3470 papers
The paper introduces CLIPO, a method that integrates contrastive learning into policy optimization to generalize Reinforcement Learning with Verifiable Rewards (RLVR) by capturing invariant structures across correct reasoning paths, thereby mitigating hallucinations and improving the generalization and robustness of Large Language Models.
This paper argues that the "Lost in the Middle" phenomenon in large language models is an inherent geometric property of causal decoder architectures present at initialization, caused by the interplay of causal masking and residual connections that creates a structurally hostile "dead zone" in the middle of the context, a bias that persists even after standard pretraining.
The paper proposes AR-VLA, a standalone autoregressive Action Expert that maintains long-lived memory to generate continuous, context-aware action sequences, effectively addressing the frequency mismatch between fast control and slow reasoning while outperforming traditional reactive Vision-Language-Action models in trajectory smoothness and task success.
This paper proposes an agentic control center that automates the continuous optimization of data products by employing specialized AI agents to generate supporting assets, monitor quality metrics, and integrate human-in-the-loop oversight to balance automation with trust.
This paper proposes a unified framework for the generation-recognition asymmetry in formal language theory by identifying six distinct dimensions of divergence, challenging the oversimplified view that generation is inherently easy while parsing is hard, and exploring the implications of these operational differences for fields ranging from compiler design to large language models.
This paper demonstrates that projecting users and entities into a joint social embedding space derived from large-scale Twitter data enables effective zero-shot cross-domain preference modeling and personalization, revealing that socio-demographic factors encoded in these embeddings correlate with user interests across diverse topics.
The paper proposes Mashup Learning, a method that accelerates LLM finetuning and improves downstream accuracy by identifying and merging relevant historical checkpoints to serve as an optimized initialization for new tasks, thereby reducing training time by up to 37% compared to training from scratch.
This paper introduces the first systematic framework for identifying and exploiting "compatibility-abusing attacks" in the Model Context Protocol (MCP) by utilizing a language-agnostic intermediate representation and LLM-guided static analysis to uncover security vulnerabilities stemming from optional clause implementations across diverse SDKs.
This paper addresses the lack of systematic security evaluation for open-source Model Context Protocol (MCP) servers by applying static code analysis to identify Common Weakness Enumeration (CWE) vulnerabilities, mapping them to MITRE CAPEC attack patterns, and introducing a multi-metric risk-assessment framework to guide secure-by-design development.
This paper introduces Adaptive Activation Cancellation (AAC), a real-time, training-free inference framework that mitigates hallucinations in large language models by identifying and suppressing hallucination-associated neural activations as structured interference, thereby improving factual accuracy across multiple model scales without degrading general capabilities or fluency.
Delta-K is a training-free, plug-and-play inference framework that resolves concept omission in multi-instance image generation by extracting and injecting differential keys into the shared cross-attention Key space to establish coherent semantic representations without requiring architectural modifications or additional training.
This research proposes a novel multilingual password strength meter that leverages AI-generated datasets (specifically ChatGPT) and Jaro similarity-based matching to outperform traditional models like PassGAN, demonstrating that incorporating non-English training data significantly enhances detection accuracy for language-specific vulnerabilities, particularly in the Indian context.
This paper establishes that a continuous-time diffusion approximation of policy gradient for stochastic bandits achieves logarithmic regret with a learning rate of , while demonstrating that a significantly smaller learning rate of is necessary to avoid linear regret in specific instances.
This paper proposes a real-time, deformation-aware framework that integrates robotic ultrasound with a lightweight USCORUNet network to dynamically update static intraoperative CBCT scans, thereby enabling continuous soft-tissue monitoring and navigation refinement without repeated radiation exposure.
This paper extends the harmonic loss framework by systematically evaluating various non-Euclidean distance metrics across vision and language models, demonstrating that cosine-based variants offer superior trade-offs in accuracy, interpretability, and sustainability compared to traditional cross-entropy and Euclidean approaches.
This paper proposes a variational quantum sensing framework that utilizes a quantum circuit-optimized probe to learn environmental features from radio-frequency signals, demonstrating through ray-tracing simulations that this approach enables robust localization with no deployment channel measurements and superior sensitivity to weak or obstructed signals compared to classical baselines.
This paper demonstrates that a natively spiking neuromorphic algorithm for solving partial differential equations possesses intrinsic fault tolerance, maintaining accuracy even when up to 32% of neurons and 90% of spikes are dropped, with this robustness being tunable via structural hyperparameters.
This paper introduces DUCTILE, an agentic LLM orchestration framework that separates adaptive decision-making from deterministic tool execution to automate engineering analysis in product development, successfully handling input deviations in an aerospace case study while highlighting the emerging tension between task automation and the creation of exhausting supervisory roles.
This paper proposes a unified cross-view contrastive learning framework that aligns global imaging and local ROI-graph representations to enhance brain disorder classification performance and reveal their complementary discriminative patterns, as validated on the ADHD-200 and ABIDE datasets.
This paper introduces ADMM-PnP with a novel AC-DC denoiser that resolves manifold mismatch and establishes convergence guarantees for score-based generative models within the ADMM framework, thereby improving solution quality across various inverse problems.