The various limitations of Generative AI, such as hallucinations and model failures, have made it crucial to understand the role of different modalities in Visual Language Model (VLM) predictions. Our work investigates how the integration of information from image and text modalities influences the performance and behavior of VLMs in visual question answering (VQA) and reasoning tasks. We measure this effect through answer accuracy, reasoning quality, model uncertainty, and modality relevance. We study the interplay between text and image modalities in different configurations where visual content is essential for solving the VQA task. Our contributions include (1) the Semantic Interventions (SI)-VQA dataset, (2) a benchmark study of various VLM architectures under different modality configurations, and (3) the Interactive Semantic Interventions (ISI) tool. The SI-VQA dataset serves as the foundation for the benchmark, while the ISI tool provides an interface to test and apply semantic interventions in image and text inputs, enabling more fine-grained analysis. Our results show that complementary information between modalities improves answer and reasoning quality, while contradictory information harms model performance and confidence. Image text annotations have minimal impact on accuracy and uncertainty, slightly increasing image relevance. Attention analysis confirms the dominant role of image inputs over text in VQA tasks. In this study, we evaluate state-of-the-art VLMs that allow us to extract attention coefficients for each modality. A key finding is PaliGemma's harmful overconfidence, which poses a higher risk of silent failures compared to the LLaVA models. This work sets the foundation for rigorous analysis of modality integration, supported by datasets specifically designed for this purpose.

Identifying predictive biomarkers, which forecast individual treatment effectiveness, is crucial for personalized medicine and informs decision-making across diverse disciplines. These biomarkers are extracted from pre-treatment data, often within randomized controlled trials, and have to be distinguished from prognostic biomarkers, which are independent of treatment assignment. Our study focuses on the discovery of predictive imaging biomarkers, aiming to leverage pre-treatment images to unveil new causal relationships. Previous approaches relied on labor-intensive handcrafted or manually derived features, which may introduce biases. In response, we present a new task of discovering predictive imaging biomarkers directly from the pre-treatment images to learn relevant image features. We propose an evaluation protocol for this task to assess a model's ability to identify predictive imaging biomarkers and differentiate them from prognostic ones. It employs statistical testing and a comprehensive analysis of image feature attribution. We explore the suitability of deep learning models originally designed for estimating the conditional average treatment effect (CATE) for this task, which previously have been primarily assessed for the precision of CATE estimation, overlooking the evaluation of imaging biomarker discovery. Our proof-of-concept analysis demonstrates promising results in discovering and validating predictive imaging biomarkers from synthetic outcomes and real-world image datasets.

Reliable application of machine learning-based decision systems in the wild is one of the major challenges currently investigated by the field. A large portion of established approaches aims to detect erroneous predictions by means of assigning confidence scores. This confidence may be obtained by either quantifying the model's predictive uncertainty, learning explicit scoring functions, or assessing whether the input is in line with the training distribution. Curiously, while these approaches all state to address the same eventual goal of detecting failures of a classifier upon real-life application, they currently constitute largely separated research fields with individual evaluation protocols, which either exclude a substantial part of relevant methods or ignore large parts of relevant failure sources. In this work, we systematically reveal current pitfalls caused by these inconsistencies and derive requirements for a holistic and realistic evaluation of failure detection. To demonstrate the relevance of this unified perspective, we present a large-scale empirical study for the first time enabling benchmarking confidence scoring functions w.r.t all relevant methods and failure sources. The revelation of a simple softmax response baseline as the overall best performing method underlines the drastic shortcomings of current evaluation in the abundance of publicized research on confidence scoring. Code and trained models are at https://github.com/IML-DKFZ/fd-shifts.