Floris den Hengst

Abhishek Kolari, Mohammadhossein Khojasteh, Yifan Jiang et al.

While vision-language models (VLMs) have made remarkable progress on many popular visual question answering (VQA) benchmarks, it remains unclear whether they abstract and reason over depicted objects. Inspired by human object categorisation, object property reasoning involves identifying and recognising low-level details and higher-level abstractions. While current VQA benchmarks consider a limited set of object property attributes like size, they typically blend perception and reasoning, and lack representativeness in terms of reasoning and image categories. To this end, we introduce a systematic evaluation framework with images of three representative types, three reasoning levels of increasing complexity, and four object property dimensions driven by prior work on commonsense reasoning. We develop a procedure to instantiate this benchmark into ORBIT, a multi-level reasoning VQA benchmark for object properties comprising 360 images paired with a total of 1,080 count-based questions. Experiments with 12 state-of-the-art VLMs in zero-shot settings reveal significant limitations compared to humans, with the best-performing model only reaching 40\% accuracy. VLMs struggle particularly with realistic (photographic) images, counterfactual reasoning about physical and functional properties, and higher counts. ORBIT points to the need to develop methods for scalable benchmarking, generalize annotation guidelines, and explore additional reasoning VLMs. We make the ORBIT benchmark and the experimental code available to support such endeavors.

Syed Ihtesham Hussain Shah, Floris den Hengst, Aneta Lisowska et al.

Constraint inference is widely considered essential to align reinforcement learning agents with safety boundaries and operational guidelines by observing expert demonstrations. However, existing approaches typically assume homogeneous demonstrations (i.e., generated by a single expert or multiple experts with identical objectives). They also have limited ability to capture individual preferences and often suffer from computational inefficiencies. In this paper, we introduce Multi-Objective Constraint Inference (MOCI), a novel framework designed to jointly extract shared constraints and individual preferences from heterogeneous expert trajectories, where multiple experts pursue different objectives. MOCI effectively models and learns from diverse, and potentially conflicting, behaviors. Empirical evaluations demonstrate that MOCI significantly outperforms existing baselines, achieving improved predictive performance, and maintaining competitive computational efficiency on a standard grid-world benchmark. These results establish MOCI as an accurate, flexible, and computationally practical approach for real-world constraint inference and preference learning tasks.

Floris den Hengst, Ralf Wolter, Patrick Altmeyer et al.

We present Conformal Intent Classification and Clarification (CICC), a framework for fast and accurate intent classification for task-oriented dialogue systems. The framework turns heuristic uncertainty scores of any intent classifier into a clarification question that is guaranteed to contain the true intent at a pre-defined confidence level. By disambiguating between a small number of likely intents, the user query can be resolved quickly and accurately. Additionally, we propose to augment the framework for out-of-scope detection. In a comparative evaluation using seven intent recognition datasets we find that CICC generates small clarification questions and is capable of out-of-scope detection. CICC can help practitioners and researchers substantially in improving the user experience of dialogue agents with specific clarification questions.

Floris den Hengst, Inès Blin, Majid Mohammadi et al.

Conformal prediction (CP) is a powerful framework for quantifying uncertainty in machine learning models, offering reliable predictions with finite-sample coverage guarantees. When applied to classification, CP produces a prediction set of possible labels that is guaranteed to contain the true label with high probability, regardless of the underlying classifier. However, standard CP treats classes as flat and unstructured, ignoring domain knowledge such as semantic relationships or hierarchical structure among class labels. This paper presents hierarchical conformal classification (HCC), an extension of CP that incorporates class hierarchies into both the structure and semantics of prediction sets. We formulate HCC as a constrained optimization problem whose solutions yield prediction sets composed of nodes at different levels of the hierarchy, while maintaining coverage guarantees. To address the combinatorial nature of the problem, we formally show that a much smaller, well-structured subset of candidate solutions suffices to ensure coverage while upholding optimality. An empirical evaluation on three new benchmarks consisting of audio, image, and text data highlights the advantages of our approach, and a user study shows that annotators significantly prefer hierarchical over flat prediction sets.

Cas Oude Hoekstra, Floris den Hengst

Symbolic Regression (SR) is a well-established framework for generating interpretable or white-box predictive models. Although SR has been successfully applied to create interpretable estimates of the average of the outcome, it is currently not well understood how it can be used to estimate the relationship between variables at other points in the distribution of the target variable. Such estimates of e.g. the median or an extreme value provide a fuller picture of how predictive variables affect the outcome and are necessary in high-stakes, safety-critical application domains. This study introduces Symbolic Quantile Regression (SQR), an approach to predict conditional quantiles with SR. In an extensive evaluation, we find that SQR outperforms transparent models and performs comparably to a strong black-box baseline without compromising transparency. We also show how SQR can be used to explain differences in the target distribution by comparing models that predict extreme and central outcomes in an airline fuel usage case study. We conclude that SQR is suitable for predicting conditional quantiles and understanding interesting feature influences at varying quantiles.

Milena de Swart, Floris den Hengst, Jieying Chen

This paper addresses the critical need for detecting bias in government documents, an underexplored area with significant implications for governance. Existing methodologies often overlook the unique context and far-reaching impacts of governmental documents, potentially obscuring embedded biases that shape public policy and citizen-government interactions. To bridge this gap, we introduce the Dutch Government Data for Bias Detection (DGDB), a dataset sourced from the Dutch House of Representatives and annotated for bias by experts. We fine-tune several BERT-based models on this dataset and compare their performance with that of generative language models. Additionally, we conduct a comprehensive error analysis that includes explanations of the models' predictions. Our findings demonstrate that fine-tuned models achieve strong performance and significantly outperform generative language models, indicating the effectiveness of DGDB for bias detection. This work underscores the importance of labeled datasets for bias detection in various languages and contributes to more equitable governance practices.

Hüseyin Aydın, Kevin Godin-Dubois, Libio Goncalvez Braz et al.

Reinforcement learning (RL) offers a general approach for modeling and training AI agents, including human-AI interaction scenarios. In this paper, we propose SHARPIE (Shared Human-AI Reinforcement Learning Platform for Interactive Experiments) to address the need for a generic framework to support experiments with RL agents and humans. Its modular design consists of a versatile wrapper for RL environments and algorithm libraries, a participant-facing web interface, logging utilities, deployment on popular cloud and participant recruitment platforms. It empowers researchers to study a wide variety of research questions related to the interaction between humans and RL agents, including those related to interactive reward specification and learning, learning from human feedback, action delegation, preference elicitation, user-modeling, and human-AI teaming. The platform is based on a generic interface for human-RL interactions that aims to standardize the field of study on RL in human contexts.

Michal Nauman, Floris Den Hengst

In this paper, we propose World Model Policy Gradient (WMPG), an approach to reduce the variance of policy gradient estimates using learned world models (WM's). In WMPG, a WM is trained online and used to imagine trajectories. The imagined trajectories are used in two ways. Firstly, to calculate a without-replacement estimator of the policy gradient. Secondly, the return of the imagined trajectories is used as an informed baseline. We compare the proposed approach with AC and MAC on a set of environments of increasing complexity (CartPole, LunarLander and Pong) and find that WMPG has better sample efficiency. Based on these results, we conclude that WMPG can yield increased sample efficiency in cases where a robust latent representation of the environment can be learned.

Floris den Hengst, Mark Hoogendoorn, Frank van Harmelen et al.

Language systems have been of great interest to the research community and have recently reached the mass market through various assistant platforms on the web. Reinforcement Learning methods that optimize dialogue policies have seen successes in past years and have recently been extended into methods that personalize the dialogue, e.g. take the personal context of users into account. These works, however, are limited to personalization to a single user with whom they require multiple interactions and do not generalize the usage of context across users. This work introduces a problem where a generalized usage of context is relevant and proposes two Reinforcement Learning (RL)-based approaches to this problem. The first approach uses a single learner and extends the traditional POMDP formulation of dialogue state with features that describe the user context. The second approach segments users by context and then employs a learner per context. We compare these approaches in a benchmark of existing non-RL and RL-based methods in three established and one novel application domain of financial product recommendation. We compare the influence of context and training experiences on performance and find that learning approaches generally outperform a handcrafted gold standard.