8512 papers
This paper proposes a goal-driven, system-level security framework that integrates system modeling, Attack-Defense Trees, and CVSS scoring to assess and mitigate risks in LLM-based systems, demonstrating through a healthcare case study that diverse threats often converge on shared system choke points, enabling targeted defenses to effectively reduce exploitability.
This paper introduces the Dual-Stream Transformer, an architecture that decomposes the residual stream into separate token and context streams with tunable mixing strategies to achieve a balance between high interpretability and minimal performance loss while demonstrating robustness to attention amplification.
This paper proposes a human-centred out-of-distribution spectrum that redefines perceptual difficulty based on human accuracy to enable principled comparisons of model-human error alignment, revealing that while vision-language models show the most consistent alignment across conditions, the relative performance of CNNs and ViTs depends on the specific regime of perceptual challenge.
This paper proposes a lightweight, self-supervised framework that augments a frozen speech enhancement backbone with low-rank adapters, enabling efficient on-device adaptation to dynamic real-world noise conditions by updating fewer than 1% of parameters while achieving significant signal quality improvements.
This paper introduces a conditional diffusion-transformer framework that generates diverse ensembles of 3D *E. coli* genome conformations guided by Hi-C contact maps, effectively reconstructing heterogeneous structures whose ensemble averages align with experimental data while preserving conformational diversity.
This paper reveals that MCP-based AI systems are fundamentally insecure due to a lack of caller identity authentication, which allows persistent authorization states and missing per-tool checks to enable unauthorized access to sensitive operations by untrusted callers.
This paper demonstrates that vision-language models can recover and generalize taxonomic knowledge (hypernyms) from language representations even when deprived of explicit visual evidence during training, provided that the counterfactual image-label mappings maintain high visual coherence within categories.
This paper proposes and validates the Late Fusion Architecture (LFA), a transformer variant that enforces interpretability by design through architectural stream independence, which separates symbolic and contextual processing to prevent premature entanglement and significantly improve model stability and functional modularity compared to standard transformers.
This paper proposes a novel end-to-end joint neural system that simultaneously optimizes concept recognition, assertion classification, and relation extraction for clinical text, significantly outperforming traditional pipeline baselines across all three tasks.
This paper proposes the Hierarchical Autonomy Evolution (HAE) framework to address critical security vulnerabilities in evolving AI agents by categorizing threats and defenses across three tiers: Cognitive, Execution, and Collective Autonomy.
SeDa is a unified framework that aggregates over 7.6 million datasets from more than 200 platforms to enable trustworthy, semantically enriched, and multi-entity augmented exploration through standardized metadata, a dynamic tag graph, and provenance assurance.
This paper establishes a unified theoretical framework demonstrating that drifting models, which optimize kernel-based mean-shift discrepancies, are mathematically equivalent to score-matching objectives on kernel-smoothed distributions, thereby precisely connecting them to diffusion models and clarifying their relationship with Distribution Matching Distillation.
InterReal is a unified physics-based imitation learning framework that enables humanoid robots to robustly learn and execute complex human-object interaction skills in real-world settings through a novel motion data augmentation scheme and an automatic reward learner.
This paper introduces SketchGraphNet, a memory-efficient hybrid graph transformer that models free-hand sketches as structured graphs to achieve state-of-the-art recognition accuracy on the newly constructed 3.44-million-sample SketchGraph benchmark while significantly reducing computational resource requirements.
This paper proposes a neural dynamics-informed pre-trained framework that overcomes the limitations of traditional atlas-based methods by extracting personalized neural activity representations to guide brain parcellation and correlation estimation, thereby achieving superior performance in constructing personalized brain functional networks across heterogeneous scenarios.
This paper introduces UniLongGen, a training-free inference strategy that improves long-horizon interleaved image generation by dynamically curating context to discard accumulated visual noise, thereby overcoming the reliability collapse caused by dense visual token interference in unified multimodal models.
DreamSAC is a framework that enhances extrapolative generalization in physics simulations by combining an unsupervised symmetry exploration strategy, which actively probes conservation laws via a Hamiltonian-based curiosity bonus, with a Hamiltonian-based world model that learns invariant physical states from raw observations through a novel contrastive objective.
The paper introduces COOL-MC, a framework that verifies and explains reinforcement learning policies for multi-bridge network maintenance by applying probabilistic model checking and explainability methods to a PRISM-encoded Markov decision process, thereby revealing safety violation probabilities and systematic policy biases.
This paper proposes a learning-free text-to-speech framework that generates L2-accented speech by applying phonological rules to phoneme sequences within a multilingual TTS model, enabling explicit accent control without requiring large-scale accented training datasets.
This paper introduces a novel Speech Generation Speaker Poisoning (SGSP) framework to address privacy risks in zero-shot text-to-speech by modifying trained models to prevent the generation of specific speaker identities while maintaining utility for others, demonstrating effective protection for up to 15 speakers but revealing scalability challenges with larger sets due to identity overlap.