Ira Assent

Andrew Draganov, Tyrus Berry, Jakob Rødsgaard Jørgensen et al.

TSNE and UMAP are two of the most popular dimensionality reduction algorithms due to their speed and interpretable low-dimensional embeddings. However, while attempts have been made to improve on TSNE's computational complexity, no existing method can obtain TSNE embeddings at the speed of UMAP. In this work, we show that this is indeed possible by combining the two approaches into a single method. We theoretically and experimentally evaluate the full space of parameters in the TSNE and UMAP algorithms and observe that a single parameter, the normalization, is responsible for switching between them. This, in turn, implies that a majority of the algorithmic differences can be toggled without affecting the embeddings. We discuss the implications this has on several theoretic claims underpinning the UMAP framework, as well as how to reconcile them with existing TSNE interpretations. Based on our analysis, we propose a new dimensionality reduction algorithm, GDR, that combines previously incompatible techniques from TSNE and UMAP and can replicate the results of either algorithm by changing the normalization. As a further advantage, GDR performs the optimization faster than available UMAP methods and thus an order of magnitude faster than available TSNE methods. Our implementation is plug-and-play with the traditional UMAP and TSNE libraries and can be found at github.com/Andrew-Draganov/GiDR-DUN.

Joachim Nyborg, Charlotte Pelletier, Ira Assent

Large-scale crop type classification is a task at the core of remote sensing efforts with applications of both economic and ecological importance. Current state-of-the-art deep learning methods are based on self-attention and use satellite image time series (SITS) to discriminate crop types based on their unique growth patterns. However, existing methods generalize poorly to regions not seen during training mainly due to not being robust to temporal shifts of the growing season caused by variations in climate. To this end, we propose Thermal Positional Encoding (TPE) for attention-based crop classifiers. Unlike previous positional encoding based on calendar time (e.g. day-of-year), TPE is based on thermal time, which is obtained by accumulating daily average temperatures over the growing season. Since crop growth is directly related to thermal time, but not calendar time, TPE addresses the temporal shifts between different regions to improve generalization. We propose multiple TPE strategies, including learnable methods, to further improve results compared to the common fixed positional encodings. We demonstrate our approach on a crop classification task across four different European regions, where we obtain state-of-the-art generalization results.

Pernille Matthews, Lena Krieger, Tommaso Amico et al.

Clustering is an unsupervised technique for grouping data points by similarity. While explainability methods exist for supervised machine learning, they are not directly applicable to clustering, making it challenging to understand cluster assignments. This interpretability gap is particularly evident in the popular density-based method DBSCAN, which assigns points as inliers (cluster members in dense regions) or outliers (noise points in sparse regions). DBSCAN does not provide insight into why a particular point receives its assignment or whether its assignment is robust to small changes in the data. To address the lack of explainability, we introduce ExDBSCAN, a density-aware, post-hoc explanation method. ExDBSCAN offers actionable counterfactual explanations, with theoretical guarantees for validity. It generates multiple counterfactuals using a density connected weighted graph, adopting a physics-inspired model that repels counterfactual candidates from one another (diversity), while pulling them toward the instance to explain (proximity). Empirical evaluation on 30 tabular datasets comparing against four baselines shows that ExDBSCAN outperforms all baselines while attaining perfect validity and retrieving diverse, proximal counterfactuals.

Rafael Pablos Sarabia, Joachim Nyborg, Morten Birk et al.

This paper presents a solution to the Weather4Cast 2023 competition, where the goal is to forecast high-resolution precipitation with an 8-hour lead time using lower-resolution satellite radiance images. We propose a simple, yet effective method for spatiotemporal feature learning using a 2D U-Net model, that outperforms the official 3D U-Net baseline in both performance and efficiency. We place emphasis on refining the dataset, through importance sampling and dataset preparation, and show that such techniques have a significant impact on performance. We further study an alternative cross-entropy loss function that improves performance over the standard mean squared error loss, while also enabling models to produce probabilistic outputs. Additional techniques are explored regarding the generation of predictions at different lead times, specifically through Conditioning Lead Time. Lastly, to generate high-resolution forecasts, we evaluate standard and learned upsampling methods. The code and trained parameters are available at https://github.com/rafapablos/w4c23-rainai.

Xuan Zhao, Zhuo Cao, Arya Bangun et al.

Counterfactual explanations provide actionable insights by identifying minimal input changes required to achieve a desired model prediction. Beyond their interpretability benefits, counterfactuals can also be leveraged for model reconstruction, where a surrogate model is trained to replicate the behavior of a target model. In this work, we demonstrate that model reconstruction can be significantly improved by recognizing that counterfactuals, which typically lie close to the decision boundary, can serve as informative though less representative samples for both classes. This is particularly beneficial in settings with limited access to labeled data. We propose a method that integrates original data samples with counterfactuals to approximate class prototypes using the Wasserstein barycenter, thereby preserving the underlying distributional structure of each class. This approach enhances the quality of the surrogate model and mitigates the issue of decision boundary shift, which commonly arises when counterfactuals are naively treated as ordinary training instances. Empirical results across multiple datasets show that our method improves fidelity between the surrogate and target models, validating its effectiveness.

Alessio Quercia, Arya Bangun, Ira Assent et al.

Low-Rank Adaptation (LoRA) methods have emerged as crucial techniques for adapting large pre-trained models to downstream tasks under computational and memory constraints. However, they face a fundamental challenge in balancing task-specific performance gains against catastrophic forgetting of pre-trained knowledge, where existing methods provide inconsistent recommendations. This paper presents a comprehensive analysis of the performance-forgetting trade-offs inherent in low-rank adaptation using principal components as initialization. Our investigation reveals that fine-tuning intermediate components leads to better balance and show more robustness to high learning rates than first (PiSSA) and last (MiLoRA) components in existing work. Building on these findings, we provide a practical approach for initialization of LoRA that offers superior trade-offs. We demonstrate in a thorough empirical study on a variety of computer vision and NLP tasks that our approach improves accuracy and reduces forgetting, also in continual learning scenarios.

Rafael Pablos Sarabia, Joachim Nyborg, Morten Birk et al.

High-resolution rainfall observations are crucial for weather forecasting, water management, and hazard mitigation. Traditional operational measurements are often biased and low-resolution, limiting their ability to capture local rainfall. Accurate high-resolution rainfall maps require integrating sparse surface observations, yet existing deep learning densification methods are hindered by rainfall's skewed, localized nature, noise, and limited spatio-temporal fusion. We present DropsToGrid, a Neural Process-based method that generates dense rainfall fields by fusing temporal sequences from noisy, irregularly distributed private weather stations with spatial context from radar. Leveraging multi-scale feature extraction, temporal attention, and multi-modal fusion, the model produces stochastic, continuous rainfall estimates and explicitly quantifies uncertainty. Evaluations on real-world datasets demonstrate that DropsToGrid outperforms both operational and deep learning baselines, generating accurate high-resolution rainfall maps with well-calibrated uncertainty, even when only few stations are available and in cross-regional scenarios.

Viet-Hung Tran, Ngoc-Bao Nguyen, Son T. Mai et al.

Model Inversion (MI) attacks pose a significant privacy threat by reconstructing private training data from machine learning models. While existing defenses primarily concentrate on model-centric approaches, the impact of data on MI robustness remains largely unexplored. In this work, we explore Random Erasing (RE), a technique traditionally used for improving model generalization under occlusion, and uncover its surprising effectiveness as a defense against MI attacks. Specifically, our novel feature space analysis shows that models trained with RE-images introduce a significant discrepancy between the features of MI-reconstructed images and those of the private data. At the same time, features of private images remain distinct from other classes and well-separated from different classification regions. These effects collectively degrade MI reconstruction quality and attack accuracy while maintaining reasonable natural accuracy. Furthermore, we explore two critical properties of RE including Partial Erasure and Random Location. Partial Erasure prevents the model from observing entire objects during training. We find this has a significant impact on MI, which aims to reconstruct the entire objects. Random Location of erasure plays a crucial role in achieving a strong privacy-utility trade-off. Our findings highlight RE as a simple yet effective defense mechanism that can be easily integrated with existing privacy-preserving techniques. Extensive experiments across 37 setups demonstrate that our method achieves state-of-the-art (SOTA) performance in the privacy-utility trade-off. The results consistently demonstrate the superiority of our defense over existing methods across different MI attacks, network architectures, and attack configurations. For the first time, we achieve a significant degradation in attack accuracy without a decrease in utility for some configurations.

Amalie Brogaard Pauli, Maria Barrett, Max Müller-Eberstein et al.

Large language models (LLMs) are increasingly used for everyday communication tasks, including drafting interpersonal messages intended to influence and persuade. Prior work has shown that LLMs can successfully persuade humans and amplify persuasive language. It is therefore essential to understand how user instructions affect the generation of persuasive language, and to understand whether the generated persuasive language differs, for example, when targeting different groups. In this work, we propose a framework for evaluating how persuasive language generation is affected by recipient gender, sender intent, or output language. We evaluate 13 LLMs and 16 languages using pairwise prompt instructions. We evaluate model responses on 19 categories of persuasive language using an LLM-as-judge setup grounded in social psychology and communication science. Our results reveal significant gender differences in the persuasive language generated across all models. These patterns reflect biases consistent with gender-stereotypical linguistic tendencies documented in social psychology and sociolinguistics.

Alessio Quercia, Erenus Yildiz, Zhuo Cao et al.

Monocular depth estimation (MDE) is a challenging task in computer vision, often hindered by the cost and scarcity of high-quality labeled datasets. We tackle this challenge using auxiliary datasets from related vision tasks for an alternating training scheme with a shared decoder built on top of a pre-trained vision foundation model, while giving a higher weight to MDE. Through extensive experiments we demonstrate the benefits of incorporating various in-domain auxiliary datasets and tasks to improve MDE quality on average by ~11%. Our experimental analysis shows that auxiliary tasks have different impacts, confirming the importance of task selection, highlighting that quality gains are not achieved by merely adding data. Remarkably, our study reveals that using semantic segmentation datasets as Multi-Label Dense Classification (MLDC) often results in additional quality gains. Lastly, our method significantly improves the data efficiency for the considered MDE datasets, enhancing their quality while reducing their size by at least 80%. This paves the way for using auxiliary data from related tasks to improve MDE quality despite limited availability of high-quality labeled data. Code is available at https://jugit.fz-juelich.de/ias-8/mdeaux.

Andrew Draganov, Jakob Rødsgaard Jørgensen, Katrine Scheel Nellemann et al.

tSNE and UMAP are popular dimensionality reduction algorithms due to their speed and interpretable low-dimensional embeddings. Despite their popularity, however, little work has been done to study their full span of differences. We theoretically and experimentally evaluate the space of parameters in both tSNE and UMAP and observe that a single one -- the normalization -- is responsible for switching between them. This, in turn, implies that a majority of the algorithmic differences can be toggled without affecting the embeddings. We discuss the implications this has on several theoretic claims behind UMAP, as well as how to reconcile them with existing tSNE interpretations. Based on our analysis, we provide a method (\ourmethod) that combines previously incompatible techniques from tSNE and UMAP and can replicate the results of either algorithm. This allows our method to incorporate further improvements, such as an acceleration that obtains either method's outputs faster than UMAP. We release improved versions of tSNE, UMAP, and \ourmethod that are fully plug-and-play with the traditional libraries at https://github.com/Andrew-Draganov/GiDR-DUN

Mads Toftrup, Søren Asger Sørensen, Manuel R. Ciosici et al.

Language Identification is the task of identifying a document's language. For applications like automatic spell checker selection, language identification must use very short strings such as text message fragments. In this work, we reproduce a language identification architecture that Apple briefly sketched in a blog post. We confirm the bi-LSTM model's performance and find that it outperforms current open-source language identifiers. We further find that its language identification mistakes are due to confusion between related languages.

Anna Beer, Martin Heinrigs, Claudia Plant et al.

We introduce MOSCITO (MOlecular Dynamics Subspace Clustering with Temporal Observance), a subspace clustering for molecular dynamics data. MOSCITO groups those timesteps of a molecular dynamics trajectory together into clusters in which the molecule has similar conformations. In contrast to state-of-the-art methods, MOSCITO takes advantage of sequential relationships found in time series data. Unlike existing work, MOSCITO does not need a two-step procedure with tedious post-processing, but directly models essential properties of the data. Interpreting clusters as Markov states allows us to evaluate the clustering performance based on the resulting Markov state models. In experiments on 60 trajectories and 4 different proteins, we show that the performance of MOSCITO achieves state-of-the-art performance in a novel single-step method. Moreover, by modeling temporal aspects, MOSCITO obtains better segmentation of trajectories, especially for small numbers of clusters.

Zhuo Cao, Lena Krieger, Hanno Scharr et al.

Galaxy morphologies play an essential role in the study of the evolution of galaxies. The determination of morphologies is laborious for a large amount of data giving rise to machine learning-based approaches. Unfortunately, most of these approaches offer no insight into how the model works and make the results difficult to understand and explain. We here propose to extend a classical encoder-decoder architecture with invertible flow, allowing us to not only obtain a good predictive performance but also provide additional information about the decision process with counterfactual explanations.

Rafael Pablos Sarabia, Joachim Nyborg, Morten Birk et al.

We present a deep learning model for high-resolution probabilistic precipitation forecasting over an 8-hour horizon in Europe, overcoming the limitations of radar-only deep learning models with short forecast lead times. Our model efficiently integrates multiple data sources - including radar, satellite, and physics-based numerical weather prediction (NWP) - while capturing long-range interactions, resulting in accurate forecasts with robust uncertainty quantification through consistent probabilistic maps. Featuring a compact architecture, it enables more efficient training and faster inference than existing models. Extensive experiments demonstrate that our model surpasses current operational NWP systems, extrapolation-based methods, and deep-learning nowcasting models, setting a new standard for high-resolution precipitation forecasting in Europe, ensuring a balance between accuracy, interpretability, and computational efficiency.

Alessio Quercia, Zhuo Cao, Arya Bangun et al.

Parameter-Efficient Fine-Tuning (PEFT) methods have transformed the approach to fine-tuning large models for downstream tasks by enabling the adjustment of significantly fewer parameters than those in the original model matrices. In this work, we study the "very low rank regime", where we fine-tune the lowest amount of parameters per linear layer for each considered PEFT method. We propose 1LoRA (Summation Low-Rank Adaptation), a compute, parameter and memory efficient fine-tuning method which uses the feature sum as fixed compression and a single trainable vector as decompression. Differently from state-of-the-art PEFT methods like LoRA, VeRA, and the recent MoRA, 1LoRA uses fewer parameters per layer, reducing the memory footprint and the computational cost. We extensively evaluate our method against state-of-the-art PEFT methods on multiple fine-tuning tasks, and show that our method not only outperforms them, but is also more parameter, memory and computationally efficient. Moreover, thanks to its memory efficiency, 1LoRA allows to fine-tune more evenly across layers, instead of focusing on specific ones (e.g. attention layers), improving performance further.

Rafael Pablos Sarabia, Joachim Nyborg, Morten Birk et al.

Precipitation nowcasting is crucial across various industries and plays a significant role in mitigating and adapting to climate change. We introduce an efficient deep learning model for precipitation nowcasting, capable of predicting rainfall up to 8 hours in advance with greater accuracy than existing operational physics-based and extrapolation-based models. Our model leverages multi-source meteorological data and physics-based forecasts to deliver high-resolution predictions in both time and space. It captures complex spatio-temporal dynamics through temporal attention networks and is optimized using data quality maps and dynamic thresholds. Experiments demonstrate that our model outperforms state-of-the-art, and highlight its potential for fast reliable responses to evolving weather conditions.

Tommaso Amico, Pernille Matthews, Lena Krieger et al.

Outlier detection identifies data points that significantly deviate from the majority of the data distribution. Explaining outliers is crucial for understanding the underlying factors that contribute to their detection, validating their significance, and identifying potential biases or errors. Effective explanations provide actionable insights, facilitating preventive measures to avoid similar outliers in the future. Counterfactual explanations clarify why specific data points are classified as outliers by identifying minimal changes required to alter their prediction. Although valuable, most existing counterfactual explanation methods overlook the unique challenges posed by outlier detection, and fail to target classical, widely adopted outlier detection algorithms. Local Outlier Factor (LOF) is one the most popular unsupervised outlier detection methods, quantifying outlierness through relative local density. Despite LOF's widespread use across diverse applications, it lacks interpretability. To address this limitation, we introduce Density-based Counterfactuals for Outliers (DCFO), a novel method specifically designed to generate counterfactual explanations for LOF. DCFO partitions the data space into regions where LOF behaves smoothly, enabling efficient gradient-based optimisation. Extensive experimental validation on 50 OpenML datasets demonstrates that DCFO consistently outperforms benchmarked competitors, offering superior proximity and validity of generated counterfactuals.

Zhuo Cao, Xuan Zhao, Lena Krieger et al.

The growing integration of machine learning (ML) and artificial intelligence (AI) models into high-stakes domains such as healthcare and scientific research calls for models that are not only accurate but also interpretable. Among the existing explainable methods, counterfactual explanations offer interpretability by identifying minimal changes to inputs that would alter a model's prediction, thus providing deeper insights. However, current counterfactual generation methods suffer from critical limitations, including gradient vanishing, discontinuous latent spaces, and an overreliance on the alignment between learned and true decision boundaries. To overcome these limitations, we propose LeapFactual, a novel counterfactual explanation algorithm based on conditional flow matching. LeapFactual generates reliable and informative counterfactuals, even when true and learned decision boundaries diverge. Following a model-agnostic approach, LeapFactual is not limited to models with differentiable loss functions. It can even handle human-in-the-loop systems, expanding the scope of counterfactual explanations to domains that require the participation of human annotators, such as citizen science. We provide extensive experiments on benchmark and real-world datasets showing that LeapFactual generates accurate and in-distribution counterfactual explanations that offer actionable insights. We observe, for instance, that our reliable counterfactual samples with labels aligning to ground truth can be beneficially used as new training data to enhance the model. The proposed method is broadly applicable and enhances both scientific knowledge discovery and non-expert interpretability.

Klim Zaporojets, Daniel Daza, Edoardo Barba et al.

Knowledge Graphs (KGs) are structured knowledge repositories containing entities and relations between them. In this paper, we investigate the problem of automatically updating KGs over time with respect to the evolution of knowledge in unstructured textual sources. This problem requires identifying a wide range of update operations based on the state of an existing KG at a specific point in time. This contrasts with traditional information extraction pipelines, which extract knowledge from text independently of the current state of a KG. To address this challenge, we propose a method for lifelong construction of a dataset consisting of Wikidata KG snapshots over time and Wikipedia passages paired with the corresponding edit operations that they induce in a particular KG snapshot. The resulting dataset comprises 376K Wikipedia passages aligned with a total of 1.25M KG edits over 10 different snapshots of Wikidata from 2019 to 2025. Our experimental results highlight challenges in updating KG snapshots based on emerging textual knowledge, positioning the dataset as a valuable benchmark for future research. We will publicly release our dataset and model implementations.

Anna Beer, Lena Krieger, Pascal Weber et al.

In density-based clustering, clusters are areas of high object density separated by lower object density areas. This notion supports arbitrarily shaped clusters and automatic detection of noise points that do not belong to any cluster. However, it is challenging to adequately evaluate the quality of density-based clustering results. Even though some existing cluster validity indices (CVIs) target arbitrarily shaped clusters, none of them captures the quality of the labeled noise. In this paper, we propose DISCO, a Density-based Internal Score for Clustering Outcomes, which is the first CVI that also evaluates the quality of noise labels. DISCO reliably evaluates density-based clusters of arbitrary shape by assessing compactness and separation. It also introduces a direct assessment of noise labels for any given clustering. Our experiments show that DISCO evaluates density-based clusterings more consistently than its competitors. It is additionally the first method to evaluate the complete labeling of density-based clustering methods, including noise labels.

Amalie Brogaard Pauli, Isabelle Augenstein, Ira Assent

Large language models (LLMs) make it easy to rewrite a text in any style -- e.g. to make it more polite, persuasive, or more positive -- but evaluation thereof is not straightforward. A challenge lies in measuring content preservation: that content not attributable to style change is retained. This paper presents a large meta-evaluation of metrics for evaluating style and attribute transfer, focusing on content preservation. We find that meta-evaluation studies on existing datasets lead to misleading conclusions about the suitability of metrics for content preservation. Widely used metrics show a high correlation with human judgments despite being deemed unsuitable for the task -- because they do not abstract from style changes when evaluating content preservation. We show that the overly high correlations with human judgment stem from the nature of the test data. To address this issue, we introduce a new, challenging test set specifically designed for evaluating content preservation metrics for style transfer. We construct the data by creating high variation in the content preservation. Using this dataset, we demonstrate that suitable metrics for content preservation for style transfer indeed are style-aware. To support efficient evaluation, we propose a new style-aware method that utilises small language models, obtaining a higher alignment with human judgements than prompting a model of a similar size as an autorater. ater.

Andrew Draganov, Pascal Weber, Rasmus Skibdahl Melanchton Jørgensen et al.

Hierarchical clustering is a powerful tool for exploratory data analysis, organizing data into a tree of clusterings from which a partition can be chosen. This paper generalizes these ideas by proving that, for any reasonable hierarchy, one can optimally solve any center-based clustering objective over it (such as $k$-means). Moreover, these solutions can be found exceedingly quickly and are themselves necessarily hierarchical. Thus, given a cluster tree, we show that one can quickly access a plethora of new, equally meaningful hierarchies. Just as in standard hierarchical clustering, one can then choose any desired partition from these new hierarchies. We conclude by verifying the utility of our proposed techniques across datasets, hierarchies, and partitioning schemes.

Amalie Brogaard Pauli, Isabelle Augenstein, Ira Assent

We are exposed to much information trying to influence us, such as teaser messages, debates, politically framed news, and propaganda - all of which use persuasive language. With the recent interest in Large Language Models (LLMs), we study the ability of LLMs to produce persuasive text. As opposed to prior work which focuses on particular domains or types of persuasion, we conduct a general study across various domains to measure and benchmark to what degree LLMs produce persuasive language - both when explicitly instructed to rewrite text to be more or less persuasive and when only instructed to paraphrase. We construct the new dataset Persuasive-Pairs of pairs of a short text and its rewrite by an LLM to amplify or diminish persuasive language. We multi-annotate the pairs on a relative scale for persuasive language: a valuable resource in itself, and for training a regression model to score and benchmark persuasive language, including for new LLMs across domains. In our analysis, we find that different 'personas' in LLaMA3's system prompt change persuasive language substantially, even when only instructed to paraphrase.

Joachim Nyborg, Ira Assent

The detection of clouds in satellite images is an essential preprocessing task for big data in remote sensing. Convolutional neural networks (CNNs) have greatly advanced the state-of-the-art in the detection of clouds in satellite images, but existing CNN-based methods are costly as they require large amounts of training images with expensive pixel-level cloud labels. To alleviate this cost, we propose Fixed-Point GAN for Cloud Detection (FCD), a weakly-supervised approach. Training with only image-level labels, we learn fixed-point translation between clear and cloudy images, so only clouds are affected during translation. Doing so enables our approach to predict pixel-level cloud labels by translating satellite images to clear ones and setting a threshold to the difference between the two images. Moreover, we propose FCD+, where we exploit the label-noise robustness of CNNs to refine the prediction of FCD, leading to further improvements. We demonstrate the effectiveness of our approach on the Landsat-8 Biome cloud detection dataset, where we obtain performance close to existing fully-supervised methods that train with expensive pixel-level labels. By fine-tuning our FCD+ with just 1% of the available pixel-level labels, we match the performance of fully-supervised methods.

Joachim Nyborg, Charlotte Pelletier, Sébastien Lefèvre et al.

The recent developments of deep learning models that capture complex temporal patterns of crop phenology have greatly advanced crop classification from Satellite Image Time Series (SITS). However, when applied to target regions spatially different from the training region, these models perform poorly without any target labels due to the temporal shift of crop phenology between regions. Although various unsupervised domain adaptation techniques have been proposed in recent years, no method explicitly learns the temporal shift of SITS and thus provides only limited benefits for crop classification. To address this, we propose TimeMatch, which explicitly accounts for the temporal shift for improved SITS-based domain adaptation. In TimeMatch, we first estimate the temporal shift from the target to the source region using the predictions of a source-trained model. Then, we re-train the model for the target region by an iterative algorithm where the estimated shift is used to generate accurate target pseudo-labels. Additionally, we introduce an open-access dataset for cross-region adaptation from SITS in four different regions in Europe. On our dataset, we demonstrate that TimeMatch outperforms all competing methods by 11% in average F1-score across five different adaptation scenarios, setting a new state-of-the-art in cross-region adaptation.

Frederik Hvilshøj, Alexandros Iosifidis, Ira Assent

As counterfactual examples become increasingly popular for explaining decisions of deep learning models, it is essential to understand what properties quantitative evaluation metrics do capture and equally important what they do not capture. Currently, such understanding is lacking, potentially slowing down scientific progress. In this paper, we consolidate the work on evaluating visual counterfactual examples through an analysis and experiments. We find that while most metrics behave as intended for sufficiently simple datasets, some fail to tell the difference between good and bad counterfactuals when the complexity increases. We observe experimentally that metrics give good scores to tiny adversarial-like changes, wrongly identifying such changes as superior counterfactual examples. To mitigate this issue, we propose two new metrics, the Label Variation Score and the Oracle score, which are both less vulnerable to such tiny changes. We conclude that a proper quantitative evaluation of visual counterfactual examples should combine metrics to ensure that all aspects of good counterfactuals are quantified.

Simon Enni, Ira Assent

The influence of machine learning (ML) is quickly spreading, and a number of recent technological innovations have applied ML as a central technology. However, ML development still requires a substantial amount of human expertise to be successful. The deliberation and expert judgment applied during ML development cannot be revisited or scrutinized if not properly documented, and this hinders the further adoption of ML technologies--especially in safety critical situations. In this paper, we present a method consisting of eight design questions, that outline the deliberation and normative choices going into creating a ML model. Our method affords several benefits, such as supporting critical assessment through methodological transparency, aiding in model debugging, and anchoring model explanations by committing to a pre hoc expectation of the model's behavior. We believe that our method can help ML practitioners structure and justify their choices and assumptions when developing ML models, and that it can help bridge a gap between those inside and outside the ML field in understanding how and why ML models are designed and developed the way they are.

Frederik Hvilshøj, Alexandros Iosifidis, Ira Assent

Counterfactual examples identify how inputs can be altered to change the predicted class of a classifier, thus opening up the black-box nature of, e.g., deep neural networks. We propose a method, ECINN, that utilizes the generative capacities of invertible neural networks for image classification to generate counterfactual examples efficiently. In contrast to competing methods that sometimes need a thousand evaluations or more of the classifier, ECINN has a closed-form expression and generates a counterfactual in the time of only two evaluations. Arguably, the main challenge of generating counterfactual examples is to alter only input features that affect the predicted outcome, i.e., class-dependent features. Our experiments demonstrate how ECINN alters class-dependent image regions to change the perceptual and predicted class of the counterfactuals. Additionally, we extend ECINN to also produce heatmaps (ECINNh) for easy inspection of, e.g., pairwise class-dependent changes in the generated counterfactual examples. Experimentally, we find that ECINNh outperforms established methods that generate heatmap-based explanations.

Holger Trittenbach, Klemens Böhm, Ira Assent

Support Vector Data Description is a popular method for outlier detection. However, its usefulness largely depends on selecting good hyperparameter values -- a difficult problem that has received significant attention in literature. Existing methods to estimate hyperparameter values are purely heuristic, and the conditions under which they work well are unclear. In this article, we propose LAMA (Local Active Min-Max Alignment), the first principled approach to estimate SVDD hyperparameter values by active learning. The core idea bases on kernel alignment, which we adapt to active learning with small sample sizes. In contrast to many existing approaches, LAMA provides estimates for both SVDD hyperparameters. These estimates are evidence-based, i.e., rely on actual class labels, and come with a quality score. This eliminates the need for manual validation, an issue with current heuristics. LAMA outperforms state-of-the-art competitors in extensive experiments on real-world data. In several cases, LAMA even yields results close to the empirical upper bound.

Manuel Ciosici, Tobias Sommer, Ira Assent

Abbreviations often have several distinct meanings, often making their use in text ambiguous. Expanding them to their intended meaning in context is important for Machine Reading tasks such as document search, recommendation and question answering. Existing approaches mostly rely on manually labeled examples of abbreviations and their correct long-forms. Such data sets are costly to create and result in trained models with limited applicability and flexibility. Importantly, most current methods must be subjected to a full empirical evaluation in order to understand their limitations, which is cumbersome in practice. In this paper, we present an entirely unsupervised abbreviation disambiguation method (called UAD) that picks up abbreviation definitions from unstructured text. Creating distinct tokens per meaning, we learn context representations as word vectors. We demonstrate how to further boost abbreviation disambiguation performance by obtaining better context representations using additional unstructured text. Our method is the first abbreviation disambiguation approach with a transparent model that allows performance analysis without requiring full-scale evaluation, making it highly relevant for real-world deployments. In our thorough empirical evaluation, UAD achieves high performance on large real-world data sets from different domains and outperforms both baseline and state-of-the-art methods. UAD scales well and supports thousands of abbreviations with multiple different meanings within a single model. In order to spur more research into abbreviation disambiguation, we publish a new data set, that we also use in our experiments.

Barbora Micenková, Brian McWilliams, Ira Assent

The problem of detecting a small number of outliers in a large dataset is an important task in many fields from fraud detection to high-energy physics. Two approaches have emerged to tackle this problem: unsupervised and supervised. Supervised approaches require a sufficient amount of labeled data and are challenged by novel types of outliers and inherent class imbalance, whereas unsupervised methods do not take advantage of available labeled training examples and often exhibit poorer predictive performance. We propose BORE (a Bagged Outlier Representation Ensemble) which uses unsupervised outlier scoring functions (OSFs) as features in a supervised learning framework. BORE is able to adapt to arbitrary OSF feature representations, to the imbalance in labeled data as well as to prediction-time constraints on computational cost. We demonstrate the good performance of BORE compared to a variety of competing methods in the non-budgeted and the budgeted outlier detection problem on 12 real-world datasets.