To Örebro University

Recent advances in treatment effect estimation (TEE) leverage deep neural networks to learn latent representations that balance treatment and control groups. However, despite their empirical success, we know remarkably little about the learned representations themselves—what structure they capture, why they lead to improved performance, and how far we can trust them. Existing evaluations focus almost exclusively on causal outcome metrics such as PEHE or ATE bias, leaving the representational mechanisms largely unexplored. This lack of understanding limits our ability to explain why models perform well, to identify when they fail, and to design more interpretable or trustworthy causal systems. In this work, we present a comprehensive analysis for representation-level evaluation of TEE methods. We benchmark CFR, DRCFR, TEDVAE and modern disentanglement-driven models using a unified suite of quantitative and graphical measures. Quantitatively, we assess AUC(T | Z) for confounding separability, Mutual Information (MI) retention for informativeness, MMD2 and SW1 for balance, CH-ratio for compactness, and Participation Ratio (PR) for effective dimensionality. Graphically, we analyze bias removal through embedding visualizations, kNN-mixing histograms, covariance heatmaps, and outcome-smoothness histograms—providing an interpretable view of latent geometry. Our analysis shows that modern VAE-based disentangling models yield more structured, balanced, and informative representations, which also correlate strongly with causal reliability. We provide a practical way to see inside TEE representations—revealing hidden biases, interpreting latent behavior, and understanding when conventional embedding properties (such as neighborhood preservation) may actually hinder bias mitigation. Overall, our findings emphasize that evaluating how models learn is essential to understanding why they perform well, paving the way toward more interpretable and trustworthy causal inference.

Estimating causal effects from observational data is challenging due to selection bias, which leads to imbalanced covariate distributions across treatment groups. Propensity score-based weighting methods are widely used to address this issue by reweighting samples to simulate a randomized controlled trial (RCT). However, the effectiveness of these methods heavily depends on the observed data and the accuracy of the propensity score estimator. For example, inverse propensity weighting (IPW) assigns weights based on the inverse of the propensity score, which can lead to instable weights when propensity scores have high variance-either due to data or model misspecification-ultimately degrading the ability of handling selection bias and treatment effect estimation. To overcome these limitations, we propose Deconfounding Factor Weighting (DFW), a novel propensity score-based approach that leverages the deconfounding factor-to construct stable and effective sample weights. DFW prioritizes less confounded samples while mitigating the influence of highly confounded ones, producing a pseudopopulation that better approximates a RCT. Our approach ensures bounded weights, lower variance, and improved covariate balance.While DFW is formulated for binary treatments, it naturally extends to multi-treatment settings, as the deconfounding factor is computed based on the estimated probability of the treatment actually received by each sample. Through extensive experiments on real-world benchmark and synthetic datasets, we demonstrate that DFW outperforms existing methods, including IPW and CBPS, in both covariate balancing and treatment effect estimation.

Machine learning predictive models rely on data to make predictions for new input data. However, accurate predictions are not always the end goal; practitioners often aim to make informed decisions through optimization problems (OPs) based on these predictions. While the idea that better predictions lead to better decisions was widely accepted, the latest literature highlights that even small inaccuracies in predictions can lead to poor decisions depending on the structure of the OP. Therefore, recent research has been focused on end-to-end learning approaches that directly improve decision quality without considering prediction accuracy when solving data-driven OPs. Some of these end-to-end learning approaches are mainly called "predict-and-optimize" (PaO), and they aim to learn a predictor based on the quality of the downstream task decisions by incorporating mathematical programming into the learning process. This literature review discusses the variations of and approaches to PaO problems by proposing a unified notation and a taxonomy for them. Throughout the paper, we aim to provide a valuable roadmap for researchers and practitioners in the field, guiding them to choose data-driven methods to solve their decision problems effectively.

Learning latent costs of transitions on graphs from trajectories demonstrations under various contextual features is challenging but useful for path planning. Yet, existing methods either oversimplify cost assumptions or scale poorly with the number of observed trajectories. This paper introduces DataSP, a differentiable all-to-all shortest path algorithm to facilitate learning latent costs from trajectories. It allows to learn from a large number of trajectories in each learning step without additional computation. Complex latent cost functions from contextual features can be represented in the algorithm through a neural network approximation. We further propose a method to sample paths from DataSP in order to reconstruct/mimic observed paths' distributions. We prove that the inferred distribution follows the maximum entropy principle. We show that DataSP outperforms state-of-the-art differentiable combinatorial solver and classical machine learning approaches in predicting paths on graphs.

Behavior trees (BTs) are increasingly popular in the robotics community. Yet in the growing body of published work on this topic, there is a lack of consensus on what to measure and how to quantify BTs when reporting results. This is not only due to the lack of standardized measures, but due to the sometimes ambiguous use of definitions to describe BT properties. This work provides a comprehensive overview of BT properties the community is interested in, how they relate to each other, the metrics currently used to measure BTs, and whether the metrics appropriately quantify those properties of interest. Finally, we provide the practitioner with a set of metrics to measure, as well as insights into the properties that can be derived from those metrics. By providing this holistic view of properties and their corresponding evaluation metrics, we hope to improve clarity when using BTs in robotics. This more systematic approach will make reported results more consistent and comparable when evaluating BTs.

Gas Distribution Mapping (GDM) is essential in monitoring hazardous environments, where uneven sampling and spatial sparsity of data present significant challenges. Traditional methods for GDM often fall short in accuracy and expressiveness. Modern learning-based approaches employing Convolutional Neural Networks (CNNs) require regular-sized input data, limiting their adaptability to irregular and sparse datasets typically encountered in GDM. This study addresses these shortcomings by showcasing Graph Neural Networks (GNNs) for learningbased GDM on irregular and spatially sparse sensor data. Our Radius-Based, Bi-Directionally connected GNN (RABI-GNN) was trained on a synthetic gas distribution dataset on which it outperforms our previous CNN-based model while overcoming its constraints. We demonstrate the flexibility of RABI-GNN by applying it to real-world data obtained in an industrial steel factory, highlighting promising opportunities for more accurate GDM models.

In real-world scenarios, gas sensor responses to mixtures of different compositions can be costly to determine a-priori, posing difficulties in identifying the presence of target analytes. In this paper, we propose the use of graph convolutional networks (GCN) to handle gas mixtures with few labelled data. We transform sensor responses into a graph structure using manifold learning and clustering, and then apply GCN for semisupervised node classification. Our approach does not require extensive training data of gas mixtures like many competing approaches, but it outperforms classical semi-supervised learning methods and achieves classification accuracy exceeding 88.5% and over 0.85 Cohen's kappa score given only 5% labelled data for training. This result demonstrates the potential towards realistic gas identification when varied mixtures are present.

Long-term human motion prediction (LHMP) is essential for safely operating autonomous robots and vehicles in populated environments. It is fundamental for various applications, including motion planning, tracking, human-robot interaction and safety monitoring. However, accurate prediction of human trajectories is challenging due to complex factors, including, for example, social norms and environmental conditions. The influence of such factors can be captured through Maps of Dynamics (MoDs), which encode spatial motion patterns learned from (possibly scattered and partial) past observations of motion in the environment and which can be used for data-efficient, interpretable motion prediction (MoD-LHMP). To address the limitations of prior work, especially regarding accuracy and sensitivity to anomalies in long-term prediction, we propose the Laminar Component Enhanced LHMP approach (LaCE-LHMP). Our approach is inspired by data-driven airflow modelling, which estimates laminar and turbulent flow components and uses predominantly the laminar components to make flow predictions. Based on the hypothesis that human trajectory patterns also manifest laminar flow (that represents predictable motion) and turbulent flow components (that reflect more unpredictable and arbitrary motion), LaCE-LHMP extracts the laminar patterns in human dynamics and uses them for human motion prediction. We demonstrate the superior prediction performance of LaCE-LHMP through benchmark comparisons with state-of-the-art LHMP methods, offering an unconventional perspective and a more intuitive understanding of human movement patterns.

Mathematical solvers use parametrized Optimization Problems (OPs) as inputs to yield optimal decisions. In many real-world settings, some of these parameters are unknown or uncertain. Recent research focuses on predicting the value of these unknown parameters using available contextual features, aiming to decrease decision regret by adopting end-to-end learning approaches. However, these approaches disregard prediction uncertainty and therefore make the mathematical solver susceptible to provide erroneous decisions in case of low-confidence predictions. We propose a novel framework that models prediction uncertainty with Bayesian Neural Networks (BNNs) and propagates this uncertainty into the mathematical solver with a Stochastic Programming technique. The differentiable nature of BNNs and differentiable mathematical solvers allow for two different learning approaches: In the Decoupled learning approach, we update the BNN weights to increase the quality of the predictions' distribution of the OP parameters, while in the Combined learning approach, we update the weights aiming to directly minimize the expected OP's cost function in a stochastic end-to-end fashion. We do an extensive evaluation using synthetic data with various noise properties and a real dataset, showing that decisions regret are generally lower (better) with both proposed methods. The code is available at https://github.com/AlanLahoud/BNNSOP.