ClawKeeper adds safety mechanisms to OpenClaw autonomous agent runtime, addressing vulnerabilities in tool integration and command execution.
OneSearch-V2 improves generative retrieval for search systems with latent reasoning and self-distillation. Industrial-scale framework.
Large-scale annotated video demonstration dataset for computer-use agents enabling automation of complex desktop workflows with continuous video sequences.
Integration of causal machine learning into clinical decision support systems with clinician-facing interfaces for interpretable treatment-specific reasoning.
Autoresearch pipeline using Claude Code LLM agent to autonomously discover novel white-box adversarial attack algorithms outperforming 30+ existing methods.
Multi-dimensional evaluation framework for uncertainty attribution methods in explainable AI with aligned proxy tasks and metrics.
Mobile GUI agent using rejection fine-tuning to learn from failed trajectories and improve credit assignment for long-horizon tasks.
Video-language foundation model pretraining on surgical procedure videos for zero-shot event recognition in intraoperative settings.
Framework combining diffusion-based world models with selective enhancement for temporally coherent augmented reality applications.
Empirical study comparing chunking strategies for RAG systems in oil and gas documents, evaluating fixed-size, recursive, semantic, and structure-aware approaches.
Agentic video understanding framework using Vision-Language Models with active planning to seek evidence from raw video during reasoning.
Free-Market Algorithm metaheuristic using distributed supply-and-demand dynamics for open-ended optimization with emergent fitness.
Adversarial attack methods to protect images from malicious diffusion-based image-to-video generation models.
Vision-Language Models for converting rasterized figures into editable SVG vector graphics automatically.
Study of RAG systems applied to AI policy analysis using AGORA corpus, examining reliability challenges in dense legal language domains.
Vision-Language Models for supporting human decision-making in high-stakes domains like medical diagnosis through collaborative human-AI systems.
Study of AI agents powered by LLMs in multi-echelon supply chain simulation investigating emergent strategic behavior and dynamics like the bullwhip effect.
Graph-based evaluation framework for domain-specific LLM benchmarking using clinical guidelines transformed into queryable knowledge graphs with dynamic query instantiation.
GeoSketch: Neural-symbolic approach for geometric reasoning in MLLMs using auxiliary line construction and affine transformations for problem solving.
SAG-Agent: LLM-based agent using dynamic knowledge graphs for long-horizon reasoning in strategy games via GUI interaction without APIs.
CastMind: Agentic reasoning framework for time series forecasting using iterative refinement with temporal features, domain knowledge, and case-based references.
Generative Adversarial Reasoner: Framework using adversarial RL to improve LLM reasoning capabilities and reduce mathematical errors through co-evolved reasoner-discriminator training.
Research on enabling ultra-long-horizon autonomous agents with cognitive accumulation for multi-week ML engineering experiments.
Evaluates LLM performance on perspective-taking and knowledge state estimation tasks comparing cognitive abilities to chimpanzees.
CollectiveKV framework reduces inference latency in Transformer-based sequential recommendation systems through KV cache optimization.
CIRCLE framework for evaluating AI systems across six lifecycle stages, bridging gap between benchmarks and real-world deployment outcomes.
Framework for evaluating logical reasoning agents with agentified assessment, standardized interfaces, and structured failure tracking.
TikZilla: Dataset and reinforcement learning approach for scaling text-to-TikZ scientific figure generation from high-quality training data.
GPT4o-Receipt: Benchmark of 1,235 receipt images comparing AI-generated vs authentic documents evaluated by LLMs and humans.
Framework for relationship-aware safety unlearning in multimodal LLMs addressing relational safety failures without collateral damage.
DomAgent: Framework combining knowledge graphs and case-based reasoning with LLMs for domain-specific code generation tasks.
Counterfactual learning approach for CVR estimation in recommender systems addressing data sparsity and sample selection bias.
Survey on enterprise financial risk analysis using big data and LLM technologies for financial prediction and management.
Dynamic Neural Potential Field: Learning-enhanced MPC framework coupling Transformer-based predictor with classical optimization for robot obstacle avoidance.
Evaluation framework for large language models addressing randomization in coupled token generation with causal modeling approach.
Unicorn: Multi-agent reinforcement learning approach for adaptive traffic signal control in heterogeneous urban networks.
VidhikDastaavej: Model-agnostic wrapper for automated legal document generation in India using LLMs, with anonymized private document dataset.
arXiv paper on VidhikDastaavej, model-agnostic wrapper for structured legal document generation in India with large-scale anonymized dataset.
arXiv paper comparing linguistic features of AI-generated vs human responses to mental health queries. Application of LLMs in healthcare.
arXiv paper introducing Distance Explainer for generating post-hoc explanations of embedded vector spaces using saliency-based techniques.
arXiv paper on Bottlenecked Transformers using periodic KV cache consolidation for improved reasoning with auxiliary latent-space computation.
arXiv paper on RestoreVAR using visual autoregressive modeling for fast all-in-one image restoration, replacing slower diffusion approaches.
arXiv paper demonstrating in-context reinforcement learning (ICRL) emerges during LLM inference. Proposes ICRL prompting for inference-time self-improvement.
arXiv paper exploring persona prompts as jailbreak attack vector against LLMs. Security analysis of prompt injection vulnerabilities.
arXiv paper on SafeSieve, progressive pruning algorithm for LLM-based multi-agent systems to reduce token overhead and communication redundancy.
arXiv paper introducing TimeAlign, representation learning method using contrastive alignment for time series forecasting.
arXiv paper on PromptLoop for iterative prompt refinement in diffusion models via latent RL feedback. Improves generalization and robustness.
Proposes declarative OS interfaces to improve computer-use agents' ability to interact with GUIs, replacing error-prone imperative action sequences.
SyTTA: Label-free test-time adaptation for LLMs using only 4 extra tokens, enabling domain-specific deployment without fine-tuning on expensive labeled data.
Network traffic characterization study analyzing ChatGPT, Copilot, and Gemini usage patterns and their impact on internet infrastructure.