Task-specific LoRA adapters offer a modular way to specialize large language and vision-language models. However, existing adapter composition methods are mostly static and cannot adapt to individual test inputs. To address these issues, we propose LiST, a label-free test-time LoRA fusion framework that converts an existing LoRA bank into a target-conditioned local simplex and searches sample-specific fusion weights at inference time. LiST builds joint task representations from LoRA parameter anchors and prompt-level behavior vectors, retrieves neighboring adapters as a local search space, and performs branch-preserving fusion without updating the backbone or adapters. Candidate weights are selected by a prompt-level energy with prior, geometric, and stochastic-consistency constraints, and are deployed only when they pass a safe acceptance rule. Otherwise, LiST falls back to a target-conditioned prior. Experiments on multimodal and language benchmarks show that LiST outperforms static LoRA merging and conventional test-time adaptation baselines, while preserving task-specific adapter utility and improving robustness on unseen tasks.

Text-to-image models learn associations between concepts - in the case of this paper, people’s professions, which we refer to as roles - and visual attributes. These associations can underpin many observed forms of stereotypical bias. A key open question in this area is whether these associations are stable or change when visual representations of people in professional roles are placed in different prompted contexts. We introduce contextbias, a controlled evaluation framework, and contextbench, a benchmark spanning 92 roles and 1,656 semantically controlled prompts, designed to isolate the effect of contextual variation on role-linked visual representations. Evaluating four state-of-the-art models on 66,240 generated images, we find that placing a role in a semantically unrelated context does not suppress role-linked attributes; instead, cross-role attribute concentration increases (pooled BI +0.047). Demographic cues, characteristic garments, and role-specific tools remain highly prevalent across context-free, related, and unrelated conditions, and are robust to semantic prompt reformulation. Scene composition and camera framing show the greatest context-sensitivity. These findings reveal a form of stereotypical persistence that remains largely invisible to context-free evaluations, highlighting the need for controlled contextual variation in bias benchmarking. Code and dataset: https://huggingface.co/datasets/shaghayegh/ContextBias , https://github.com/Sina-Emami/ContextBias

Neural networks over non-Euclidean manifolds have attracted increasing attention across various applications. A fundamental step in these networks is to map Euclidean features onto a target manifold. Existing methods mainly use exponential maps and projections, operate through tangent or ambient Euclidean spaces, and may therefore distort intrinsic manifold geometry. On the other hand, point-to-hyperplane distances have recently shown great success in building manifold-valued networks. Based on this, we propose the Point-to-Hyperplane Distance Map (PHDM), a parameter-free mapping that extends Euclidean fully connected layers to Euclidean-to-manifold transformations. Specially, the exponential map is recovered as a pseudo-distance-based special case. We manifest PHDM on the Lorentz, Poincaré, spherical, and projected-spherical models, covering both negative and positive curvature. Experiments on arithmetic reasoning across four LLMs, image classification, graph node classification, and link prediction demonstrate consistent improvements over the corresponding baselines. The code is available at https://github.com/Longchentong/PHDM.