variety of experiments. Furthermore, Branches supports non-binary features unlike the other methods, we show that this property can further induce larger gains in computational efficiency.

Alban Puech, Jesse Read

Yaw misalignment, measured as the difference between the wind direction and the nacelle position of a wind turbine, has consequences on the power output, the safety and the lifetime of the turbine and its wind park as a whole. We use reinforcement learning to develop a yaw control agent to minimise yaw misalignment and optimally reallocate yaw resources, prioritising high-speed segments, while keeping yaw usage low. To achieve this, we carefully crafted and tested the reward metric to trade-off yaw usage versus yaw alignment (as proportional to power production), and created a novel simulator (environment) based on real-world wind logs obtained from a REpower MM82 2MW turbine. The resulting algorithm decreased the yaw misalignment by 5.5% and 11.2% on two simulations of 2.7 hours each, compared to the conventional active yaw control algorithm. The average net energy gain obtained was 0.31% and 0.33% respectively, compared to the traditional yaw control algorithm. On a single 2MW turbine, this amounts to a 1.5k-2.5k euros annual gain, which sums up to very significant profits over an entire wind park.

Fernando Llorente, Luca Martino, Jesse Read et al.

In this work, we analyze the noisy importance sampling (IS), i.e., IS working with noisy evaluations of the target density. We present the general framework and derive optimal proposal densities for noisy IS estimators. The optimal proposals incorporate the information of the variance of the noisy realizations, proposing points in regions where the noise power is higher. We also compare the use of the optimal proposals with previous optimality approaches considered in a noisy IS framework.

Heitor Murilo Gomes, Maciej Grzenda, Rodrigo Mello et al.

Unlabelled data appear in many domains and are particularly relevant to streaming applications, where even though data is abundant, labelled data is rare. To address the learning problems associated with such data, one can ignore the unlabelled data and focus only on the labelled data (supervised learning); use the labelled data and attempt to leverage the unlabelled data (semi-supervised learning); or assume some labels will be available on request (active learning). The first approach is the simplest, yet the amount of labelled data available will limit the predictive performance. The second relies on finding and exploiting the underlying characteristics of the data distribution. The third depends on an external agent to provide the required labels in a timely fashion. This survey pays special attention to methods that leverage unlabelled data in a semi-supervised setting. We also discuss the delayed labelling issue, which impacts both fully supervised and semi-supervised methods. We propose a unified problem setting, discuss the learning guarantees and existing methods, explain the differences between related problem settings. Finally, we review the current benchmarking practices and propose adaptations to enhance them.

Luca Martino, Jesse Read

The expressive power of Bayesian kernel-based methods has led them to become an important tool across many different facets of artificial intelligence, and useful to a plethora of modern application domains, providing both power and interpretability via uncertainty analysis. This article introduces and discusses two methods which straddle the areas of probabilistic Bayesian schemes and kernel methods for regression: Gaussian Processes and Relevance Vector Machines. Our focus is on developing a common framework with which to view these methods, via intermediate methods a probabilistic version of the well-known kernel ridge regression, and drawing connections among them, via dual formulations, and discussion of their application in the context of major tasks: regression, smoothing, interpolation, and filtering. Overall, we provide understanding of the mathematical concepts behind these models, and we summarize and discuss in depth different interpretations and highlight the relationship to other methods, such as linear kernel smoothers, Kalman filtering and Fourier approximations. Throughout, we provide numerous figures to promote understanding, and we make numerous recommendations to practitioners. Benefits and drawbacks of the different techniques are highlighted. To our knowledge, this is the most in-depth study of its kind to date focused on these two methods, and will be relevant to theoretical understanding and practitioners throughout the domains of data-science, signal processing, machine learning, and artificial intelligence in general.

Kayo Yin, Jesse Read

Sign Language Translation (SLT) first uses a Sign Language Recognition (SLR) system to extract sign language glosses from videos. Then, a translation system generates spoken language translations from the sign language glosses. This paper focuses on the translation system and introduces the STMC-Transformer which improves on the current state-of-the-art by over 5 and 7 BLEU respectively on gloss-to-text and video-to-text translation of the PHOENIX-Weather 2014T dataset. On the ASLG-PC12 corpus, we report an increase of over 16 BLEU. We also demonstrate the problem in current methods that rely on gloss supervision. The video-to-text translation of our STMC-Transformer outperforms translation of GT glosses. This contradicts previous claims that GT gloss translation acts as an upper bound for SLT performance and reveals that glosses are an inefficient representation of sign language. For future SLT research, we therefore suggest an end-to-end training of the recognition and translation models, or using a different sign language annotation scheme.

Jesse Read, Bernhard Pfahringer, Geoff Holmes et al.

The family of methods collectively known as classifier chains has become a popular approach to multi-label learning problems. This approach involves linking together off-the-shelf binary classifiers in a chain structure, such that class label predictions become features for other classifiers. Such methods have proved flexible and effective and have obtained state-of-the-art empirical performance across many datasets and multi-label evaluation metrics. This performance led to further studies of how exactly it works, and how it could be improved, and in the recent decade numerous studies have explored classifier chains mechanisms on a theoretical level, and many improvements have been made to the training and inference procedures, such that this method remains among the state-of-the-art options for multi-label learning. Given this past and ongoing interest, which covers a broad range of applications and research themes, the goal of this work is to provide a review of classifier chains, a survey of the techniques and extensions provided in the literature, as well as perspectives for this approach in the domain of multi-label classification in the future. We conclude positively, with a number of recommendations for researchers and practitioners, as well as outlining a number of areas for future research.

Jesse Read, Luca Martino

A large number and diversity of techniques have been offered in the literature in recent years for solving multi-label classification tasks, including classifier chains where predictions are cascaded to other models as additional features. The idea of extending this chaining methodology to multi-output regression has already been suggested and trialed: regressor chains. However, this has so-far been limited to greedy inference and has provided relatively poor results compared to individual models, and of limited applicability. In this paper we identify and discuss the main limitations, including an analysis of different base models, loss functions, explainability, and other desiderata of real-world applications. To overcome the identified limitations we study and develop methods for regressor chains. In particular we present a sequential Monte Carlo scheme in the framework of a probabilistic regressor chain, and we show it can be effective, flexible and useful in several types of data. We place regressor chains in context in general terms of multi-output learning with continuous outputs, and in doing this shed additional light on classifier chains.

Jesse Read

Learning from data streams is an increasingly important topic in data mining, machine learning, and artificial intelligence in general. A major focus in the data stream literature is on designing methods that can deal with concept drift, a challenge where the generating distribution changes over time. A general assumption in most of this literature is that instances are independently distributed in the stream. In this work we show that, in the context of concept drift, this assumption is contradictory, and that the presence of concept drift necessarily implies temporal dependence; and thus some form of time series. This has important implications on model design and deployment. We explore and highlight the these implications, and show that Hoeffding-tree based ensembles, which are very popular for learning in streams, are not naturally suited to learning \emph{within} drift; and can perform in this scenario only at significant computational cost of destructive adaptation. On the other hand, we develop and parameterize gradient-descent methods and demonstrate how they can perform \emph{continuous} adaptation with no explicit drift-detection mechanism, offering major advantages in terms of accuracy and efficiency. As a consequence of our theoretical discussion and empirical observations, we outline a number of recommendations for deploying methods in concept-drifting streams.

Antoine J. -P. Tixier, Maria-Evgenia G. Rossi, Fragkiskos D. Malliaros et al.

Some of the most effective influential spreader detection algorithms are unstable to small perturbations of the network structure. Inspired by bagging in Machine Learning, we propose the first Perturb and Combine (P&C) procedure for networks. It (1) creates many perturbed versions of a given graph, (2) applies a node scoring function separately to each graph, and (3) combines the results. Experiments conducted on real-world networks of various sizes with the k-core, generalized k-core, and PageRank algorithms reveal that P&C brings substantial improvements. Moreover, this performance boost can be obtained at almost no extra cost through parallelization. Finally, a bias-variance analysis suggests that P&C works mainly by reducing bias, and that therefore, it should be capable of improving the performance of all vertex scoring functions, including stable ones.

Jesse Read, Luca Martino, Jaakko Hollmén

The number of methods available for classification of multi-label data has increased rapidly over recent years, yet relatively few links have been made with the related task of classification of sequential data. If labels indices are considered as time indices, the problems can often be seen as equivalent. In this paper we detect and elaborate on connections between multi-label methods and Markovian models, and study the suitability of multi-label methods for prediction in sequential data. From this study we draw upon the most suitable techniques from the area and develop two novel competitive approaches which can be applied to either kind of data. We carry out an empirical evaluation investigating performance on real-world sequential-prediction tasks: electricity demand, and route prediction. As well as showing that several popular multi-label algorithms are in fact easily applicable to sequencing tasks, our novel approaches, which benefit from a unified view of these areas, prove very competitive against established methods.

Diego Marrón, Jesse Read, Albert Bifet et al.

Big Data streams are being generated in a faster, bigger, and more commonplace. In this scenario, Hoeffding Trees are an established method for classification. Several extensions exist, including high-performing ensemble setups such as online and leveraging bagging. Also, $k$-nearest neighbors is a popular choice, with most extensions dealing with the inherent performance limitations over a potentially-infinite stream. At the same time, gradient descent methods are becoming increasingly popular, owing in part to the successes of deep learning. Although deep neural networks can learn incrementally, they have so far proved too sensitive to hyper-parameter options and initial conditions to be considered an effective `off-the-shelf' data-streams solution. In this work, we look at combinations of Hoeffding-trees, nearest neighbour, and gradient descent methods with a streaming preprocessing approach in the form of a random feature functions filter for additional predictive power. We further extend the investigation to implementing methods on GPUs, which we test on some large real-world datasets, and show the benefits of using GPUs for data-stream learning due to their high scalability. Our empirical evaluation yields positive results for the novel approaches that we experiment with, highlighting important issues, and shed light on promising future directions in approaches to data-stream classification.

Jesse Read, Jaakko Hollmén

Competitive methods for multi-label classification typically invest in learning labels together. To do so in a beneficial way, analysis of label dependence is often seen as a fundamental step, separate and prior to constructing a classifier. Some methods invest up to hundreds of times more computational effort in building dependency models, than training the final classifier itself. We extend some recent discussion in the literature and provide a deeper analysis, namely, developing the view that label dependence is often introduced by an inadequate base classifier, rather than being inherent to the data or underlying concept; showing how even an exhaustive analysis of label dependence may not lead to an optimal classification structure. Viewing labels as additional features (a transformation of the input), we create neural-network inspired novel methods that remove the emphasis of a prior dependency structure. Our methods have an important advantage particular to multi-label data: they leverage labels to create effective units in middle layers, rather than learning these units from scratch in an unsupervised fashion with gradient-based methods. Results are promising. The methods we propose perform competitively, and also have very important qualities of scalability.

Jesse Read, Fernando Perez-Cruz

In multi-label classification, the main focus has been to develop ways of learning the underlying dependencies between labels, and to take advantage of this at classification time. Developing better feature-space representations has been predominantly employed to reduce complexity, e.g., by eliminating non-helpful feature attributes from the input space prior to (or during) training. This is an important task, since many multi-label methods typically create many different copies or views of the same input data as they transform it, and considerable memory can be saved by taking advantage of redundancy. In this paper, we show that a proper development of the feature space can make labels less interdependent and easier to model and predict at inference time. For this task we use a deep learning approach with restricted Boltzmann machines. We present a deep network that, in an empirical evaluation, outperforms a number of competitive methods from the literature