Yulong Wu

Yulong Wu, Viktor Schlegel, Riza Batista-Navarro

As neural language models achieve human-comparable performance on Machine Reading Comprehension (MRC) and see widespread adoption, ensuring their robustness in real-world scenarios has become increasingly important. Current robustness evaluation research, though, primarily develops synthetic perturbation methods, leaving unclear how well they reflect real life scenarios. Considering this, we present a framework to automatically examine MRC models on naturally occurring textual perturbations, by replacing paragraph in MRC benchmarks with their counterparts based on available Wikipedia edit history. Such perturbation type is natural as its design does not stem from an arteficial generative process, inherently distinct from the previously investigated synthetic approaches. In a large-scale study encompassing SQUAD datasets and various model architectures we observe that natural perturbations result in performance degradation in pre-trained encoder language models. More worryingly, these state-of-the-art Flan-T5 and Large Language Models (LLMs) inherit these errors. Further experiments demonstrate that our findings generalise to natural perturbations found in other more challenging MRC benchmarks. In an effort to mitigate these errors, we show that it is possible to improve the robustness to natural perturbations by training on naturally or synthetically perturbed examples, though a noticeable gap still remains compared to performance on unperturbed data.

Yidan Sun, Viktor Schlegel, Srinivasan Nandakumar et al.

Data-driven decision support in high-stakes domains like healthcare and finance faces significant barriers to data sharing due to regulatory, institutional, and privacy concerns. While recent generative AI models, such as large language models, have shown impressive performance in open-domain tasks, their adoption in sensitive environments remains limited by unpredictable behaviors and insufficient privacy-preserving datasets for benchmarking. Existing anonymization methods are often inadequate, especially for unstructured text, as redaction and masking can still allow re-identification. Differential Privacy (DP) offers a principled alternative, enabling the generation of synthetic data with formal privacy assurances. In this work, we address these challenges through three key contributions. First, we introduce a comprehensive evaluation framework with standardized utility and fidelity metrics, encompassing nine curated datasets that capture domain-specific complexities such as technical jargon, long-context dependencies, and specialized document structures. Second, we conduct a large-scale empirical study benchmarking state-of-the-art DP text generation methods and LLMs of varying sizes and different fine-tuning strategies, revealing that high-quality domain-specific synthetic data generation under DP constraints remains an unsolved challenge, with performance degrading as domain complexity increases. Third, we develop a membership inference attack (MIA) methodology tailored for synthetic text, providing first empirical evidence that the use of public datasets - potentially present in pre-training corpora - can invalidate claimed privacy guarantees. Our findings underscore the urgent need for rigorous privacy auditing and highlight persistent gaps between open-domain and specialist evaluations, informing responsible deployment of generative AI in privacy-sensitive, high-stakes settings.

John K. Tsotsos, Iuliia Kotseruba, Alexander Andreopoulos et al.

Why do some continue to wonder about the success and dominance of deep learning methods in computer vision and AI? Is it not enough that these methods provide practical solutions to many problems? Well no, it is not enough, at least for those who feel there should be a science that underpins all of this and that we should have a clear understanding of how this success was achieved. Here, this paper proposes that the dominance we are witnessing would not have been possible by the methods of deep learning alone: the tacit change has been the evolution of empirical practice in computer vision and AI over the past decades. We demonstrate this by examining the distribution of sensor settings in vision datasets and performance of both classic and deep learning algorithms under various camera settings. This reveals a strong mismatch between optimal performance ranges of classical theory-driven algorithms and sensor setting distributions in the common vision datasets, while data-driven models were trained for those datasets. The head-to-head comparisons between data-driven and theory-driven models were therefore unknowingly biased against the theory-driven models.