Machine learning models should be robust, in the sense of remaining predictively consistent under permissible variations. A model's predictions should ideally remain unchanged when it is replaced by a functionally equivalent one, or when its inputs are subject to minor, admissible perturbations. If such changes alter a prediction significantly, then the prediction is "ambiguous" with respect to the model. Models should abstain from making such ambiguous predictions and/or should flag them for human inspection, especially in high-stakes decision-making scenarios. However, in practice, such ambiguity is not easy to identify once a model is deployed. Here, the Robust Ambiguity Detection (RAD) framework is advanced for quantifying predictive ambiguity using two complementary metrics: Model-Space Consistency and Feature-Space Consistency. These two scores, the RAD Score-Pair, visualised through the RAD Plot, provide an interpretable characterisation of the sources of ambiguity and the actions a user may consider in response. RAD is evaluated on synthetic datasets with systematically controlled overlap, as well as several real-world datasets where the level of ambiguity cannot be directly inspected. Finally, we demonstrate a downstream application of RAD where samples are ranked by their RAD Pareto-Rank and the most ambiguous are abstained from prediction, achieving performance comparable to existing rejection-based approaches.
Many-shot in-context learning (ICL) lets vision-language models (VLMs) adapt from image--label demonstrations without weight updates, and is widely assumed to improve as more demonstrations are supplied. We show the opposite: as demonstrations accumulate, a subset of VLMs undergo an \emph{in-context collapse}, a sharp, sometimes catastrophic accuracy drop spanning synthetic classification, natural-image classification, and VQA benchmarks, in some models falling below chance while outputs remain well-formed. Across an open VLM panel ($0.5$B--$11$B) and a frontier model (Claude Sonnet 4.5), the collapse is graded. Two capabilities turn out to be dissociable: robustness to accumulating demonstrations and the ability to learn a novel rule in context, their combinations yield three reproducible regimes. A parameter-matched lesion-and-rescue causally localizes the collapse to the vision-language integration pathway: an adapter on the connector and early/mid layers restores genuine learning (remap accuracy $0.39!\rightarrow!0.91$ at 16 shots), while an equal-capacity adapter on the late readout does not. We propose \textsc{CircA}, whose core is a one-time integration vaccine: trained once on one synthetic task, it transfers collapse-resistance to unseen task families (chance$\rightarrow$$0.71$/$0.60$ on CIFAR/Fashion). The layers best for in-context integration are not the layers best for weight-based consolidation, the late readout achieves higher accuracy and less forgetting at fewer parameters. The collapse is an integration failure at the vision--language interface, correctable by a lightweight, transferable intervention.
To test how correct logical judgments respond to learned context, we prepend a soft prefix to an exactly labeled syllogistic reasoning benchmark while keeping the model fixed. Soft prefixes are opaque continuous vectors, so we characterize them through the behavior they induce across controlled variations in logical form and interface. By studying which prefixes succeed and how their effects generalize, we characterize how learned contextual pressure can override correct judgments and expose limits in a model's logical stability. Across Qwen3.6-35B-A3B MoE, Qwen3-8B, and Gemma 4 31B, learned prefixes redirect many correct answers and remain effective across unseen forms and interface changes. In repeated tests with Qwen3.6 MoE and Gemma, they outperform paired random controls in all 16 model--direction--split comparisons by 37 to 99 percentage points. Qwen3.6 MoE flip rates remain between 72% and 90% across wording and prompt changes, while Gemma validity prefixes retain 54% to 56% flip compared with less than 1% for matched random prefixes. Diagnostic tests show that the dominant effect is a broad preference for one answer meaning rather than fixed-symbol forcing or a logical operation that transfers reliably between tasks. The form of this bias differs across models. In both Qwen models, simple score models often predict which judgments will flip but not how far their margins will move, whereas Gemma's overall response is more closely approximated by the same models. These results show that the dominant behavioral effect of successful soft prefixes is a broad answer preference, while the remaining response reveals substantial model-specific differences in logical stability.
As neural networks are increasingly deployed in safety-critical domains, testing is essential to evaluate and improve their reliability. Existing testing methods, whether black-box or white-box, primarily use global mutation or coverage-guided strategies, both of which struggle to efficiently uncover diverse model failures while remaining proximate to the original data distribution and semantics. We propose BayesWarp, a testing framework that addresses this limitation by mutating decision-critical input regions identified via interpretable saliency techniques and adaptively guiding the testing process using an uncertainty-aware Bayesian Optimization strategy, enabling the discovery of diverse failures while preserving distributional and semantic proximity to the original data. Evaluation on MNIST, CIFAR-10, and ImageNet across six neural network models shows that BayesWarp improves failure discovery, failure diversity, test case quality, and critical neuron coverage under a fixed mutation budget. These results demonstrate that BayesWarp improves testing effectiveness. Moreover, fine-tuning with the generated failure cases leads to improvements in model performance.