Researcher profile

Stefan Feuerriegel

Stefan Feuerriegel contributes to research discovery and scholarly infrastructure.

ResearcherAffiliation not importedOpen to collaborate

Machine Learning Applications Artificial Intelligence Computation and Language cs.CY Social and Information Networks Human-Computer Interaction Computer Vision Information Retrieval Methodology Multiagent Systems Neural and Evolutionary Computing physics.soc-ph

Trust snapshot

Quick read

Trust 21 - EmergingVerification L1Unclaimed author

23works

0followers

13topics

4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Claiming links this public author record to a researcher profile and unlocks direct collaboration workflows.

Direct collaboration

Open a focused conversation when the fit is right

Claim this author entity first to unlock direct invitations.

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Adaptive Experimentation for Censored Survival Outcomes

Adaptive experimentation enables efficient estimation of causal effects, but existing methods are not designed for survival data with censoring, where event times are only partially observed (e.g., overall survival in cancer trials but with dropout). In this paper, we develop a novel framework for adaptive experimentation to estimate causal effects under right censoring. For this, we derive the semiparametric efficiency bound for the average survival effect curve as a function of the treatment allocation policy and thereby obtain a closed-form efficiency-optimal allocation policy. The policy generalizes classical Neyman allocation to survival settings by prioritizing patient strata where both event and censoring dynamics induce high uncertainty. Building on this, we propose the Adaptive Survival Estimator (ASE), an adaptive framework that learns the allocation policy and estimates the average survival effect curve sequentially. Our framework has three main benefits: (i) it accommodates arbitrary machine learning models for nuisance estimation; (ii) it is guided by a closed-form efficiency-optimal allocation policy; and (iii) it admits strong theoretical guarantees, including asymptotic normality via a martingale central limit theorem. We demonstrate our framework across various numerical experiments to show consistent efficiency gains over uniform randomization and censoring-agnostic baselines.

preprint2026arXiv

Amortizing Causal Sensitivity Analysis via Prior Data-Fitted Networks

Causal sensitivity analysis aims to provide bounds for causal effect estimates in the presence of unobserved confounding. However, existing methods for causal sensitivity analysis are per-instance procedures, meaning that changes to the dataset, causal query, sensitivity level, or treatment require new computation. Here, we instead present an in-context learning approach. Specifically, we propose an amortized approach to causal sensitivity analysis based on prior-data fitted networks. A key challenge is that the sensitivity bounds are not directly available when sampling training data. To address this, we develop a general prior-data construction that is applicable across the class of generalized treatment sensitivity models. Our construction involves a Lagrangian scalarization of the objective to generate training labels for the bounds through a tradeoff between causal effect min/max-imization and sensitivity model violation, which avoids model-specific analytical derivations. We further show that, under standard convexity and linearity conditions, our objective recovers the full Pareto frontier of solutions. Empirically, we demonstrate our amortized approach across various datasets, causal queries, and sensitivity levels, where our approach achieves a test-time computation that is orders of magnitude faster than per-instance methods. To the best of our knowledge, ours is the first foundation model for in-context learning for causal sensitivity analysis.

preprint2026arXiv

ConfoundingSHAP: Quantifying confounding strength in causal inference

In causal inference, confounders are variables that influence both treatment decisions and outcomes. However, unlike as in randomized clinical trials, the treatment assignment mechanism in observational studies is not known, and it is thus unclear which covariates act as confounders. Here, we aim to generate insight for causal inference and answer: which of the observed covariates act as confounders? We introduce ConfoundingSHAP, a Shapley-based method for attributing confounding strength to individual covariates. Our contributions are twofold. First, we propose a Shapley game targeted to infer the confounding strength of the covariates. Our resulting Shapley values differ from the standard applications of SHAP explanations on causal targets, such as understanding treatment effect heterogeneity, which are ill-suited for our task. Second, as our task requires evaluating the value function over many adjustment sets, we provide a scalable TabPFN-based estimation that avoids exhaustive refitting. We demonstrate the practical value across various datasets, where ConfoundingSHAP provides informative explanations of which observed covariates drive confounding and thereby helps to provide more insight for causal inference in practice.

preprint2026arXiv

Continual Learning of Domain-Invariant Representations

Continual learning (CL) aims to train models sequentially over multiple domains without forgetting previously learned knowledge. However, existing CL methods optimize for in-domain performance and are therefore prone to learning spurious, domain-specific cues (``shortcut learning''), which limits generalization to unseen domains after deployment. In this paper, we address this limitation through continual learning of domain-invariant representation. We introduce a broad class of CL methods that sequentially learn representations capturing invariant structures across domains. Our methods are motivated by the observation that such invariant structures often preserve the underlying causal mechanisms, which can reduce the risk of overfitting to domain-specific cues and thus offer better out-of-domain generalization. Our proposed CL methods combine replay-based training with a tailored sequential invariance alignment to learn -- and preserve -- invariant structures over time. We evaluate our methods under a deployment-oriented protocol that measures performance on unseen target domains. Across six benchmark and real-world datasets spanning vision, medicine, manufacturing, and ecology, our methods consistently outperform existing CL baselines in terms of generalization to unseen target domains. As an ablation, we further show that naïve extensions of sequential training with existing domain-invariant representation learning (DIRL) methods provide only limited benefits. To the best of our knowledge, this is the first work to develop domain-invariant representation methods for CL.

preprint2026arXiv

NARRA-Gym for Evaluating Interactive Narrative Agents

Interactive narrative tasks require LLMs to sustain a coherent, evolving story while adapting to a user over multiple turns. However, suitable benchmarks for this setting are limited: existing evaluations often focus on static prompts, isolated story generations, or post-hoc ratings, and therefore miss whether models can jointly manage story generation, long-context state and pacing, character simulation, empathic personalization, and story-grounded artifacts. We introduce NARRA-Gym, an executable evaluation environment that turns a sparse emotional seed into a complete interactive story episode and logs the full model-in-the-loop trajectory, including story construction, memory updates, planning, pacing interventions, and optional artifact synthesis. We evaluate nine frontier LLMs using a controlled LLM-as-judge sweep over eight benchmark personas and a human evaluation in which participants rate customized model outputs. Our results show substantial variation across models, personas, and evaluation dimensions: models that produce fluent stories can still fail on robustness, user experience, or resistance-sensitive personalization. These findings suggest that interactive narrative offers a useful benchmark for evaluating long-horizon, user-adaptive LLM behavior beyond isolated story quality.

preprint2026arXiv

ORTHOBO: Orthogonal Bayesian Hyperparameter Optimization

Bayesian optimization is widely used for hyperparameter optimization when model evaluations are expensive; however, noisy acquisition estimates can lead to unstable decisions. We identify acquisition estimation noise as a failure mode that was previously overlooked: even when the surrogate model and acquisition target are correctly specified, finite-sample Monte Carlo error can perturb acquisition values. This can, in turn, flip candidate rankings and lead to suboptimal BO decisions. As a remedy, we aim at variance reduction and propose an orthogonal acquisition estimator that subtracts an optimally weighted score-function control variate, which yields an acquisition residual orthogonal to posterior score directions and which thus reduces Monte Carlo variance. We further introduce OrthoBO: a Bayesian optimization framework that combines our orthogonal acquisition estimator with ensemble surrogates and an outer log transformation. We show theoretically that our estimator preserves the target, leads to variance reduction, and improves pairwise ranking stability. We further verify the theoretical properties of OrthoBO through numerical experiments where our framework reduces acquisition estimation variance, stabilizes candidate rankings, and achieves strong performance. We also demonstrate the downstream utility of OrthoBO in hyperparameter optimization for neural network training and fine-tuning.

preprint2026arXiv

SkillGen: Verified Inference-Time Agent Skill Synthesis

Skills are a promising way to improve LLM agent capabilities without retraining, while keeping the added procedure reusable and controllable. However, high-quality skills are still largely written by hand. We introduce SkillGen, a multi-agent framework that synthesizes a single auditable skill from trajectories generated by a base agent. The output is a human-readable artifact that can be inspected before use. Rather than merely summarizing trajectories, SkillGen leverages contrastive induction over both successful and failed trajectories to identify reusable success patterns, recurring failure modes, and behaviors that appear in nearby successes but are missing from failures. SkillGen then generates candidate skills and iteratively refines the skill. A key novelty in SkillGen is that we model agent skills as interventions to empirically verify the net effect of skills on the overall performance. Specifically, we compare outcomes on the same instances with and without the skill, so that we account for both repairs (cases where the skill fixes a baseline failure) and regressions (cases where the skill breaks a baseline success). Across a broad range of agents and datasets, SkillGen consistently improves held-out performance, outperforms existing skill-generation baselines, and produces skills that transfer across models.

preprint2022arXiv

"Is It My Turn?" Assessing Teamwork and Taskwork in Collaborative Immersive Analytics

Immersive analytics has the potential to promote collaboration in machine learning (ML). This is desired due to the specific characteristics of ML modeling in practice, namely the complexity of ML, the interdisciplinary approach in industry, and the need for ML interpretability. In this work, we introduce an augmented reality-based system for collaborative immersive analytics that is designed to support ML modeling in interdisciplinary teams. We conduct a user study to examine how collaboration unfolds when users with different professional backgrounds and levels of ML knowledge interact in solving different ML tasks. Specifically, we use the pair analytics methodology and performance assessments to assess collaboration and explore their interactions with each other and the system. Based on this, we provide qualitative and quantitative results on both teamwork and taskwork during collaboration. Our results show how our system elicits sustained collaboration as measured along six distinct dimensions. We finally make recommendations how immersive systems should be designed to elicit sustained collaboration in ML modeling.

preprint2022arXiv

Causal Transformer for Estimating Counterfactual Outcomes

Estimating counterfactual outcomes over time from observational data is relevant for many applications (e.g., personalized medicine). Yet, state-of-the-art methods build upon simple long short-term memory (LSTM) networks, thus rendering inferences for complex, long-range dependencies challenging. In this paper, we develop a novel Causal Transformer for estimating counterfactual outcomes over time. Our model is specifically designed to capture complex, long-range dependencies among time-varying confounders. For this, we combine three transformer subnetworks with separate inputs for time-varying covariates, previous treatments, and previous outcomes into a joint network with in-between cross-attentions. We further develop a custom, end-to-end training procedure for our Causal Transformer. Specifically, we propose a novel counterfactual domain confusion loss to address confounding bias: it aims to learn adversarial balanced representations, so that they are predictive of the next outcome but non-predictive of the current treatment assignment. We evaluate our Causal Transformer based on synthetic and real-world datasets, where it achieves superior performance over current baselines. To the best of our knowledge, this is the first work proposing transformer-based architecture for estimating counterfactual outcomes from longitudinal data.

preprint2022arXiv

Combining Observational and Randomized Data for Estimating Heterogeneous Treatment Effects

Estimating heterogeneous treatment effects is an important problem across many domains. In order to accurately estimate such treatment effects, one typically relies on data from observational studies or randomized experiments. Currently, most existing works rely exclusively on observational data, which is often confounded and, hence, yields biased estimates. While observational data is confounded, randomized data is unconfounded, but its sample size is usually too small to learn heterogeneous treatment effects. In this paper, we propose to estimate heterogeneous treatment effects by combining large amounts of observational data and small amounts of randomized data via representation learning. In particular, we introduce a two-step framework: first, we use observational data to learn a shared structure (in form of a representation); and then, we use randomized data to learn the data-specific structures. We analyze the finite sample properties of our framework and compare them to several natural baselines. As such, we derive conditions for when combining observational and randomized data is beneficial, and for when it is not. Based on this, we introduce a sample-efficient algorithm, called CorNet. We use extensive simulation studies to verify the theoretical properties of CorNet and multiple real-world datasets to demonstrate our method's superiority compared to existing methods.

preprint2022arXiv

Data-driven dynamic treatment planning for chronic diseases

In order to deliver effective care, health management must consider the distinctive trajectories of chronic diseases. These diseases recurrently undergo acute, unstable, and stable phases, each of which requires a different treatment regimen. However, the correct identification of trajectory phases, and thus treatment regimens, is challenging. In this paper, we propose a data-driven, dynamic approach for identifying trajectory phases of chronic diseases and thus suggesting treatment regimens. Specifically, we develop a novel variable-duration copula hidden Markov model (VDC-HMMX). In our VDC-HMMX, the trajectory is modeled as a series of latent states with acute, stable, and unstable phases, which are eventually recovered. We demonstrate the effectiveness of our VDC-HMMX model on the basis of a longitudinal study with 928 patients suffering from low back pain. A myopic classifier identifies correct treatment regimens with a balanced accuracy of slightly above 70%. In comparison, our VDC-HMMX model is correct with a balanced accuracy of 83.65%. This thus highlights the value of longitudinal monitoring for chronic disease management.

preprint2022arXiv

Detecting False Rumors from Retweet Dynamics on Social Media

False rumors are known to have detrimental effects on society. To prevent the spread of false rumors, social media platforms such as Twitter must detect them early. In this work, we develop a novel probabilistic mixture model that classifies true vs. false rumors based on the underlying spreading process. Specifically, our model is the first to formalize the self-exciting nature of true vs. false retweeting processes. This results in a novel mixture marked Hawkes model (MMHM). Owing to this, our model obviates the need for feature engineering; instead, it directly models the spreading process in order to make inferences of whether online rumors are incorrect. Our evaluation is based on 13,650 retweet cascades of both true. vs. false rumors from Twitter. Our model recognizes false rumors with a balanced accuracy of 64.97% and an AUC of 69.46%. It outperforms state-of-the-art baselines (both neural and feature engineering) by a considerable margin but while being fully interpretable. Our work has direct implications for practitioners: it leverages the spreading process as an implicit quality signal and, based on it, detects false content.

preprint2022arXiv

Detecting User Exits from Online Behavior: A Duration-Dependent Latent State Model

In order to steer e-commerce users towards making a purchase, marketers rely upon predictions of when users exit without purchasing. Previously, such predictions were based upon hidden Markov models (HMMs) due to their ability of modeling latent shopping phases with different user intents. In this work, we develop a duration-dependent hidden Markov model. In contrast to traditional HMMs, it explicitly models the duration of latent states and thereby allows states to become "sticky". The proposed model is superior to prior HMMs in detecting user exits: out of 100 user exits without purchase, it correctly identifies an additional 18. This helps marketers in better managing the online behavior of e-commerce customers. The reason for the superior performance of our model is the duration dependence, which allows our model to recover latent states that are characterized by a distorted sense of time. We finally provide a theoretical explanation for this, which builds upon the concept of "flow".

preprint2022arXiv

Online Emotions During the Storming of the U.S. Capitol: Evidence from the Social Media Network Parler

The storming of the U.S. Capitol on January 6, 2021 has led to the killing of 5 people and is widely regarded as an attack on democracy. The storming was largely coordinated through social media networks such as Parler. Yet little is known regarding how users interacted on Parler during the storming of the Capitol. In this work, we examine the emotion dynamics on Parler during the storming with regard to heterogeneity across time and users. For this, we segment the user base into different groups (e.g., Trump supporters and QAnon supporters). We use affective computing (Kratzwald et al. 2018) to infer the emotions in the contents, thereby allowing us to provide a comprehensive assessment of online emotions. Our evaluation is based on a large-scale dataset from Parler, comprising of 717,300 posts from 144,003 users. We find that the user base responded to the storming of the Capitol with an overall negative sentiment. Akin to this, Trump supporters also expressed a negative sentiment and high levels of unbelief. In contrast to that, QAnon supporters did not express a more negative sentiment during the storming. We further provide a cross-platform analysis and compare the emotion dynamics on Parler and Twitter. Our findings point at a comparatively less negative response to the incidents on Parler compared to Twitter accompanied by higher levels of disapproval and outrage. Our contribution to research is three-fold: (1) We identify online emotions that were characteristic of the storming; (2) we assess emotion dynamics across different user groups on Parler; (3) we compare the emotion dynamics on Parler and Twitter. Thereby, our work offers important implications for actively managing online emotions to prevent similar incidents in the future.

preprint2022arXiv

Sequential Deconfounding for Causal Inference with Unobserved Confounders

Using observational data to estimate the effect of a treatment is a powerful tool for decision-making when randomized experiments are infeasible or costly. However, observational data often yields biased estimates of treatment effects, since treatment assignment can be confounded by unobserved variables. A remedy is offered by deconfounding methods that adjust for such unobserved confounders. In this paper, we develop the Sequential Deconfounder, a method that enables estimating individualized treatment effects over time in presence of unobserved confounders. This is the first deconfounding method that can be used in a general sequential setting (i.e., with one or more treatments assigned at each timestep). The Sequential Deconfounder uses a novel Gaussian process latent variable model to infer substitutes for the unobserved confounders, which are then used in conjunction with an outcome model to estimate treatment effects over time. We prove that using our method yields unbiased estimates of individualized treatment responses over time. Using simulated and real medical data, we demonstrate the efficacy of our method in deconfounding the estimation of treatment responses over time.

preprint2022arXiv

Web Mining to Inform Locations of Charging Stations for Electric Vehicles

The availability of charging stations is an important factor for promoting electric vehicles (EVs) as a carbon-friendly way of transportation. Hence, for city planners, the crucial question is where to place charging stations so that they reach a large utilization. Here, we hypothesize that the utilization of EV charging stations is driven by the proximity to points-of-interest (POIs), as EV owners have a certain limited willingness to walk between charging stations and POIs. To address our research question, we propose the use of web mining: we characterize the influence of different POIs from OpenStreetMap on the utilization of charging stations. For this, we present a tailored interpretable model that takes into account the full spatial distributions of both the POIs and the charging stations. This allows us then to estimate the distance and magnitude of the influence of different POI types. We evaluate our model with data from approx. 300 charging stations and 4,000 POIs in Amsterdam, Netherlands. Our model achieves a superior performance over state-of-the-art baselines and, on top of that, is able to offer an unmatched level of interpretability. To the best of our knowledge, no previous paper has quantified the POI influence on charging station utilization from real-world usage data by estimating the spatial proximity in which POIs are relevant. As such, our findings help city planners in identifying effective locations for charging stations.

preprint2021arXiv

AttDMM: An Attentive Deep Markov Model for Risk Scoring in Intensive Care Units

Clinical practice in intensive care units (ICUs) requires early warnings when a patient's condition is about to deteriorate so that preventive measures can be undertaken. To this end, prediction algorithms have been developed that estimate the risk of mortality in ICUs. In this work, we propose a novel generative deep probabilistic model for real-time risk scoring in ICUs. Specifically, we develop an attentive deep Markov model called AttDMM. To the best of our knowledge, AttDMM is the first ICU prediction model that jointly learns both long-term disease dynamics (via attention) and different disease states in health trajectory (via a latent variable model). Our evaluations were based on an established baseline dataset (MIMIC-III) with 53,423 ICU stays. The results confirm that compared to state-of-the-art baselines, our AttDMM was superior: AttDMM achieved an area under the receiver operating characteristic curve (AUROC) of 0.876, which yielded an improvement over the state-of-the-art method by 2.2%. In addition, the risk score from the AttDMM provided warnings several hours earlier. Thereby, our model shows a path towards identifying patients at risk so that health practitioners can intervene early and save patient lives.

preprint2021arXiv

Directed particle swarm optimization with Gaussian-process-based function forecasting

Particle swarm optimization (PSO) is an iterative search method that moves a set of candidate solution around a search-space towards the best known global and local solutions with randomized step lengths. PSO frequently accelerates optimization in practical applications, where gradients are not available and function evaluations expensive. Yet the traditional PSO algorithm ignores the potential knowledge that could have been gained of the objective function from the observations by individual particles. Hence, we draw upon concepts from Bayesian optimization and introduce a stochastic surrogate model of the objective function. That is, we fit a Gaussian process to past evaluations of the objective function, forecast its shape and then adapt the particle movements based on it. Our computational experiments demonstrate that baseline implementations of PSO (i.e., SPSO2011) are outperformed. Furthermore, compared to, state-of-art surrogate-assisted evolutionary algorithms, we achieve substantial performance improvements on several popular benchmark functions. Overall, we find that our algorithm attains desirable properties for exploratory and exploitative behavior.

preprint2021arXiv

DocParser: Hierarchical Structure Parsing of Document Renderings

Translating renderings (e. g. PDFs, scans) into hierarchical document structures is extensively demanded in the daily routines of many real-world applications. However, a holistic, principled approach to inferring the complete hierarchical structure of documents is missing. As a remedy, we developed "DocParser": an end-to-end system for parsing the complete document structure - including all text elements, nested figures, tables, and table cell structures. Our second contribution is to provide a dataset for evaluating hierarchical document structure parsing. Our third contribution is to propose a scalable learning framework for settings where domain-specific data are scarce, which we address by a novel approach to weak supervision that significantly improves the document structure parsing performance. Our experiments confirm the effectiveness of our proposed weak supervision: Compared to the baseline without weak supervision, it improves the mean average precision for detecting document entities by 39.1 % and improves the F1 score of classifying hierarchical relations by 35.8 %.

preprint2021arXiv

Monitoring the COVID-19 epidemic with nationwide telecommunication data

In response to the novel coronavirus disease (COVID-19), governments have introduced severe policy measures with substantial effects on human behavior. Here, we perform a large-scale, spatio-temporal analysis of human mobility during the COVID-19 epidemic. We derive human mobility from anonymized, aggregated telecommunication data in a nationwide setting (Switzerland; February 10 - April 26, 2020), consisting of ~1.5 billion trips. In comparison to the same time period from 2019, human movement in Switzerland dropped by 49.1%. The strongest reduction is linked to bans on gatherings of more than 5 people, which is estimated to have decreased mobility by 24.9%, followed by venue closures (stores, restaurants, and bars) and school closures. As such, human mobility at a given day predicts reported cases 7-13 days ahead. A 1% reduction in human mobility predicts a 0.88-1.11% reduction in daily reported COVID-19 cases. When managing epidemics, monitoring human mobility via telecommunication data can support public decision-makers in two ways. First, it helps in assessing policy impact; second, it provides a scalable tool for near real-time epidemic surveillance, thereby enabling evidence-based policies.

preprint2020arXiv

Forecasting electricity prices with machine learning: Predictor sensitivity

Purpose: Trading on electricity markets occurs such that the price settlement takes place before delivery, often day-ahead. In practice, these prices are highly volatile as they largely depend upon a range of variables such as electricity demand and the feed-in from renewable energy sources. Hence, accurate forecasts are demanded. Approach: This paper aims at comparing different predictors stemming from supply-side (solar and wind power generation), demand-side, fuel-related and economic influences. For this reason, we implement a broad range of non-linear models from machine learning and draw upon the information-fusion-based sensitivity analysis. Findings: We disentangle the respective relevance of each predictor. We show that external predictors altogether decrease root mean squared errors by up to 21.96%. A Diebold-Mariano test statistically proves that the forecasting accuracy of the proposed machine learning models is superior. Originality: The benefit of adding further predictors has only recently received traction; however, little is known about how the individual variables contribute to improving forecasts in machine learning.

preprint2020arXiv

Leveraging Mobility Flows from Location Technology Platforms to Test Crime Pattern Theory in Large Cities

Crime has been previously explained by social characteristics of the residential population and, as stipulated by crime pattern theory, might also be linked to human movements of non-residential visitors. Yet a full empirical validation of the latter is lacking. The prime reason is that prior studies are limited to aggregated statistics of human visitors rather than mobility flows and, because of that, neglect the temporal dynamics of individual human movements. As a remedy, we provide the first work which studies the ability of granular human mobility in describing and predicting crime concentrations at an hourly scale. For this purpose, we propose the use of data from location technology platforms. This type of data allows us to trace individual transitions and, therefore, we succeed in distinguishing different mobility flows that (i) are incoming or outgoing from a neighborhood, (ii) remain within it, or (iii) refer to transitions where people only pass through the neighborhood. Our evaluation infers mobility flows by leveraging an anonymized dataset from Foursquare that includes almost 14.8 million consecutive check-ins in three major U.S. cities. According to our empirical results, mobility flows are significantly and positively linked to crime. These findings advance our theoretical understanding, as they provide confirmatory evidence for crime pattern theory. Furthermore, our novel use of digital location services data proves to be an effective tool for crime forecasting. It also offers unprecedented granularity when studying the connection between human mobility and crime.

preprint2020arXiv

Practical Annotation Strategies for Question Answering Datasets

Annotating datasets for question answering (QA) tasks is very costly, as it requires intensive manual labor and often domain-specific knowledge. Yet strategies for annotating QA datasets in a cost-effective manner are scarce. To provide a remedy for practitioners, our objective is to develop heuristic rules for annotating a subset of questions, so that the annotation cost is reduced while maintaining both in- and out-of-domain performance. For this, we conduct a large-scale analysis in order to derive practical recommendations. First, we demonstrate experimentally that more training samples contribute often only to a higher in-domain test-set performance, but do not help the model in generalizing to unseen datasets. Second, we develop a model-guided annotation strategy: it makes a recommendation with regard to which subset of samples should be annotated. Its effectiveness is demonstrated in a case study based on domain customization of QA to a clinical setting. Here, remarkably, annotating a stratified subset with only 1.2% of the original training set achieves 97.7% of the performance as if the complete dataset was annotated. Hence, the labeling effort can be reduced immensely. Altogether, our work fulfills a demand in practice when labeling budgets are limited and where thus recommendations are needed for annotating QA datasets more cost-effectively.

Institution

Affiliation not imported yet

This author record came from a source that does not expose affiliation metadata. Once the author claims the profile or we enrich the record from another provider, this section will link to the concrete institution.

Topic footprint

Fields this researcher appears in

Source provenance