Machine Learning – Publications
1.
Majumdar, Ayan; Kanubala, Deborah Dormah; Gupta, Kavya; Valera, Isabel
A Causal Framework to Measure and Mitigate Non-binary Treatment Discrimination Journal Article
In: CoRR, vol. abs/2503.22454, 2026.
@article{DBLP:journals/corr/abs-2503-22454,
title = {A Causal Framework to Measure and Mitigate Non-binary Treatment Discrimination},
author = {Ayan Majumdar and Deborah Dormah Kanubala and Kavya Gupta and Isabel Valera},
url = {https://doi.org/10.48550/arXiv.2503.22454},
doi = {10.48550/ARXIV.2503.22454},
year = {2026},
date = {2026-03-19},
urldate = {2026-03-19},
journal = {CoRR},
volume = {abs/2503.22454},
abstract = {Fairness studies of algorithmic decision-making systems often simplify complex decision processes, such as bail or lending decisions, into binary classification tasks (e.g., approve or not approve). However, these approaches overlook that such decisions are not inherently binary; they also involve non-binary treatment decisions (e.g., loan or bail terms) that can influence the downstream outcomes (e.g., loan repayment or reoffending). We argue that treatment decisions are integral to the decision-making process and, therefore, should be central to fairness analyses. Consequently, we propose a causal framework that extends and complements existing fairness notions by explicitly distinguishing between decision-subjects’ covariates and the treatment decisions. Our framework leverages path-specific counterfactual reasoning to: (i) measure treatment disparity and its downstream effects in historical data; and (ii) mitigate the impact of past unfair treatment decisions when automating decision-making. We use our framework to empirically analyze four widely used loan approval datasets to reveal potential disparity in non-binary treatment decisions and their discriminatory impact on outcomes, highlighting the need to incorporate treatment decisions in fairness assessments. Finally, by intervening in treatment decisions, we show that our framework effectively mitigates treatment discrimination from historical loan approval data to ensure fair risk score estimation and (non-binary) decision-making processes that benefit all stakeholders.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Fairness studies of algorithmic decision-making systems often simplify complex decision processes, such as bail or lending decisions, into binary classification tasks (e.g., approve or not approve). However, these approaches overlook that such decisions are not inherently binary; they also involve non-binary treatment decisions (e.g., loan or bail terms) that can influence the downstream outcomes (e.g., loan repayment or reoffending). We argue that treatment decisions are integral to the decision-making process and, therefore, should be central to fairness analyses. Consequently, we propose a causal framework that extends and complements existing fairness notions by explicitly distinguishing between decision-subjects’ covariates and the treatment decisions. Our framework leverages path-specific counterfactual reasoning to: (i) measure treatment disparity and its downstream effects in historical data; and (ii) mitigate the impact of past unfair treatment decisions when automating decision-making. We use our framework to empirically analyze four widely used loan approval datasets to reveal potential disparity in non-binary treatment decisions and their discriminatory impact on outcomes, highlighting the need to incorporate treatment decisions in fairness assessments. Finally, by intervening in treatment decisions, we show that our framework effectively mitigates treatment discrimination from historical loan approval data to ensure fair risk score estimation and (non-binary) decision-making processes that benefit all stakeholders.
2.
Koyuncu, Batuhan; Kwon, Byeungchun; Lombardi, Marco Jacopo; Perez-Cruz, Fernando; Shin, Hyun Song
BISTRO: a general purpose oracle for macroeconomic time series Journal Article
In: 2026.
@article{nokey,
title = {BISTRO: a general purpose oracle for macroeconomic time series},
author = {Batuhan Koyuncu and Byeungchun Kwon and Marco Jacopo Lombardi and Fernando Perez-Cruz and Hyun Song Shin},
year = {2026},
date = {2026-03-16},
urldate = {2026-03-16},
abstract = {Predictions of macroeconomic variables are a key input to economic policy, yet traditional econometric approaches have the limitation that the model needs to be tailored to the specific task. The advent of large language models (LLMs) opens up the tantalising prospect that a single general model can tackle a wide variety of tasks. This article introduces the BIS Time-series Regression Oracle (BISTRO), a general purpose time series model for macroeconomic forecasting. Building on the transformer architecture underlying LLMs, BISTRO is fine-tuned on the large repository of macroeconomic data maintained at the BIS. We put the model through its paces by assessing how well it forecasts the 2021 inflation surge. In contrast to standard benchmarks, which mechanically project a reversion to the mean, BISTRO correctly anticipates the persistence of the inflation wave. This highlights its ability to adapt to unfamiliar patterns in the data. Thus, BISTRO holds promise for producing reliable baseline forecasts and for scenario analysis.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Predictions of macroeconomic variables are a key input to economic policy, yet traditional econometric approaches have the limitation that the model needs to be tailored to the specific task. The advent of large language models (LLMs) opens up the tantalising prospect that a single general model can tackle a wide variety of tasks. This article introduces the BIS Time-series Regression Oracle (BISTRO), a general purpose time series model for macroeconomic forecasting. Building on the transformer architecture underlying LLMs, BISTRO is fine-tuned on the large repository of macroeconomic data maintained at the BIS. We put the model through its paces by assessing how well it forecasts the 2021 inflation surge. In contrast to standard benchmarks, which mechanically project a reversion to the mean, BISTRO correctly anticipates the persistence of the inflation wave. This highlights its ability to adapt to unfamiliar patterns in the data. Thus, BISTRO holds promise for producing reliable baseline forecasts and for scenario analysis.
3.
Koyuncu, Batuhan; Kwon, Byeungchun; Lombardi, Marco Jacopo; Perez-Cruz, Fernando; Shin, Hyun Song
Introducing BISTRO: a foundational model for unconditional and conditional forecasting of macroeconomic time series Journal Article
In: 2026.
@article{nokey,
title = {Introducing BISTRO: a foundational model for unconditional and conditional forecasting of macroeconomic time series},
author = {Batuhan Koyuncu and Byeungchun Kwon and Marco Jacopo Lombardi and Fernando Perez-Cruz and Hyun Song Shin},
year = {2026},
date = {2026-03-13},
urldate = {2026-03-13},
abstract = {This article introduces the BIS Time-series Regression Oracle (BISTRO), a general purpose time series model for macroeconomic forecasting. Its edge over traditional econometric approaches lies in its ability to deal with generic unconditional and conditional forecasting tasks, without requiring to adjust the model to the macroeconomic tasks being tackled. Building on the transformer architecture underlying LLMs, BISTRO is fine-tuned on the large repository of macroeconomic data maintained at the BIS. We show that BISTRO provides reliable unconditional forecasts for key macroeconomic aggregates and illustrate how using it for conditional forecasting can help unveiling patterns of nonlinearity in the data.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This article introduces the BIS Time-series Regression Oracle (BISTRO), a general purpose time series model for macroeconomic forecasting. Its edge over traditional econometric approaches lies in its ability to deal with generic unconditional and conditional forecasting tasks, without requiring to adjust the model to the macroeconomic tasks being tackled. Building on the transformer architecture underlying LLMs, BISTRO is fine-tuned on the large repository of macroeconomic data maintained at the BIS. We show that BISTRO provides reliable unconditional forecasts for key macroeconomic aggregates and illustrate how using it for conditional forecasting can help unveiling patterns of nonlinearity in the data.
4.
Koyuncu, Batuhan; Kwon, Byeungchun; Lombardi, Marco; Perez-Cruz, Fernando; Shin, Hyun Song
A foundational model for macroeconomic times series forecasting and nowcasting Journal Article
In: 2026.
@article{nokey,
title = {A foundational model for macroeconomic times series forecasting and nowcasting},
author = {Batuhan Koyuncu and Byeungchun Kwon and Marco Lombardi and Fernando Perez-Cruz and Hyun Song Shin},
year = {2026},
date = {2026-02-20},
urldate = {2026-02-20},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
5.
Wang, Yifan; Jobanputra, Mayank; Lee, Ji-Ung; Oh, Soyoung; Valera, Isabel; Demberg, Vera
Bridging Fairness and Explainability: Can Input-Based Explanations Promote Fairness in Hate Speech Detection? Journal Article
In: 2026.
@article{nokey,
title = {Bridging Fairness and Explainability: Can Input-Based Explanations Promote Fairness in Hate Speech Detection?},
author = {Yifan Wang and Mayank Jobanputra and Ji-Ung Lee and Soyoung Oh and Isabel Valera and Vera Demberg},
doi = { https://doi.org/10.48550/arXiv.2509.22291},
year = {2026},
date = {2026-02-11},
urldate = {2026-02-11},
abstract = {Natural language processing (NLP) models often replicate or amplify social bias from training data, raising concerns about fairness. At the same time, their black-box nature makes it difficult for users to recognize biased predictions and for developers to effectively mitigate them. While some studies suggest that input-based explanations can help detect and mitigate bias, others question their reliability in ensuring fairness. Existing research on explainability in fair NLP has been predominantly qualitative, with limited large-scale quantitative analysis. In this work, we conduct the first systematic study of the relationship between explainability and fairness in hate speech detection, focusing on both encoder- and decoder-only models. We examine three key dimensions: (1) identifying biased predictions, (2) selecting fair models, and (3) mitigating bias during model training. Our findings show that input-based explanations can effectively detect biased predictions and serve as useful supervision for reducing bias during training, but they are unreliable for selecting fair models among this http URL http://candidates.our/ code is available at this URL https://github.com/Ewanwong/fairness_x_explainability},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Natural language processing (NLP) models often replicate or amplify social bias from training data, raising concerns about fairness. At the same time, their black-box nature makes it difficult for users to recognize biased predictions and for developers to effectively mitigate them. While some studies suggest that input-based explanations can help detect and mitigate bias, others question their reliability in ensuring fairness. Existing research on explainability in fair NLP has been predominantly qualitative, with limited large-scale quantitative analysis. In this work, we conduct the first systematic study of the relationship between explainability and fairness in hate speech detection, focusing on both encoder- and decoder-only models. We examine three key dimensions: (1) identifying biased predictions, (2) selecting fair models, and (3) mitigating bias during model training. Our findings show that input-based explanations can effectively detect biased predictions and serve as useful supervision for reducing bias during training, but they are unreliable for selecting fair models among this http URL http://candidates.our/ code is available at this URL https://github.com/Ewanwong/fairness_x_explainability
6.
Uth, Richard; Niemitz, Nelli; Valera, Isabel; Langer, Markus
Personalizing explanations in AI-based decisions: The effects of personalization and (Mis)aligning with individual preferences Journal Article
In: Computers in Human Behavior, 2026.
@article{Uth2025PersonalizingEI,
title = {Personalizing explanations in AI-based decisions: The effects of personalization and (Mis)aligning with individual preferences},
author = {Richard Uth and Nelli Niemitz and Isabel Valera and Markus Langer},
url = {https://api.semanticscholar.org/CorpusID:283171660},
year = {2026},
date = {2026-02-03},
urldate = {2025-01-01},
journal = {Computers in Human Behavior},
abstract = {The increasing reliance on AI-based decision-making in high-stakes contexts underscores the need for transparency and justice. Here, negative outcomes drive individuals affected by AI-based decisions to seek actionable explanations that enable them to realize what they can do to achieve a better future outcome. However, actionability is subjective, varying across individuals and contexts. Personalization of explanations has been proposed to address this variability, but insights on personalized explanation processes, their potential, and challenges are scarce. This paper investigates the impact of personalization and (mis)alignment with individual needs and preferences in explanations for AI-based decisions through an experimental online study simulating denied loan applications. In a within-participants design (N=255), participants ranked the actionability of decision-relevant features and experienced five explanation conditions: personalized directive explanations based on the most, second most, or least actionable feature (as ranked by participants); a non-personalized directive explanation highlighting a random feature; and no explanation. In line with justice theory, our results show that any explanation was better than none, and that personalized explanations led to more favorable reactions than non-personalized explanations, enhancing perceptions of justice and attractiveness of the bank. Closer alignment with preferences had only small positive effects, mainly for attractiveness. These findings highlight that even simple ranking-based approaches can make explanations more effective and accessible without requiring technical expertise while cautioning against offering superficial control. This study provides insights into the effects of ranking-based personalization, informing the design of explainability tailored to diverse user needs and addressing ethical and practical considerations in personalization.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The increasing reliance on AI-based decision-making in high-stakes contexts underscores the need for transparency and justice. Here, negative outcomes drive individuals affected by AI-based decisions to seek actionable explanations that enable them to realize what they can do to achieve a better future outcome. However, actionability is subjective, varying across individuals and contexts. Personalization of explanations has been proposed to address this variability, but insights on personalized explanation processes, their potential, and challenges are scarce. This paper investigates the impact of personalization and (mis)alignment with individual needs and preferences in explanations for AI-based decisions through an experimental online study simulating denied loan applications. In a within-participants design (N=255), participants ranked the actionability of decision-relevant features and experienced five explanation conditions: personalized directive explanations based on the most, second most, or least actionable feature (as ranked by participants); a non-personalized directive explanation highlighting a random feature; and no explanation. In line with justice theory, our results show that any explanation was better than none, and that personalized explanations led to more favorable reactions than non-personalized explanations, enhancing perceptions of justice and attractiveness of the bank. Closer alignment with preferences had only small positive effects, mainly for attractiveness. These findings highlight that even simple ranking-based approaches can make explanations more effective and accessible without requiring technical expertise while cautioning against offering superficial control. This study provides insights into the effects of ranking-based personalization, informing the design of explainability tailored to diverse user needs and addressing ethical and practical considerations in personalization.
7.
Vo, Huyen; Valera, Isabel
Hellinger Multimodal Variational Autoencoders Miscellaneous
2026.
@misc{vo2026hellingermultimodalvariationalautoencoders,
title = {Hellinger Multimodal Variational Autoencoders},
author = {Huyen Vo and Isabel Valera},
url = {https://arxiv.org/abs/2601.06572},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
abstract = {Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from Hölder pooling with , which corresponds to the unique symmetric member of the family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from Hölder pooling with , which corresponds to the unique symmetric member of the family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models.
8.
Müller, Nicola J.; Oster, Moritz; Valera, Isabel; Hoffmann, Jörg; Gros, Timo P.
Per-Domain Generalizing Policies: On Learning Efficient and Robust Q-Value Functions (Extended Version with Technical Appendix) Journal Article
In: CoRR, vol. abs/2603.17544, 2026.
@article{DBLP:journals/corr/abs-2603-17544,
title = {Per-Domain Generalizing Policies: On Learning Efficient and Robust Q-Value Functions (Extended Version with Technical Appendix)},
author = {Nicola J. Müller and Moritz Oster and Isabel Valera and Jörg Hoffmann and Timo P. Gros},
url = {https://doi.org/10.48550/arXiv.2603.17544},
doi = {10.48550/ARXIV.2603.17544},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {CoRR},
volume = {abs/2603.17544},
abstract = {Learning per-domain generalizing policies is a key challenge in learning for planning. Standard approaches learn state-value functions represented as graph neural networks using supervised learning on optimal plans generated by a teacher planner. In this work, we advocate for learning Q-value functions instead. Such policies are drastically cheaper to evaluate for a given state, as they need to process only the current state rather than every successor. Surprisingly, vanilla supervised learning of Q-values performs poorly as it does not learn to distinguish between the actions taken and those not taken by the teacher. We address this by using regularization terms that enforce this distinction, resulting in Q-value policies that consistently outperform state-value policies across a range of 10 domains and are competitive with the planner LAMA-first.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Learning per-domain generalizing policies is a key challenge in learning for planning. Standard approaches learn state-value functions represented as graph neural networks using supervised learning on optimal plans generated by a teacher planner. In this work, we advocate for learning Q-value functions instead. Such policies are drastically cheaper to evaluate for a given state, as they need to process only the current state rather than every successor. Surprisingly, vanilla supervised learning of Q-values performs poorly as it does not learn to distinguish between the actions taken and those not taken by the teacher. We address this by using regularization terms that enforce this distinction, resulting in Q-value policies that consistently outperform state-value policies across a range of 10 domains and are competitive with the planner LAMA-first.
9.
Jobanputra, Mayank; Kovtunova, Alisa; Balthes, Brisca; Pogulskiy, Fedor Grigoryevich; Wang, Yifan; Borgwardt, Stefan; Demberg, Vera
ProofTeller: Exposing recency bias in LLM reasoning and its side effects on communication Proceedings Article
In: Inui, Kentaro; Sakti, Sakriani; Wang, Haofen; Wong, Derek F.; Bhattacharyya, Pushpak; Banerjee, Biplab; Ekbal, Asif; Chakraborty, Tanmoy; Singh, Dhirendra Pratap (Ed.): Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, pp. 1439–1462, The Asian Federation of Natural Language Processing and The Association for Computational Linguistics, Mumbai, India, 2025, ISBN: 979-8-89176-298-5.
@inproceedings{jobanputra-etal-2025-proofteller,
title = {ProofTeller: Exposing recency bias in LLM reasoning and its side effects on communication},
author = {Mayank Jobanputra and Alisa Kovtunova and Brisca Balthes and Fedor Grigoryevich Pogulskiy and Yifan Wang and Stefan Borgwardt and Vera Demberg},
editor = {Kentaro Inui and Sakriani Sakti and Haofen Wang and Derek F. Wong and Pushpak Bhattacharyya and Biplab Banerjee and Asif Ekbal and Tanmoy Chakraborty and Dhirendra Pratap Singh},
url = {https://aclanthology.org/2025.ijcnlp-long.80/},
doi = {10.18653/v1/2025.ijcnlp-long.80},
isbn = {979-8-89176-298-5},
year = {2025},
date = {2025-12-01},
urldate = {2025-12-01},
booktitle = {Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics},
pages = {1439–1462},
publisher = {The Asian Federation of Natural Language Processing and The Association for Computational Linguistics},
address = {Mumbai, India},
abstract = {Large language models (LLMs) are increasingly applied in domains that demand reliable and interpretable reasoning. While formal methods can generate provably correct proofs, these proofs are often inaccessible to non-expert users. This raises a natural question: can LLMs, when given a verified proof, faithfully interpret its reasoning and communicate it clearly? We introduce $ProofTeller$, a benchmark that evaluates this ability across three tasks: (1) identifying key proof steps, (2) summarizing the reasoning, and (3) explaining the result in concise natural language. The benchmark covers three domains: _Biology_, _Drones_, and _Recipes_, representing scientific, safety-critical, and everyday reasoning scenarios. We find a consistent near-conclusion bias: LLMs tend to focus on steps closest to the final proof conclusion rather than on the most informative ones. A targeted human study confirms that explanations based on such steps are rated less appropriate for end users. These findings indicate that even when reasoning is provided, current LLMs face challenges in communicating key information in a useful manner, highlighting the need for LLMs that can communicate important details reliably.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Large language models (LLMs) are increasingly applied in domains that demand reliable and interpretable reasoning. While formal methods can generate provably correct proofs, these proofs are often inaccessible to non-expert users. This raises a natural question: can LLMs, when given a verified proof, faithfully interpret its reasoning and communicate it clearly? We introduce $ProofTeller$, a benchmark that evaluates this ability across three tasks: (1) identifying key proof steps, (2) summarizing the reasoning, and (3) explaining the result in concise natural language. The benchmark covers three domains: _Biology_, _Drones_, and _Recipes_, representing scientific, safety-critical, and everyday reasoning scenarios. We find a consistent near-conclusion bias: LLMs tend to focus on steps closest to the final proof conclusion rather than on the most informative ones. A targeted human study confirms that explanations based on such steps are rated less appropriate for end users. These findings indicate that even when reasoning is provided, current LLMs face challenges in communicating key information in a useful manner, highlighting the need for LLMs that can communicate important details reliably.
10.
Martínez-García, María; Villacrés, Grace; Mitchell, David; Olmos, Pablo M
Improved Variational Inference in Discrete VAEs using Error Correcting Codes Proceedings Article
In: The 41st Conference on Uncertainty in Artificial Intelligence, 2025.
@inproceedings{martinezimproved,
title = {Improved Variational Inference in Discrete VAEs using Error Correcting Codes},
author = {María Martínez-García and Grace Villacrés and David Mitchell and Pablo M Olmos},
url = {https://proceedings.mlr.press/v286/martinez-garcia25a.html},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {The 41st Conference on Uncertainty in Artificial Intelligence},
abstract = {Despite advances in deep probabilistic models, learning discrete latent representations remains challenging. This work introduces a novel method to improve inference in discrete Variational Autoencoders by reframing the inference problem through a generative perspective. We conceptualize the model as a communication system, and propose to leverage Error-Correcting Codes (ECCs) to introduce redundancy in latent representations, allowing the variational posterior to produce more accurate estimates and reduce the variational gap. We present a proof-of-concept using a Discrete Variational Autoencoder with binary latent variables and low-complexity repetition codes, extending it to a hierarchical structure for disentangling global and local data features. Our approach significantly improves generation quality, data reconstruction, and uncertainty calibration, outperforming the uncoded models even when trained with tighter bounds such as the Importance Weighted Autoencoder objective. We also outline the properties that ECCs should possess to be effectively utilized for improved discrete variational inference.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite advances in deep probabilistic models, learning discrete latent representations remains challenging. This work introduces a novel method to improve inference in discrete Variational Autoencoders by reframing the inference problem through a generative perspective. We conceptualize the model as a communication system, and propose to leverage Error-Correcting Codes (ECCs) to introduce redundancy in latent representations, allowing the variational posterior to produce more accurate estimates and reduce the variational gap. We present a proof-of-concept using a Discrete Variational Autoencoder with binary latent variables and low-complexity repetition codes, extending it to a hierarchical structure for disentangling global and local data features. Our approach significantly improves generation quality, data reconstruction, and uncertainty calibration, outperforming the uncoded models even when trained with tighter bounds such as the Importance Weighted Autoencoder objective. We also outline the properties that ECCs should possess to be effectively utilized for improved discrete variational inference.
11.
Eisenhut, Jan; Fivser, Daniel; Valera, Isabel; Hoffmann, J¨org
On Picking Good Policies: Leveraging Action-Policy Testing in Policy Training Journal Article
In: Proceedings of the International Conference on Automated Planning and Scheduling, 2025.
@article{Eisenhut2025OnPG,
title = {On Picking Good Policies: Leveraging Action-Policy Testing in Policy Training},
author = {Jan Eisenhut and Daniel Fivser and Isabel Valera and J¨org Hoffmann},
url = {https://api.semanticscholar.org/CorpusID:281347744},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Proceedings of the International Conference on Automated Planning and Scheduling},
abstract = {Testing is a natural approach to assess the quality of learned action policies π. Prior work introduced policy testing in AI planning as searching for bugs in π, that is, states where π is sub-optimal with respect to a given testing objective. Beyond quality assurance, an obvious application of these methods is policy selection: given several π to choose from, we can use testing to select the "least buggy" one. Here, we integrate testing-based policy selection into the training process. This includes making more informed decisions when selecting the final policy after training, as well as choosing more promising intermediate policies during the training process. Our experiments with ASNets action policies show that integrating testing allows us to more reliably obtain good-quality policies.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Testing is a natural approach to assess the quality of learned action policies π. Prior work introduced policy testing in AI planning as searching for bugs in π, that is, states where π is sub-optimal with respect to a given testing objective. Beyond quality assurance, an obvious application of these methods is policy selection: given several π to choose from, we can use testing to select the "least buggy" one. Here, we integrate testing-based policy selection into the training process. This includes making more informed decisions when selecting the final policy after training, as well as choosing more promising intermediate policies during the training process. Our experiments with ASNets action policies show that integrating testing allows us to more reliably obtain good-quality policies.
12.
Eniser, Hasan Ferit; Lin, Songtuan; Müller, Nicola; Isychev, Anastasia; Wüstholz, Valentin; Valera, Isabel; Hoffmann, J¨org; Christakis, Maria
Using Action-Policy Testing in RL to Reduce the Number of Bugs Journal Article
In: Proceedings of the International Symposium on Combinatorial Search, 2025.
@article{Eniser2025UsingAT,
title = {Using Action-Policy Testing in RL to Reduce the Number of Bugs},
author = {Hasan Ferit Eniser and Songtuan Lin and Nicola Müller and Anastasia Isychev and Valentin Wüstholz and Isabel Valera and J¨org Hoffmann and Maria Christakis},
url = {https://api.semanticscholar.org/CorpusID:280219509},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Proceedings of the International Symposium on Combinatorial Search},
abstract = {Reinforcement learning is becoming ever more prominent in solving combinatorial search problems, in particular ones where states are images. Prior work has devised action-policy testing methodology, that identifies so-called bug states where policy performance is sub-optimal. Here we show how to leverage this methodology during the RL process, using action-policy testing to find bugs and injecting those as alternate start states for the training runs. Running experiments across six 2D games, we find that our testing-guided training often achieves similar expected reward while reducing the number of bugs.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Reinforcement learning is becoming ever more prominent in solving combinatorial search problems, in particular ones where states are images. Prior work has devised action-policy testing methodology, that identifies so-called bug states where policy performance is sub-optimal. Here we show how to leverage this methodology during the RL process, using action-policy testing to find bugs and injecting those as alternate start states for the training runs. Running experiments across six 2D games, we find that our testing-guided training often achieves similar expected reward while reducing the number of bugs.
13.
Javaloy, Adrián; Vergari, Antonio; Valera, Isabel
COPA: Comparing the Incomparable to Explore the Pareto Front Journal Article
In: CoRR, vol. abs/2503.14321, 2025.
@article{DBLP:journals/corr/abs-2503-14321,
title = {COPA: Comparing the Incomparable to Explore the Pareto Front},
author = {Adrián Javaloy and Antonio Vergari and Isabel Valera},
url = {https://doi.org/10.48550/arXiv.2503.14321},
doi = {10.48550/ARXIV.2503.14321},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2503.14321},
abstract = {In machine learning (ML), we often need to choose one among hundreds of trained ML models at hand, based on various objectives such as accuracy, robustness, fairness or scalability. However, it is often unclear how to compare, aggregate and, ultimately, trade-off these objectives, making it a time-consuming task that requires expert knowledge, as objectives may be measured in different units and scales. In this work, we investigate how objectives can be automatically normalized and aggregated to systematically help the user navigate their Pareto front. To this end, we make incomparable objectives comparable using their cumulative functions, approximated by their relative rankings. As a result, our proposed approach, COPA, can aggregate them while matching user-specific preferences, allowing practitioners to meaningfully navigate and search for models in the Pareto front. We demonstrate the potential impact of COPA in both model selection and benchmarking tasks across diverse ML areas such as fair ML, domain generalization, AutoML and foundation models, where classical ways to normalize and aggregate objectives fall short.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In machine learning (ML), we often need to choose one among hundreds of trained ML models at hand, based on various objectives such as accuracy, robustness, fairness or scalability. However, it is often unclear how to compare, aggregate and, ultimately, trade-off these objectives, making it a time-consuming task that requires expert knowledge, as objectives may be measured in different units and scales. In this work, we investigate how objectives can be automatically normalized and aggregated to systematically help the user navigate their Pareto front. To this end, we make incomparable objectives comparable using their cumulative functions, approximated by their relative rankings. As a result, our proposed approach, COPA, can aggregate them while matching user-specific preferences, allowing practitioners to meaningfully navigate and search for models in the Pareto front. We demonstrate the potential impact of COPA in both model selection and benchmarking tasks across diverse ML areas such as fair ML, domain generalization, AutoML and foundation models, where classical ways to normalize and aggregate objectives fall short.
14.
Almodóvar, Alejandro; Javaloy, Adrián; Parras, Juan; Zazo, Santiago; Valera, Isabel
DeCaFlow: A Deconfounding Causal Generative Model Journal Article
In: CoRR, vol. abs/2503.15114, 2025.
@article{DBLP:journals/corr/abs-2503-15114,
title = {DeCaFlow: A Deconfounding Causal Generative Model},
author = {Alejandro Almodóvar and Adrián Javaloy and Juan Parras and Santiago Zazo and Isabel Valera},
url = {https://doi.org/10.48550/arXiv.2503.15114},
doi = {10.48550/ARXIV.2503.15114},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2503.15114},
abstract = {We introduce DeCaFlow, a deconfounding causal generative model. Training once per dataset using just observational data and the underlying causal graph, DeCaFlow enables accurate causal inference on continuous variables under the presence of hidden confounders. Specifically, we extend previous results on causal estimation under hidden confounding to show that a single instance of DeCaFlow provides correct estimates for all causal queries identifiable with do-calculus, leveraging proxy variables to adjust for the causal effects when do-calculus alone is insufficient. Moreover, we show that counterfactual queries are identifiable as long as their interventional counterparts are identifiable, and thus are also correctly estimated by DeCaFlow. Our empirical results on diverse settings (including the Ecoli70 dataset, with 3 independent hidden confounders, tens of observed variables and hundreds of causal queries) show that DeCaFlow outperforms existing approaches, while demonstrating its out-of-the-box applicability to any given causal graph},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We introduce DeCaFlow, a deconfounding causal generative model. Training once per dataset using just observational data and the underlying causal graph, DeCaFlow enables accurate causal inference on continuous variables under the presence of hidden confounders. Specifically, we extend previous results on causal estimation under hidden confounding to show that a single instance of DeCaFlow provides correct estimates for all causal queries identifiable with do-calculus, leveraging proxy variables to adjust for the causal effects when do-calculus alone is insufficient. Moreover, we show that counterfactual queries are identifiable as long as their interventional counterparts are identifiable, and thus are also correctly estimated by DeCaFlow. Our empirical results on diverse settings (including the Ecoli70 dataset, with 3 independent hidden confounders, tens of observed variables and hundreds of causal queries) show that DeCaFlow outperforms existing approaches, while demonstrating its out-of-the-box applicability to any given causal graph
15.
Cinquini, Martina; Beretta, Isacco; Ruggieri, Salvatore; Valera, Isabel
A Practical Approach to Causal Inference over Time Proceedings Article
In: Walsh, Toby; Shah, Julie; Kolter, Zico (Ed.): AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Philadelphia, PA, USA, pp. 14832–14839, AAAI Press, 2025.
@inproceedings{DBLP:conf/aaai/CinquiniBRV25,
title = {A Practical Approach to Causal Inference over Time},
author = {Martina Cinquini and Isacco Beretta and Salvatore Ruggieri and Isabel Valera},
editor = {Toby Walsh and Julie Shah and Zico Kolter},
url = {https://doi.org/10.1609/aaai.v39i14.33626},
doi = {10.1609/AAAI.V39I14.33626},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {AAAI-25, Sponsored by the Association for the Advancement of Artificial
Intelligence, February 25 - March 4, 2025, Philadelphia, PA, USA},
pages = {14832–14839},
publisher = {AAAI Press},
abstract = {In this paper, we focus on estimating the causal effect of an intervention over time on a dynamical system. To that end, we formally define causal interventions and their effects over time on discrete-time stochastic processes (DSPs). Then, we show under which conditions the equilibrium states of a DSP, both before and after a causal intervention, can be captured by a structural causal model (SCM). With such an equivalence at hand, we provide an explicit mapping from vector autoregressive models (VARs), broadly applied in econometrics, to linear, but potentially cyclic and/or affected by unmeasured confounders, SCMs. The resulting causal VAR framework allows us to perform causal inference over time from observational time series data. Our experiments on synthetic and real-world datasets show that the proposed framework achieves strong performance in terms of observational forecasting while enabling accurate estimation of the causal effect of interventions on dynamical systems. We demonstrate, through a case study, the potential practical questions that can be addressed using the proposed causal VAR framework.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In this paper, we focus on estimating the causal effect of an intervention over time on a dynamical system. To that end, we formally define causal interventions and their effects over time on discrete-time stochastic processes (DSPs). Then, we show under which conditions the equilibrium states of a DSP, both before and after a causal intervention, can be captured by a structural causal model (SCM). With such an equivalence at hand, we provide an explicit mapping from vector autoregressive models (VARs), broadly applied in econometrics, to linear, but potentially cyclic and/or affected by unmeasured confounders, SCMs. The resulting causal VAR framework allows us to perform causal inference over time from observational time series data. Our experiments on synthetic and real-world datasets show that the proposed framework achieves strong performance in terms of observational forecasting while enabling accurate estimation of the causal effect of interventions on dynamical systems. We demonstrate, through a case study, the potential practical questions that can be addressed using the proposed causal VAR framework.
16.
Peis, Ignacio; Koyuncu, Batuhan; Valera, Isabel; Frellsen, Jes
Hyper-Transforming Latent Diffusion Models Journal Article
In: CoRR, vol. abs/2504.16580, 2025.
@article{DBLP:journals/corr/abs-2504-16580,
title = {Hyper-Transforming Latent Diffusion Models},
author = {Ignacio Peis and Batuhan Koyuncu and Isabel Valera and Jes Frellsen},
url = {https://doi.org/10.48550/arXiv.2504.16580},
doi = {10.48550/ARXIV.2504.16580},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2504.16580},
abstract = {We introduce a novel generative framework for functions by integrating Implicit Neural Representations (INRs) and Transformer-based hypernetworks into latent variable models. Unlike prior approaches that rely on MLP-based hypernetworks with scalability limitations, our method employs a Transformer-based decoder to generate INR parameters from latent variables, addressing both representation capacity and computational efficiency. Our framework extends latent diffusion models (LDMs) to INR generation by replacing standard decoders with a Transformer-based hypernetwork, which can be trained either from scratch or via hyper-transforming: a strategy that fine-tunes only the decoder while freezing the pre-trained latent space. This enables efficient adaptation of existing generative models to INR-based representations without requiring full retraining. We validate our approach across multiple modalities, demonstrating improved scalability, expressiveness, and generalization over existing INR-based generative models. Our findings establish a unified and flexible framework for learning structured function representations.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We introduce a novel generative framework for functions by integrating Implicit Neural Representations (INRs) and Transformer-based hypernetworks into latent variable models. Unlike prior approaches that rely on MLP-based hypernetworks with scalability limitations, our method employs a Transformer-based decoder to generate INR parameters from latent variables, addressing both representation capacity and computational efficiency. Our framework extends latent diffusion models (LDMs) to INR generation by replacing standard decoders with a Transformer-based hypernetwork, which can be trained either from scratch or via hyper-transforming: a strategy that fine-tunes only the decoder while freezing the pre-trained latent space. This enables efficient adaptation of existing generative models to INR-based representations without requiring full retraining. We validate our approach across multiple modalities, demonstrating improved scalability, expressiveness, and generalization over existing INR-based generative models. Our findings establish a unified and flexible framework for learning structured function representations.
17.
Koyuncu, Batuhan; DeVries, Rachael; Winther, Ole; Valera, Isabel
Temporal Variational Implicit Neural Representations Journal Article
In: CoRR, vol. abs/2506.01544, 2025.
@article{DBLP:journals/corr/abs-2506-01544,
title = {Temporal Variational Implicit Neural Representations},
author = {Batuhan Koyuncu and Rachael DeVries and Ole Winther and Isabel Valera},
url = {https://doi.org/10.48550/arXiv.2506.01544},
doi = {10.48550/ARXIV.2506.01544},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2506.01544},
abstract = {We introduce Temporal Variational Implicit Neural Representations (TV-INRs), a probabilistic framework for modeling irregular multivariate time series that enables efficient individualized imputation and forecasting. By integrating implicit neural representations with latent variable models, TV-INRs learn distributions over time-continuous generator functions conditioned on signal-specific covariates. Unlike existing approaches that require extensive training, fine-tuning or meta-learning, our method achieves accurate individualized predictions through a single forward pass. Our experiments demonstrate that with a single TV-INRs instance, we can accurately solve diverse imputation and forecasting tasks, offering a computationally efficient and scalable solution for real-world applications. TV-INRs excel especially in low-data regimes, where it outperforms existing methods by an order of magnitude in mean squared error for imputation task.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We introduce Temporal Variational Implicit Neural Representations (TV-INRs), a probabilistic framework for modeling irregular multivariate time series that enables efficient individualized imputation and forecasting. By integrating implicit neural representations with latent variable models, TV-INRs learn distributions over time-continuous generator functions conditioned on signal-specific covariates. Unlike existing approaches that require extensive training, fine-tuning or meta-learning, our method achieves accurate individualized predictions through a single forward pass. Our experiments demonstrate that with a single TV-INRs instance, we can accurately solve diverse imputation and forecasting tasks, offering a computationally efficient and scalable solution for real-world applications. TV-INRs excel especially in low-data regimes, where it outperforms existing methods by an order of magnitude in mean squared error for imputation task.
18.
Kalampalikis, Nektarios; Gupta, Kavya; Vitanov, Georgi; Valera, Isabel
Towards Reasonable Concept Bottleneck Models Journal Article
In: CoRR, vol. abs/2506.05014, 2025.
@article{DBLP:journals/corr/abs-2506-05014,
title = {Towards Reasonable Concept Bottleneck Models},
author = {Nektarios Kalampalikis and Kavya Gupta and Georgi Vitanov and Isabel Valera},
url = {https://doi.org/10.48550/arXiv.2506.05014},
doi = {10.48550/ARXIV.2506.05014},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2506.05014},
abstract = {In this paper, we propose textbf{C}oncept textbf{REA}soning textbf{M}odels (CREAM), a novel family of Concept Bottleneck Models (CBMs) that: (i) explicitly encodes concept-concept ({texttt{C-C}}) and concept-task ({texttt{C$rightarrow$Y}}) relationships to enforce a desired model reasoning; and (ii) use a regularized side-channel to achieve competitive task performance, while keeping high concept importance. Specifically, CREAM architecturally embeds (bi)directed concept-concept, and concept to task relationships specified by a human expert, while severing undesired information flows (e.g., to handle mutually exclusive concepts). Moreover, CREAM integrates a black-box side-channel that is regularized to encourage task predictions to be grounded in the relevant concepts, thereby utilizing the side-channel only when necessary to enhance performance. Our experiments show that: (i) CREAM mainly relies on concepts while achieving task performance on par with black-box models; and (ii) the embedded {texttt{C-C}} and {texttt{C$rightarrow$Y}} relationships ease model interventions and mitigate concept leakage.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In this paper, we propose textbf{C}oncept textbf{REA}soning textbf{M}odels (CREAM), a novel family of Concept Bottleneck Models (CBMs) that: (i) explicitly encodes concept-concept ({texttt{C-C}}) and concept-task ({texttt{C$rightarrow$Y}}) relationships to enforce a desired model reasoning; and (ii) use a regularized side-channel to achieve competitive task performance, while keeping high concept importance. Specifically, CREAM architecturally embeds (bi)directed concept-concept, and concept to task relationships specified by a human expert, while severing undesired information flows (e.g., to handle mutually exclusive concepts). Moreover, CREAM integrates a black-box side-channel that is regularized to encourage task predictions to be grounded in the relevant concepts, thereby utilizing the side-channel only when necessary to enhance performance. Our experiments show that: (i) CREAM mainly relies on concepts while achieving task performance on par with black-box models; and (ii) the embedded {texttt{C-C}} and {texttt{C$rightarrow$Y}} relationships ease model interventions and mitigate concept leakage.
19.
Azime, Israel Abebe; Kanubala, Deborah Dormah; Afonja, Tejumade; Fritz, Mario; Valera, Isabel; Klakow, Dietrich; Slusallek, Philipp
Accept or Deny? Evaluating LLM Fairness and Performance in Loan Approval across Table-to-Text Serialization Approaches Journal Article
In: CoRR, vol. abs/2508.21512, 2025.
@article{DBLP:journals/corr/abs-2508-21512,
title = {Accept or Deny? Evaluating LLM Fairness and Performance in Loan Approval across Table-to-Text Serialization Approaches},
author = {Israel Abebe Azime and Deborah Dormah Kanubala and Tejumade Afonja and Mario Fritz and Isabel Valera and Dietrich Klakow and Philipp Slusallek},
url = {https://doi.org/10.48550/arXiv.2508.21512},
doi = {10.48550/ARXIV.2508.21512},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2508.21512},
abstract = {Large Language Models (LLMs) are increasingly employed in high-stakes decision-making tasks, such as loan approvals. While their applications expand across domains, LLMs struggle to process tabular data, ensuring fairness and delivering reliable predictions. In this work, we assess the performance and fairness of LLMs on serialized loan approval datasets from three geographically distinct regions: Ghana, Germany, and the United States. Our evaluation focuses on the model’s zero-shot and in-context learning (ICL) capabilities. Our results reveal that the choice of serialization format significantly affects both performance and fairness in LLMs, with certain formats such as GReaT and LIFT yielding higher F1 scores but exacerbating fairness disparities. Notably, while ICL improved model performance by 4.9-59.6% relative to zero-shot baselines, its effect on fairness varied considerably across datasets. Our work underscores the importance of effective tabular data representation methods and fairness-aware models to improve the reliability of LLMs in financial decision-making.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Large Language Models (LLMs) are increasingly employed in high-stakes decision-making tasks, such as loan approvals. While their applications expand across domains, LLMs struggle to process tabular data, ensuring fairness and delivering reliable predictions. In this work, we assess the performance and fairness of LLMs on serialized loan approval datasets from three geographically distinct regions: Ghana, Germany, and the United States. Our evaluation focuses on the model’s zero-shot and in-context learning (ICL) capabilities. Our results reveal that the choice of serialization format significantly affects both performance and fairness in LLMs, with certain formats such as GReaT and LIFT yielding higher F1 scores but exacerbating fairness disparities. Notably, while ICL improved model performance by 4.9-59.6% relative to zero-shot baselines, its effect on fairness varied considerably across datasets. Our work underscores the importance of effective tabular data representation methods and fairness-aware models to improve the reliability of LLMs in financial decision-making.
20.
Gros, Timo P.; Müller, Nicola J.; Fiser, Daniel; Valera, Isabel; Wolf, Verena; Hoffmann, Jörg
Per-Domain Generalizing Policies: On Validation Instances and Scaling Behavior Journal Article
In: CoRR, vol. abs/2505.00439, 2025.
@article{DBLP:journals/corr/abs-2505-00439,
title = {Per-Domain Generalizing Policies: On Validation Instances and Scaling Behavior},
author = {Timo P. Gros and Nicola J. Müller and Daniel Fiser and Isabel Valera and Verena Wolf and Jörg Hoffmann},
url = {https://doi.org/10.48550/arXiv.2505.00439},
doi = {10.48550/ARXIV.2505.00439},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {CoRR},
volume = {abs/2505.00439},
abstract = {Recent work has shown that successful per-domain generalizing action policies can be learned. Scaling behavior, from small training instances to large test instances, is the key objective; and the use of validation instances larger than training instances is one key to achieve it. Prior work has used fixed validation sets. Here, we introduce a method generating the validation set dynamically, on the fly, increasing instance size so long as informative and this http://feasible.we/ also introduce refined methodology for evaluating scaling behavior, generating test instances systematically to guarantee a given confidence in coverage performance for each instance size. In experiments, dynamic validation improves scaling behavior of GNN policies in all 9 domains used.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Recent work has shown that successful per-domain generalizing action policies can be learned. Scaling behavior, from small training instances to large test instances, is the key objective; and the use of validation instances larger than training instances is one key to achieve it. Prior work has used fixed validation sets. Here, we introduce a method generating the validation set dynamically, on the fly, increasing instance size so long as informative and this http://feasible.we/ also introduce refined methodology for evaluating scaling behavior, generating test instances systematically to guarantee a given confidence in coverage performance for each instance size. In experiments, dynamic validation improves scaling behavior of GNN policies in all 9 domains used.