Build Loop

Empower Arabic speakers to create and analyze poetry through an AI-driven, instruction-based platform.

Transform GUI automation through Reinforcement Learning to create scalable, intelligent digital agents.

A lightweight, real-time crack classification framework for UAV bridge inspections, balancing accuracy and speed.

D3-Gym provides verifiable environments for data-driven scientific discovery, enabling AI agents to solve real-world research tasks.

NeocorRAG enhances RAG systems by optimizing retrieval quality through evidence chains, achieving state-of-the-art performance with reduced token usage.

Automated diagnostics for transformer models that detect, categorize, and pinpoint root causes of faults, improving repair accuracy for practitioners.

Improve accuracy and efficiency of Text-to-SQL queries for enterprise databases using Template Constrained Decoding.

A unified framework for multi-agent systems that connects Bayesian inference, game theory, and thermodynamics through collective free energy minimization.

MCPHunt is a benchmark framework for evaluating cross-boundary data propagation in multi-server MCP agents, identifying vulnerabilities and testing mitigation strategies.

A comprehensive SHACL validation framework for TMF Intent Ontologies to ensure correctness of network intents before deployment.

A hierarchical alignment transformer for radiology report generation that precisely maps medical images to structured reports.

MED-VRAG enhances medical question answering by integrating multimodal retrieval with visual content from document pages.

KellyBench is a new benchmark and API for evaluating long-horizon sequential decision-making in sports betting markets, revealing significant room for improvement in frontier models.

RuC is a grammar-driven benchmark generator for RTL code completion that enables controlled and scalable evaluation of LLMs in hardware development.

TransVLM is a vision-language model for detecting shot transitions in videos, enhanced with optical flow and deployed to production.

A novel attack that steals secrets from local LLM fine-tuning by exploiting supply-chain model code backdoors, achieving high secret leakage without compromising primary tasks.

A high-fidelity benchmark for translating natural language intents into Ethereum transactions, revealing LLM limitations in real-world Web3 interactions.

Language models and numerical optimizers collaborate to systematically improve mechanical linkage designs, reducing geometric error and diagnosing failure modes.

This paper presents a method for debiasing reward models in LLMs, enhancing alignment with human preferences without performance trade-offs.

A lightweight module accelerates LLM-based recommendation inference by intelligently drafting and verifying item tokens.

A dynamic agentic framework for time series forecasting that refines predictions through iterative planning, action, and reflection, leveraging specialized LLMs and ensemble methods.

An end-to-end governance framework for EHR-embedded AI agents, demonstrating continuous improvement and operational efficiency for clinical AI systems.

A new benchmark for evaluating autonomous GUI agents on complex, multi-application professional workflows, revealing significant performance gaps in current leading models.

This study presents a deep learning approach for non-invasive segmentation of peritoneal cancer regions from CT scans.

SpecVQA provides a benchmark dataset for advancing spectral image understanding and visual question answering with multimodal AI models.

A live benchmark for LLM agents that evaluates performance on evolving real-world workflows with verifiable execution traces.

A self-supervised contrastive framework for learning disentangled representations from tabular remote sensing data.

InteractWeb-Bench is a new benchmark for multimodal agents in website generation, simulating non-expert user interactions and revealing limitations in current agents' ability to avoid blind execution.

DALS is a unified optimizer framework that dynamically scales learning rates per layer and time, achieving superior performance across diverse training regimes.

MetaSymbO is a multi-agent framework for language-guided metamaterial discovery, using symbolic evolution to generate novel and valid microstructures.

Eywa is a framework enabling language models to orchestrate and reason over diverse scientific foundation models for complex, multi-modal tasks.

A self-supervised foundation model trained on millions of brain MRIs that generalizes across diverse clinical tasks with minimal labeled data.

A Python library for extracting Target-Event-Agent networks from text to perform interpretable emotion detection and semantic analysis, comparing human and LLM responses.

IMPRESS framework improves graph few-shot learning by using hyperbolic space and denoising diffusion for better representation and generalization.

A survey and practical guidance for building, evaluating, and integrating LLM systems across the full peer review workflow.

A new benchmark for evaluating fine-grained image-text alignment in interleaved multimodal contexts, crucial for real-world document understanding.

This paper evaluates LLM agents for scientific visualization, offering insights for practical deployment.

MIFair is a unified framework for bias assessment and mitigation based on mutual information, addressing intersectionality and multiclass fairness with a flexible metric and regularization-based training.

A multi-agent framework with retrieval augmentation and debate stages for robust multimodal stance detection.

A multimodal AI that reliably translates circuit diagrams into hardware code, overcoming a critical 'mirage' defect and achieving state-of-the-art performance.

A schema-grounded AI memory system that reliably extracts and validates facts, moving beyond simple text retrieval for agents.

RAY-TOLD is a hybrid control architecture for robots that integrates LiDAR data into latent dynamics and uses MPPI with RL for dense crowd navigation.

Leveraging LLMs to refine graph structures in EEG signals for more accurate and interpretable seizure diagnosis.

WaferSAGE is an LLM-powered framework for wafer defect analysis using synthetic data and reinforcement learning, enabling on-premise deployment of small, specialized models.

An open-source, low-cost, and scalable piezoresistive tactile sensing solution for robotic end-effectors, enabling advanced tactile learning pipelines.

BoostLoRA enhances parameter-efficient fine-tuning by iteratively merging small adapters to grow effective rank, achieving state-of-the-art performance without inference overhead.

PARA is a data-free method to significantly reduce LoRA parameters by 75-90% while preserving performance, integrating seamlessly into existing fine-tuning pipelines.

A framework for video moment retrieval that improves accuracy by learning relationships between answer segments and predicting temporal boundaries.

A unified framework for robust representation learning on heterogeneous graphs with heterophily, addressing noisy connectivity for improved performance.

An open-source toolkit for automated vibrational analysis and structural remediation of crystalline materials using machine-learned potentials, enabling faster discovery of stable polymorphs.

A new metric for LLM evaluation that detects reliable changes in model performance, revealing bidirectional improvements and deteriorations missed by aggregate scores.

A methodology for systematically engineering LLM agents by involving subject matter experts, developers, and helper agents, improving development efficiency and complex-query performance.

Machine Collective Intelligence autonomously discovers governing equations from data, offering explainable and extrapolatable models superior to deep neural networks.

A framework that uses LLMs and Answer Set Programming with self-correction to enable task-agnostic nonmonotonic reasoning.

A production-native framework for evaluating Text-to-SQL accuracy without requiring database schema or reference queries.

An end-to-end LLM framework automating SOC operations for threat detection, query generation, and incident resolution, reducing triage time from hours to minutes.

Develops an activation-level detection system for multi-turn LLM prompt injection attacks, achieving high accuracy across different model families.

An empirical study comparing Google Search, Gemini, and AI Overviews reveals significant differences in information retrieval, source diversity, and robustness, with implications for the generative search ecosystem.

An adaptive aggregation method for federated learning that provides multi-layer defense against Byzantine attacks without requiring server-side data.

A method for discovering and integrating implicit knowledge into deep neural networks for robust image recognition, improving generalization.

A novel method for aligning multimodal reinforcement learning models without introducing significant distributional drift.

Pragmos: A prototype system for collaborative, explainable process modeling using LLMs and specialized tools, co-creating evolving models with human users.

High-signal data filtering and repetition outperform diverse data for sample-efficient German language modeling, achieving state-of-the-art results.

A hybrid engine that uses LLM agents and a protocol-aware DSL with templates to generate UVM testbenches and sequences for IC verification.

Develop controllable generative motion models with integrated physical priors for enhanced animation realism.

TopBench: A new benchmark for evaluating LLMs on implicit prediction and reasoning over tabular data, revealing current model limitations.

This paper explores how historical political institutions can inform the design of multi-agent systems.

A controlled environment and methodology to rigorously assess and improve the out-of-distribution generalization of neural program synthesis models.

TypeBandit: A model-agnostic methodology for heterogeneous attribute completion that addresses type-dependent information asymmetry.

This research challenges the 'Wisdom of the Crowd' in AI agent swarms, demonstrating the 'Inverse-Wisdom Law' where architectural tribalism leads to consensus on erroneous trajectories, and proposes a 'Heterogeneity Mandate' for resilient architectures.

Guidelines for creating adversarial, difficult, and legible benchmark tasks for terminal-agent evaluations to improve LLM capability assessment.

A configurable, grid-aware agent-based model for analyzing electric vehicle charging systems, integrating heterogeneous EV behavior and power allocation.

A dataset mapping LLM mathematical reasoning, anxiety, and confidence across simulated student and AI personas.

Identifies four distinct profiles of AI dependency among Filipino students, highlighting a need for AI literacy integration in higher education.

This research highlights the critical impact of prompt optimization on LLM evaluation, suggesting a need for more dynamic and model-specific assessment frameworks.

A taxonomy and formalization of graph world models for improved environmental representation, prediction, and planning in AI agents.

An all-MLP framework for multivariate time series forecasting that uses iterative refinement and external attention, tuned by Harris Hawks Optimization.

A framework for auditing financial NLP benchmarks to ensure reliable model selection and deployment by addressing measurement risk.

A formal framework for runtime delegation safety in hierarchical multi-agent systems, balancing safety and efficiency.

Evaluating the emotional nuance preservation of small language models in machine translation using a fine-grained emotion dataset.

Mechanized proofs for structural governance in cognitive workflow systems, including coinductive safety predicates and governance invariance.

This research reveals that LLM political bias audits are significantly influenced by sycophancy, where models adapt responses based on inferred user identity rather than fixed ideology.

A semantics-aware runtime that bridges the agent-OS gap for efficient and correct checkpoint/restore in agent sandboxes.

Develops cognitive models to understand and improve human comprehension of AI explanations, reducing the cost of user studies.

A training-free framework for tunnel defect inspection and engineering interpretation using visual recalibration and entity reconstruction.

A formal framework for analyzing structural gaps in AI governance, proposing coterminous governance as a testable criterion.

ObjectGraph creates a seamless bridge between document incorporation and knowledge traversal for efficient agent interaction.

A synthetic corpus and interactive platform for analyzing LLM discourse across diverse human personas and societal topics.

A methodological evolution graph infrastructure for AI scientists and agents to understand research lineage and drive automated discovery.

A domain-specific language and agent system for generating spatially accurate and collision-free 3D indoor scenes from natural language.

ValuePlanner introduces a hierarchical framework for proactive embodied agents to resolve motivational conflicts.

A risk-sensitive bandit controller for LLM coding agents that learns when to use external memory to avoid unsafe injections.

A scalable methodology for creating synthetic computer environments and simulating long-horizon productivity tasks to train agents for improved performance.

A generative multimodal model of human physiology that forecasts trajectories and simulates interventions for personalized medicine.

A self-evolving framework that autonomously discovers, refines, and selects context-specific skills for language models without human supervision.

Investigates factors influencing the non-development or abandonment of AI systems throughout the development lifecycle.

An anchored-curriculum framework where a unified policy alternates between proposing novel specifications and solving them with verified solutions for self-improvement.

A novel rule-generation perspective for estimating LLM compositionality, addressing explainability and data leakage issues.

A lightweight agent architecture with explicit state dynamics generates continuous escalation pressure signals, making LLM agents more clinically legible by revealing accumulating risk prior to escalation.

A leakage-safe survival framework for predicting Alzheimer's disease progression and dynamic tracking using temporal Transformers.

Examines factors influencing pre-service teachers' intention to use AI-enabled educational tools, finding internal motivation to be key.

A new method for procedural tasks using in-context prompting eliminates the need for external agent orchestration frameworks, leading to better performance and fewer failures.

This paper critiques current agentic memory systems and proposes a new understanding of memory in AI.

Agentic AI and digital twins optimize traffic signals in real-time, reducing waiting times and improving traffic flow.

Consumer attitudes in Australia show a preference for AI-generated health summaries based on quality and empathy, despite accuracy concerns.

A unified framework for generative structure search that combines diffusion models with physical forces to accelerate the discovery of diverse molecular and crystal structures.

A pipeline for on-demand persona generation in agentic platforms allows for dynamic crafting of AI personas to match user needs and task contexts, improving efficiency and appropriateness.

A novel diffusion model for generating continuous-time, continuous-space stochastic processes conditioned on partial observations, improving dynamics and enabling arbitrary subset conditioning.

Investigating how visual priming affects the cooperative behavior of Vision-Language Models in the Iterated Prisoner's Dilemma.

Demonstrates that in-context examples can suppress scientific knowledge recall in LLMs, shifting computation towards empirical pattern fitting.

A protocol for verifying and deploying LLM-generated reward hypotheses in reinforcement learning based on policy competence and training phase.

Reviews fairness notions and metrics for distribution network planning and operations, analyzing their mathematical complexity and impact on resource allocation.

Identifies vulnerabilities in cross-modal encoders by detecting hub texts that can lead to misleading similarity scores in applications like image captioning and retrieval.

A neuro-symbolic framework for synthesizing, verifying, and evaluating causal rules grounded in legal and safety principles for safety-critical domains.

Characterizing the inconsistent persona of emergent misalignment in LLMs, revealing distinct patterns of coherent and inverted misalignment.

This paper proposes an energy-geography framework for geo-distributed AI inference, optimizing placement based on electricity prices, carbon intensity, and latency constraints.

A proposed framework for governing LLM updates focuses on production contracts, risk-category-based testing, and compatibility gates to manage behavioral drift in hosted LLM services.

Investigating LLMs and diffusion models for procedural content generation of trading cards, focusing on personalized designs and player engagement.

This paper audits vision-language models for medical visual question answering, highlighting significant trustworthiness issues.

This paper explores design challenges and emerging methodologies for ensuring dependability in autonomous and embedded systems integrating AI/ML components.

This paper introduces a protocol and empirical findings for evaluating how AI reading assistants manage interpretive work, identifying weaknesses where systems redistribute too much meaning-making labor away from the user.

The paper introduces a VLA model that reformulates pretraining through goal-conditioned reinforcement learning for robotic control.

AgentEconomist translates economic intuitions into executable computational experiments through a modular interactive system.

The paper introduces a framework for adaptive inference control in black-box LLM services to improve response reliability.

This paper presents a theoretical framework for clinician overrides in clinical AI without practical implementation.

This paper explores the conceptual foundations and future directions of agentic reinforcement learning within large language models, focusing on autonomous agents capable of complex reasoning and planning.

A framework for analyzing information contamination in multi-agent systems by injecting perturbations and measuring trace divergence.

This paper explores LLM-based modularization for engineering design but lacks practical deployment signals.

This framework constructs knowledge graphs from AI policy documents to enhance compliance reasoning.

A layered review of security risks and defense strategies for autonomous agent frameworks, using OpenClaw as a case study.

This paper proposes an architectural pattern language for visual agents but lacks practical application signals.

Sampler-Robust Optimization (SRO) optimizes decisions against the worst-case sampler induced by perturbing learned generative models.

This paper proposes a conceptual framework for enhancing human-AI information seeking through knowledge affordances.

A tensor learning framework for constructing statistical channel fingerprints in massive MIMO systems.

Novel splitting techniques for argumentation formalisms that incorporate collective attacks and supports, generalizing existing frameworks.

A theoretical framework for improving the computational efficiency of assumption-based argumentation by splitting the knowledge base.

This paper introduces a new equilibrium concept in game theory that guarantees existence by minimizing coalitional deviation incentives, with algorithms for average and maximum gain objectives.

A conceptual framework to help leaders understand and manage the evolving spectrum of human and AI roles in decision-making teams.

Develops a theoretical framework to understand how sparse autoencoders capture concept manifolds, revealing suboptimal recovery and motivating new interpretability methods.

A framework for theoretically grounded interpretation of AI performances to counter behavioristic tendencies in AI research.

Examines how network architecture, task similarity, and representational dimensionality jointly shape learning in continual learning systems.

A framework for mapping classroom interaction research dimensions (scale, duration, modality) to guide research and AI tool design.

A theoretical framework recasts Information Bottleneck as a rate-distortion problem, offering principled distributional regularizers for learning.

Critiques AI sign language translation tools for their ableist nature, arguing they standardize language for profit and fail to capture the human experience of deaf individuals.

Developing a gradient descent-based, physics-constrained attractor Fuzzy Cognitive Map with residual memory and backpropagation through time.