Multimodal Large Language Models (MLLMs) have demonstrated great performance, yet they often suffer from severe modality misalignment when confronted with densely constrained spaces for interior design. Due to the loss of high-frequency local topological details and fine-grained aesthetic shifts during visual encoding, existing MLLMs frequently hallucinate, yielding physical spatial collisions and visual aesthetic dissonance. To address this, we propose Dual-prior Activation Residual Task-vectors Injection mechanism (DART-I) for MLLMs. It shifts the paradigm from lossy text-prompting to direct latent intervention, utilizing weak-to-strong deterministic rules to anchor the causal reasoning of MLLMs for interior design. Specifically, DART-I operates in three steps: it first explicitly extracts continuous spatial distance and color typography features from images using extremely lightweight weak experts; subsequently, it transforms these deterministic priors into directional task vectors via a linear projection network; these vectors are dynamically injected as residual terms into the latent space of the frozen MLLMs, steering MLLMs towards precise reasoning for interior design. Stepping outside the conventional paradigms, our method achieves precise reasoning without fine-tuning the MLLMs, effectively bypassing expensive computational costs and catastrophic forgetting. Extensive experiments on various benchmarks demonstrate the effectiveness and advantages of DART-I.
Timely detection of spreading events and accurate inference of their sources are central challenges in multi-layer networks, where heterogeneous topologies and interactions across layers shape diffusion. We formulate these coupled tasks as the Multi-layer outbreak Detection and source Inference (MDI) problem. MDI is a bi-objective vital-node identification problem that selects observer sets of fixed cardinality to minimize detection time and inference cost under limited sensing resources. We propose MOEA/D-TCM, a decomposition-based multi-objective evolutionary framework tailored to node-set optimization in multi-layer networks. It represents each solution as an observer set and combines a set-structure-adaptive evolutionary strategy with stable-state replacement and neighborhood adjustment to coordinate exploration and exploitation. Whereas the baseline methods return a ranked list or a fixed observer set, MOEA/D-TCM returns multiple non-dominated observer sets that represent different trade-offs between detection timeliness and inference cost. Experiments on 128 synthetic and empirical multilayer networks show that MOEA/D-TCM outperforms nine representative baselines on average, with mean improvements of 21.84% on synthetic networks and 28.90% on real-world networks. These results support bi-objective vital-node identification as a useful framework for monitoring and source localization in multi-layer diffusion systems.
Bo Gao, Yuxuan Yang, Xi Wang et al.· International Journal of Mod...· 0 citations
This paper investigates language-conditioned safety filtering, in which a Hamilton-Jacobi safety actor and critic are conditioned on language-specified constraints, and provides evidence that language-conditioned safety filters reduce constraint violations and exhibit partial transfer to unseen constraint instances.
