arXiv:2606.12412v1 Announce Type: new Abstract: Vision-language models (VLMs) project images into hundreds to thousands of visual tokens, making decoder inference expensive in both attention computation and KV-cache memory. Existing visual-token reduction methods largely follow a rank-and-remove paradigm: they score visual tokens, keep a compact subset, and permanently discard the rest. We show that this irreversible action is fragile because visual-token importance changes across decoder depth; tokens ranked low at one stage may become relevant in later layers, especially for grounding-sensitive queries. We propose Reroute, a training-free plug-in that replaces removal with recoverable routing. At each routing stage, selected vision tokens pass through decoder blocks, while deferred tokens bypass the stage and re-enter the candidate pool at the next routing decision. Reroute reuses existing attention-score ranking rules and stage-wise schedules, preserving the theoretical TFLOPs and KV-cache budget class of the pruning method it augments. Across FastV, PDrop, and N\"uwa variants on LLaVA-1.5 and Qwen backbones, reroute improves grounding under aggressive token reduction while maintaining general VQA performance. These results suggest that VLM token reduction should not be viewed only as irreversible pruning, but also as recoverable routing. The code can be found here: https://github.com/elmma/mllm-reroute/
Science Journals
arXiv:2606.09469v1 Announce Type: cross Abstract: Denial-of-service (DoS) and distributed denial-of-service (DDoS) mitigation requires separating malicious traffic from benign traffic while minimizing disruption to legitimate users. Prior work proposed mapping honeypot traffic partitioning to a weighted MaxCut problem and solving the resulting graphs with variational quantum algorithms. We extend this proof of principle direction with a reproducible event-level honeypot-to-QUBO pipeline, labeled temporal bipartite benchmark graphs with 16, 32, 66, and 110 event nodes, QAOA executions on IBM quantum hardware, classical heuristic baselines, a noiseless matrix product state reference, and a routing overhead analysis across quantum processor architectures. The largest benchmark is a 110-node, 181-edge instance executed on three IBM backends. Our results show that a shallow QAOA can execute real traffic partitioning workloads at the utility scale, while backend architecture and routing overhead affect objective quality, security metrics, and observed runtime. Because simple classical heuristics can solve the current labeled benchmark graphs, these experiments are not a quantum advantage claim. Instead, we deliberately use a fixed, shallow QAOA implementation to enable controlled comparisons across problem sizes and hardware architectures. This work establishes a hardware feasibility and architecture benchmark framework, and demonstrates that MaxCut cost, security quality, routing overhead, and runtime must be reported as separate metrics for cybersecurity relevant quantum optimization.
arXiv:2305.06145v2 Announce Type: replace Abstract: In cloth-changing person re-identification (CCReID), it is critical to learn clothes-invariant feature, which can provide discriminative ID features that remain robust against clothing changes. However, a spurious correlation currently limits existing ReID methods from effectively extracting these clothing-invariant features. This spurious correlation arises from clothing ownership: clothing is rarely shared across different identities, so models tend to memorize clothing cues for identity recognition, and this strategy generalizes poorly to unseen clothing. In this paper, we propose Causal Clothes-Invariant Learning (CCIL), which explicitly shifts CC-ReID from likelihood learning P (Y|X) to causal intervention learning P (Y|do(X)) to block the clothing shortcut. CCIL realizes this intervention through three modules: a Confounder Dictionary, an Intervention Module, and Disentangle Regularization. The causality-based modeling makes the entire model naturally clothes-invariant, effectively preventing the capture of spurious correlations in feature learning. Extensive experiments validate the effectiveness of CCIL. On PRCC and DeepChange datasets, CCIL achieves Rank-1 accuracies of 66.4% and 59.2%, outperforming state-of-the-art methods by 1.4 and 4.1 percentage points, respectively.
arXiv:2404.10119v2 Announce Type: replace Abstract: We developed an open-source scalar wave transport model to estimate the generalized scattering matrix (S matrix) of a disordered medium in the diffusion regime. The term generalized refers to the incorporation of evanescent wave field modes alongside propagating modes in the estimation of the S matrix. To achieve this, we employed the scalar Kirchhoff-Helmholtz boundary integral formulation together with the Green's function perturbation method, thereby extending the conventional Fisher-Lee relations to include evanescent modes. The estimated S matrix, which satisfies the generalized unitarity and reciprocity relations, is modeled for a 2D disordered waveguide. The generalized transmission matrix contained within the S matrix is utilized to estimate the optimal phase-conjugate wavefront for focusing onto an evanescent mode. The phenomenon of a universal transmission value of 2/3 for such an optimal phase conjugate wavefront is demonstrated in the context of evanescent wave mode focusing through a diffusive disorder. The presented code framework may be of interest to wavefront shaping researchers for visualizing and estimating wave transport properties in general.
arXiv:2408.13043v3 Announce Type: replace Abstract: Differential equations posed on quadratic matrix Lie groups arise in the context of classical mechanics and quantum dynamical systems. Lie group numerical integrators preserve the constants of motions defining the Lie group. Thus, they respect important physical laws of the dynamical system, such as unitarity and energy conservation in the context of quantum dynamical systems, for instance. In this article we develop a high-order commutator free Lie group integrator for non-autonomous differential equations evolving on quadratic Lie groups. Instead of matrix exponentials, which are expensive to evaluate and need to be approximated by appropriate rational functions in order to preserve the Lie group structure, the proposed method is obtained as a composition of Cayley transforms which naturally respect the structure of quadratic Lie groups while being computationally efficient to evaluate. Unlike Cayley--Magnus methods the method is also free from nested matrix commutators.
arXiv:2605.15161v2 Announce Type: replace Abstract: Understanding when linear immersions of nonlinear dynamical systems exist is important since such immersions allow us to leverage the rich tools of linear system theory to analyze nonlinear dynamics. Recently, Liu et al. (2023) showed that continuous-time dynamical systems that admit countably many but more than one omega-limit sets cannot be immersed into finite dimensional linear systems with a one-to-one and continuous mapping. In this paper, we extend these results to discrete-time dynamics and show that similar obstructions exist also in discrete time. We further consider a generalization involving alpha-limit sets. Several examples are provided to demonstrate the results.
arXiv:2605.16651v2 Announce Type: replace Abstract: Explanation mechanisms are increasingly used to support transparency and trust in vision-language models (VLMs), particularly in settings where model decisions require human oversight. However, the robustness of these explanations remains insufficiently understood. In this work, we investigate whether explanation heatmaps in VLMs, particularly CLIP-based models, faithfully reflect model reasoning under adversarial conditions. We show that explanation maps can be systematically manipulated while preserving the model's original prediction, revealing a disconnect between predictive behavior and explanation faithfulness. To study this vulnerability, we introduce X-Shift, a novel grey-box attack that perturbs patch-level visual representations to redirect explanation heatmaps toward semantically irrelevant regions without altering the predicted output. Unlike conventional adversarial attacks that aim to induce misclassification, X-Shift specifically targets the integrity of the explanation process itself. The attack operates without modifying model parameters and generalizes across multiple CLIP architectures and explanation methods. We evaluate the proposed approach on ImageNet-1k, MS-COCO, and Flickr30K, demonstrating consistent degradation in explanation alignment under imperceptible perturbations while maintaining prediction stability. Furthermore, standard prediction-oriented adversarial attacks fail to reproduce the same explanation-shifting behavior even under substantially larger perturbation budgets. Our findings highlight a fundamental limitation of current explanation mechanisms in VLMs and raise concerns about their use as reliable indicators of model trustworthiness in high-impact applications.
arXiv:2505.03296v2 Announce Type: replace Abstract: We present Mixture of Discrete-time Gaussian Processes (MiDiGap), a novel approach for flexible policy representation and imitation learning in robot manipulation. MiDiGap enables learning from as few as five demonstrations using only camera observations and generalizes across a wide range of challenging tasks. It excels at long-horizon behaviors such as making coffee, highly constrained motions such as opening doors, dynamic actions such as scooping with a spatula, and multimodal tasks such as hanging a mug. MiDiGap learns these tasks on a CPU in less than a minute and scales linearly to large datasets. We also develop a rich suite of tools for inference-time steering using evidence such as collision signals and robot kinematic constraints. This steering enables novel generalization capabilities, including obstacle avoidance and cross-embodiment policy transfer. MiDiGap achieves state-of-the-art performance on diverse few-shot manipulation benchmarks. On constrained RLBench tasks, it improves policy success by 76 percentage points and reduces trajectory cost by 67%. On multimodal tasks, it improves policy success by 48 percentage points and increases sample efficiency by a factor of 20. In cross-embodiment transfer, it more than doubles policy success. We make the code publicly available at https://midigap.cs.uni-freiburg.de.
arXiv:2505.05305v2 Announce Type: replace Abstract: The global energy landscape is undergoing a transformative shift towards renewable energy and advanced storage solutions, driven by the urgent need for sustainable and resilient power systems. Isolated offshore communities, such as islands and offshore platforms, which traditionally rely on mainland grids or diesel generators, stand to gain significantly from renewable energy integration. Promising offshore renewable technologies include wind turbines, wave and tidal energy converters, and floating photovoltaic systems, paired with a storage solution like battery energy storage systems. This paper introduces a renewable energy microgrid optimizer (REMO), a tool designed to identify the optimal sizes of renewable generation and storage resources for offshore microgrids. A key challenge in such models is accurately accounting for battery degradation costs. To address this, the REMO model integrates a deep neural network-based battery degradation (DNN-BD) module, which factors in variables like ambient temperature, charge/discharge rates, state of charge, depth of discharge and battery health. Simulations on six test regions demonstrate that the REMO-DNN-BD approach minimizes lifetime energy costs while maintaining high reliability and sustainability, making it a viable design solution for offshore microgrid systems.
arXiv:2506.18401v3 Announce Type: replace Abstract: This paper studies the relation between agreement and strongly linearizable implementations of various objects. This leads to new results about implementations of concurrent objects from various primitives including window registers, interfering primitives and stacks. We identify that lock-free, respectively, wait-free, strongly linearizable implementations of several concurrent objects entail a form of agreement that does not require full consensus but cannot be implemented in a strongly-linearizable manner with combinations of non-universal primitives. In both cases, lock-free and wait-free, this form of agreement requires a distinguished process to referee a competition that involves all other processes. Our results show that consistent refereeing of such competitions (i.e., once resolved, the outcome cannot be revised in any extension of the execution) requires high coordination power. More specifically, two contest objects are defined and used to help characterize coordination constraints imposed by strong linearizability in lock-free and wait-free implementations, respectively. Both objects are strictly weaker than consensus, in the sense that they have a wait-free linearizable (in fact, decisively linearizable) implementation from reads and writes. The contest objects capture strong linearizability in two complementary ways. First, they admit strongly linearizable implementations from several high-level objects such as queues, snapshots, and counters, so impossibility results for the contest objects immediately carry over to these objects. Second, they admit powerful impossibility results for strong linearizability that involve window registers, interfering primitives and stacks, which are non-universal.
arXiv:2506.19604v2 Announce Type: replace Abstract: Polynomial factorisation is a fundamental problem in computational algebra. Over the past half century, a variety of algorithmic techniques have been developed to tackle different variants of this problem. In parallel, algebraic complexity theory classifies polynomials into complexity classes based on their computational hardness. This raises a natural question: Are these complexity classes closed under factorisation? In this survey, we revisit pivotal techniques in polynomial factorisation: Hensel lifting, Newton iteration, and Lagrange inversion. These techniques have played an essential role in resolving key factoring questions in algebraic complexity for more than half a century. We examine and organise the known results through the lens of these techniques, discussing their underlying mathematical equivalence while reflecting on how their applications vary depending on the problem context. We focus on prominent algebraic complexity classes, including $\text{VP}$ (circuits of polynomial size and degree), its closure $\overline{\text{VP}}$, the class $\text{VNP}$ (verifier circuits of polynomial size and degree), $\text{VBP}$ (polynomial-size branching programs), $\text{VF}$ (polynomial-size formulas), and $\text{VP}_{\text{nb}}$ (circuits of polynomial size and exponential degree). We also discuss bounded-depth circuits and sparse polynomials. Along the way, we highlight several unresolved open problems.
arXiv:2511.13452v2 Announce Type: replace Abstract: Understanding how group interactions influence opinion dynamics is fundamental to the study of collective behavior. In this work, we propose and study a model of opinion dynamics on $d$-uniform hypergraphs, where individuals interact through group-based (higher-order) structures rather than simple pairwise connections. Each one of the two opinions $A$ and $B$ is characterized by a quality, $Q_A$ and $Q_B$, and agents update their opinions according to a general mechanism that takes into account the weighted fraction of agents supporting either opinion and the pooling error, $\alpha$, a proxy for the information lost during the interaction. Through bifurcation analysis of the mean-field model, we identify two critical thresholds, $\alpha_{\text{crit}}^{(1)}$ and $\alpha_{\text{crit}}^{(2)}$, which delimit stability regimes for the consensus states. These analytical predictions are validated through extensive agent-based simulations on both random and scale-free hypergraphs. Moreover, the analytical framework demonstrates that the bifurcation structure and critical thresholds are independent of the underlying topology of the higher-order network, depending solely on the parameters $d$, i.e., the size of the interaction groups, and the quality ratio. Finally, we bring to the fore a nontrivial effect: the large sizes of the interaction groups, could drive the system toward the adoption of the worst option.
arXiv:2512.03077v2 Announce Type: replace Abstract: The accelerated development, deployment and adoption of artificial intelligence systems has been fuelled by the increasing presence of big tech in the AI field. This trend has been accompanied by growing ethical concerns and intensified societal and environmental impacts. This position paper argues that irresponsible AI development is strongly driven by big tech's influence and involvement in the field. First, we examine the growing and disproportionate influence of big tech in AI research and argue that its drive for scaling and general-purpose systems is fundamentally at odds with the responsible, ethical, and sustainable development of AI. Second, we review key current environmental and societal negative impacts of AI and trace their connections to big tech's influence. Third, we discuss the underlying economic forces driving big tech's actions. Finally, as a call to action, we invite AI researchers to counter big tech's influence in irresponsible AI development through strategies that build on the responsibility of implicated actors and collective action.
arXiv:2604.25018v2 Announce Type: replace Abstract: The Internet of Everything (IoE) represents an evolution of the Internet of Things (IoT) by integrating people, data, processes, and things into a unified intelligent ecosystem. IoE aims to enhance automation, decision-making, and service efficiency across multiple application domains such as smart cities, healthcare, industry, and next-generation wireless networks. This paper provides a structured overview of the IoE concept, its core components, architectural foundations, enabling technologies, and major research challenges. Finally, open research directions toward 6G-enabled intelligent IoE systems are discussed, with emphasis on scalability, security, privacy, and energy efficiency.
arXiv:2509.15680v3 Announce Type: replace Abstract: We present SAM, a State-space Audio-language Model that integrates an audio encoder with a Mamba-2 backbone. SAM-2.7B achieves 21.1 mAP on AudioSet and 17.6 SPICE on AudioCaps, matching or surpassing larger 7B transformer-based models with fewer parameters. We further provide the first systematic, representation-level analysis of how SSMs interact with audio encoder outputs: (1) joint audio encoder finetuning is essential, supported by accuracy gains and observed adaptation of token representation rank and similarity across different SSM sizes; (2) despite linear scaling, SSMs benefit more from compact, information-rich audio token representations than from excessively long token sequences; and (3) incorporating instruction-following supervision substantially improves reasoning ability, boosting MMAU-Sound accuracy from 22.8 to 56.8. Through comprehensive experiments and analysis, we establish practical design principles for SSMs as strong, scalable backbones for audio-language models.
arXiv:2510.01157v4 Announce Type: replace Abstract: Speech language models (SLMs) are systems of systems: independent components that unite to achieve a common goal. Despite their heterogeneous nature, SLMs are often studied end-to-end; how information flows through the pipeline remains obscure. We investigate this question through the lens of backdoor attacks. We first establish that backdoors can propagate through the SLM, leaving all tasks highly vulnerable. From this, we design a component analysis to discover the role each component takes in backdoor learning. We find that backdoor persistence or erasure is highly dependent on the targeted component. Beyond propagation, we examine how backdoors are encoded in shared multitask embeddings, showing that poisoned samples are not directly separable from benign ones, challenging a common separability assumption used in filtering defenses. Our findings emphasize the need to treat multimodal pipelines as intricate systems with unique vulnerabilities, not solely extensions of unimodal ones.
arXiv:2510.02660v2 Announce Type: replace Abstract: When researchers claim AI systems possess ToM or mental models, they are fundamentally discussing behavioral predictions and bias corrections rather than genuine mental states. This position paper argues that the current discourse conflates sophisticated pattern matching with authentic cognition, missing a crucial distinction between simulation and experience. While recent studies show LLMs achieving human-level performance on ToM laboratory tasks, these results are based only on behavioral mimicry. More importantly, the entire testing paradigm may be flawed in applying individual human cognitive tests to AI systems, but assessing human cognition directly in the moment of human-AI interaction. I suggest shifting focus toward mutual ToM frameworks that acknowledge the simultaneous contributions of human cognition and AI algorithms, emphasizing the interaction dynamics, instead of testing AI in isolation.
arXiv:2604.25363v2 Announce Type: replace Abstract: Regression testing in Continuous Integration (CI) pipelines is increasingly costly due to the growing size and execution frequency of test suites. Test Case Prioritization (TCP) mitigates this problem by reordering tests to expose faults earlier. However, most existing techniques rely primarily on historical execution data and coverage metrics, neglecting the rich structural information contained in code changes. This paper proposes a commit-aware, learning-based TCP method that combines structural properties of version-control diffs, test coverage relations, and historical execution behavior into a unified predictive model. Given a new commit, the method estimates the probability that each test suite will reveal at least one failure and prioritizes test execution accordingly. We evaluate our method on five Defects4J projects using a leave-one-project-out cross-project validation setting. Results show that the commit-aware TCP significantly outperform non-commit-aware-baselines in both classification and prioritization effectiveness. Our findings show that including commit structural semantics substantially enhances regression fault detection and enables robust, generalizable learning-based TCP in CI environments.
arXiv:2605.11911v2 Announce Type: replace Abstract: Predictive Coding (PC) is an influential account of cortical learning. Much of recent work has focused on comparing PC to Backpropagation (BP) to find whether PC offers any advantages. Small scale experiments show that PC enables learning that is more sample efficient and effective in many contexts, though a thorough theoretical understanding of the phenomena remains elusive. To address this, we quantify the efficiency of learning in BP and PC through a metric called ``target alignment'', which measures how closely the change in the output of the network is aligned to the output prediction error. We then derive and empirically validate analytical expressions for target alignment in Deep Linear Networks. We show that learning in PC is more efficient than BP, which is especially pronounced in deep, narrow and pre-trained networks. We also derive exact conditions for guaranteed optimal target alignment in PC and validate our findings through experiments. We study full training trajectories of linear and non-linear models, and find the predicted benefits of PC persist in practice even when some assumptions are violated. Overall, this work provides a mechanistic understanding of the higher learning efficiency observed for PC over BP in previous works, and can guide how PC should be parametrised to learn most effectively.
arXiv:2510.18289v3 Announce Type: replace Abstract: Food assistance referral requires conversational agents to translate underspecified, often noisy help-seeking dialogues into locally valid resource recommendations. We present Food4All, an agentic food-resource referral framework and benchmark grounded in 686 structured Indiana food resources. Food4All couples a food-specific search tool with 300 multi-turn evaluation tasks spanning single food needs, composite cases with access or document constraints, and five non-ideal user interaction traits: unreasonable demands, rambling responses, impatience, incomplete answers, and inconsistent information. We evaluate six Large Language Models (LLMs) on requirement grounding, resource retrieval, final referral correctness, and interaction efficiency. Although the strongest model achieves 96.33% referral accuracy, our diagnostics reveal persistent failures in grounding schedule, eligibility, intake, and document constraints, as well as failures to preserve valid retrieved resources in the final recommendation. Trait-level analysis further shows that different non-ideal behaviors stress different parts of the referral pipeline. Food4All provides a controlled testbed for studying tool-calling agents in constraint-sensitive food assistance referral under realistic user interaction challenges.
arXiv:2510.22335v2 Announce Type: replace Abstract: Reconstructing visual stimuli from fMRI signals is a central challenge bridging machine learning and neuroscience. Recent diffusion-based methods typically map fMRI activity to a single neural embedding, using it as static guidance throughout the entire generation process. However, this fixed guidance collapses hierarchical neural information and is misaligned with the stage-dependent demands of image reconstruction. In response, we propose MindHier, a coarse-to-fine fMRI-to-image reconstruction framework built on scale-wise autoregressive modeling. MindHier introduces three components: a Hierarchical fMRI Encoder to extract multi-level neural embeddings, a Hierarchy-to-Hierarchy Alignment scheme to enforce layer-wise correspondence with CLIP features, and a Scale-Aware Coarse-to-Fine Neural Guidance strategy to inject these embeddings into autoregression at matching scales. These designs make MindHier an efficient and cognitively aligned alternative to diffusion-based methods by enabling a hierarchical reconstruction process that synthesizes global semantics before refining local details, akin to human visual perception. Extensive experiments on the NSD dataset show that MindHier achieves superior semantic fidelity, 4.67$\times$ faster inference, and more deterministic results than the diffusion-based baselines.
arXiv:2512.17467v2 Announce Type: replace Abstract: CsSn$_x$Ge$_{1-x}$I$_3$ as lead-free perovskites are promising for next generation NIR emitting perovskite LEDs due to their tunable bandgaps and stability. However, they suffer from poor light extraction efficiency, and accurate composition-specific optical data for these materials remain scarce. This study presents a DFT-FDTD framework to optimize light extraction via compositional tuning and plasmonic enhancement. First, DFT calculations were performed to obtain composition-specific complex refractive index and extinction coefficient values for $x = 0, 0.25, 0.5, 0.75$, and $1$. Results show bandgap increased from 1.331 eV for CsSnI$_3$ to 1.927 eV for CsGeI$_3$ with increasing Ge content, while refractive index ranges from 2.2 to 2.6 across compositions. These optical constants were then used as inputs for FDTD simulations of a PeLED structure with optimized Au/SiO$_2$ core-shell nanorods for plasmonic enhancement. A 12.1-fold Purcell enhancement was achieved for CsSn$_{0.25}$Ge$_{0.75}$I$_3$, while light extraction efficiency reached 25% for CsSn$_{0.5}$Ge$_{0.5}$I$_3$. LEE enhancement of 36% was obtained for CsSnI$_3$, and spectral overlap between emitter and plasmon resonance reached 96% for Sn-rich compositions. Design guidelines indicate CsSn$_{0.5}$Ge$_{0.5}$I$_3$ offers optimal balance of extraction efficiency (25%), Purcell enhancement (5.3$\times$), spectral overlap (93%), and oxidation stability for wearable and flexible optoelectronic applications, while CsSn$_{0.25}$Ge$_{0.75}$I$_3$ is recommended for applications prioritizing spontaneous emission rate.
arXiv:2512.18033v2 Announce Type: replace Abstract: We construct a local Fortin projection for the Scott-Vogelius finite element pair for polynomial degree $k \ge 4$ on general shape-regular triangulations in two dimensions. In particular, the triangulation may contain singular vertices. In addition to preserving the divergence in the dual of the pressure space, the projection preserves discrete boundary data and satisfies local stability estimates.
arXiv:2601.00181v3 Announce Type: replace Abstract: We address two persistent gaps in Emotion Recognition in Conversation: which modeling choices materially affect performance, and how recognition findings connect to interpretable discourse-level patterns. We study both through a systematic investigation on IEMOCAP with cross-dataset validation on MELD. For recognition, we run controlled ablations with 10 random seeds and paired significance tests with multiple-comparisons correction, yielding three findings. First, conversational context is the dominant factor, but performance saturates quickly: roughly 90% of the gain is captured within the most recent 10-30 preceding turns, depending on the label set. Second, hierarchical sentence representations help most in utterance-only settings and show a clear advantage on MELD, but their benefit disappears once turn-level context is available, suggesting that conversational history subsumes much of the intra-utterance structure. Third, integrating an external affective lexicon does not improve results, consistent with pretrained encoders already capturing most of the affective signal needed for ERC. Under a strictly causal setting, our simple models achieve strong performance (82.69% 4-way; 67.07% 6-way weighted F1), showing that competitive accuracy is achievable without future turns. For linguistic analysis, we examine 5,286 discourse-marker occurrences and find a reliable association between emotion and marker position (p < .0001). Sad utterances show reduced left-periphery marker usage (21.9%) relative to other emotions (28-32%), consistent with accounts linking left-periphery markers to active discourse management. This aligns with our recognition results, where Sad benefits most from conversational context (+22 percentage points), suggesting sadness may be more context-dependent than emotions with stronger local pragmatic cues.
arXiv:2604.13326v2 Announce Type: replace Abstract: The robustness of machine learning models can be compromised by spurious correlations between non-causal features in the input data and target labels. A common way to test for such correlations is to train on data where the label is strongly tied to some non-causal cue, then evaluate on examples where that tie no longer holds. This idea is well established for classification tasks, but for semantic segmentation the specific failure modes are not well understood. We show that a model may achieve reasonable overlap while assigning the wrong semantic label, swapping one plausible foreground class for another, even when object boundaries are largely correct. We focus on this semantic label-flip behaviour and quantify it with a simple diagnostic (Flip) that counts how often ground truth foreground pixels are assigned the wrong foreground identity while remaining predicted as foreground. In a setting where category and scene are correlated during training, increasing the correlation consistently widens the gap between common and rare test conditions and increases these within-object label swaps on counterfactual groups. Overall, our results motivate assessing segmentation robustness under distribution shift beyond overlap by decomposing foreground errors into correct pixels, flipped-identity pixels, and missed-to-background pixels. We also propose an entropy-based, ground truth label-free `flip-risk' score, which is computed from foreground identity uncertainty, and show that it can flag flip-prone cases at inference time. Code is available at https://github.com/acharaakshit/label-flips.