Build Loop

A new benchmark for testing multimodal AI models on complex multi-hop reasoning across texts and visuals.

Faithfulness-QA enhances retrieval-augmented language models by providing a large dataset for context-faithful output training.

Automate data-driven video story generation with a multi-agent framework and a new benchmark for evaluation.

Develop an AI platform for strategic coordination and multiplayer game scenarios to enhance cooperation in competitive environments.

AI-driven short video recommendation system that captures user preferences to enhance engagement.

A plug-and-play method to reduce hallucinations in vision-language models by intervening during the prefill stage, improving initial representations.

A zero-trust semantic gateway for secure, validated autonomous agent interaction in enterprise systems.

TSN-Affinity is a continual offline reinforcement learning method that enables task-specific parameterization and knowledge sharing through similarity-guided architectural reuse.

A health-system-scale semantic search system for clinical notes that reduces chart abstraction time by up to 89% at a low operational cost.

QFlash enables end-to-end integer-only FlashAttention for Vision Transformers, significantly speeding up computation and reducing energy consumption.

AutoResearchBench revolutionizes AI-driven scientific literature discovery with a challenging new benchmark.

Achieve topology-faithful dimensionality reduction at scale, outperforming UMAP in preserving global structure.

An agentic AI framework that uses LLMs and simulation to explore and optimize computer architecture designs, outperforming state-of-the-art.

Physics-informed test-time adaptation for robust human activity recognition on mobile devices, enabling on-device personalization without centralizing private data.

A recursive multi-agent framework that enhances reasoning and efficiency in complex tasks by enabling agent collaboration through a unified latent-space computation.

AHASD is a novel asynchronous heterogeneous architecture for LLM speculative decoding on mobile devices, significantly improving throughput and energy efficiency.

LLM-ReSum is a self-reflective summarization framework that improves factual accuracy by up to 33% and coverage by 39% through an LLM-based feedback loop.

Open-source, highly sparse multilingual Mixture-of-Experts language models with efficient upcycling and best-in-class performance-to-compute ratio.

A benchmark and metric for evaluating LLM-generated REST API tests from natural language requirements, improving functional validation.

OxyGent is an open-source framework for building modular, observable, and evolvable multi-agent systems using a Lego-like abstraction.

A multi-agent framework that decomposes instructed code editing into specialized roles to improve reliability and reduce unintended changes.

A multi-source benchmark and evaluation pipeline for assessing structured output quality in large language models across text, image, and audio inputs.

A fast, zero-shot detector for machine-generated text that leverages perplexity shifts under text shuffling, outperforming existing methods.

Generate synthetic training data for custom LLM guardrails using debate and dimension decomposition, eliminating the need for human annotation.

Interpretability-Guided Data Selection (IGDS) framework that leverages internal LLM task features to select highly effective data for fine-tuning, improving efficiency and performance.

A Korean legal-domain LLM trained with a use-case-driven framework and rigorous data curation for practical legal tasks.

PSI-Bench is an interpretable evaluation framework for depression patient simulators, providing clinically grounded diagnostics to improve realism and guide future development.

A multimodal framework that leverages code and comments to significantly improve software vulnerability detection accuracy.

DRAGON is a new benchmark and dataset for evaluating evidence-grounded visual reasoning in diagrams, enabling more reliable assessment of VLM capabilities in diagram question answering.

SymphonyGen is a 3D hierarchical framework for generating cinematic orchestral music with controllable harmony skeletons and improved musicality.

A benchmark and fine-tuned LLM for automatically evaluating K-12 science instructional materials.

A dataset of 10,217 confirmed GPT-image-2 generated images from Twitter, including analysis of content and a finding that C2PA credentials are stripped.

An LLM agent framework that automates end-to-end 3D cutscene generation by integrating with game engines and orchestrating specialist sub-agents.

SnapGuard: A lightweight and fast method for detecting prompt injection attacks in screenshot-based web agents by analyzing visual and textual signals.

A label-free reinforcement learning framework for LLMs that decouples answer proposal from reward verification, improving mathematical reasoning and code generation.

A framework for synthesizing diverse terminal tasks using skill graphs to improve agent training for command-line execution.

A decoupled reinforcement learning framework for reasoning-driven image editing that optimizes planning independently from the generative model.

GraphPL, a graph neural network approach for robust modality imputation in patchwork learning, enabling downstream tasks like disease prediction.

A neurocognitive governance framework embeds self-governance into autonomous AI agents, achieving 95% compliance accuracy in a retail supply chain workflow.

A language-agnostic guardrail for LLMs that detects cross-lingual jailbreaks using semantic similarity to an English prompt codebook, without retraining.

R3-SQL is a Text-to-SQL framework that improves query ranking consistency and recall through unified reward and agentic resampling, achieving state-of-the-art results.

An agentic workflow using LLMs and structured queries to automate the initial stages of security alert investigation, improving accuracy and reducing manual workload.

An ultra-low-power FPGA-based CNN for on-device cardiac feature extraction, enabling autonomous health monitoring for astronauts.

A framework for continual brain lesion segmentation that integrates visual features with structured concepts to simulate clinical reasoning and guide model growth.

This research demonstrates that unstructured pruning can augment test-time reasoning performance in LLMs, challenging the notion that pruning always degrades capabilities.

A framework that decouples retrieval and reranking for more accurate multi-modal knowledge graph completion.

A speaker-adaptive network for emotion recognition in conversations that calibrates features, gates modalities, and regularizes speaker identity for improved accuracy.

An empirical study of uncertainty estimation for audio-aware LLMs, revealing that semantic-level methods are better for reasoning but model-dependent for trustworthiness.

Sample-efficient neuro-symbolic PPO that uses logical policy specifications to accelerate learning in sparse-reward environments.

CORAL is an adaptive retrieval methodology for multilingual RAG that iteratively refines retrieval space and query for culturally-aligned answers, improving accuracy by up to 3.58%p.

A new evaluation metric for watermark removal that focuses on forensic detectability, revealing artifacts missed by current methods.

SIEVES improves multimodal LLM coverage on out-of-distribution tasks by learning to score the quality of visual evidence for selective prediction.

A real-time adaptive traffic signal system that uses computer vision to extend crossing times for vulnerable road users, improving safety and reducing delays.

Optimized communication-efficient federated learning solution to fine-tune language models on edge devices.

A benchmark showing that providing explicit business semantics to LLMs significantly improves accuracy and reduces hallucinations in data analytics queries.

A production-ready inference architecture for scalable, cost-effective deployment of complex, multi-model AI systems.

An architecture for long-horizon knowledge synthesis that orchestrates LLM agents with explicit state tracking and robust error handling for complex tasks.

Generative UI can create adaptive interfaces for user-generated content platforms, bridging accessibility gaps for visually impaired and older users.

Develop a platform for optimizing complex models using memory-enhanced decentralized debate among AI agents.

Evaluating 15 frontier AI chatbots on psychiatric triage, this study reveals high accuracy for emergencies but significant over-triage for lower-risk presentations.

A reinforcement learning framework that learns expert-level chip placement by inferring implicit rewards from expert layouts, improving wirelength optimization.

A multi-task EEG analysis framework using LoRA to enable simultaneous adaptation of pre-trained models to multiple downstream tasks, improving efficiency and performance.

A small transformer model trained to execute programs in MicroPy, demonstrating the potential for transformers to act as universal computers.

TrialCalibre is a multi-agent system designed to automate and scale the BenchExCal framework for causal effect estimation in real-world evidence studies.

This research evaluates existing code evaluation metrics for their effectiveness in detecting source code plagiarism, suggesting that some metrics can be competitive with dedicated tools.

This paper categorizes imperfect rewards in reinforcement learning, showing that some errors can be beneficial for policy gradient optimization.

A benchmark for evaluating bandgap prediction models in semiconductors under experimental conditions, highlighting generalization limitations.

VAE-Inf is a novel framework that combines variational autoencoders with hypothesis testing for statistically interpretable imbalanced classification, offering robust error control.

A cooperative path planning system for ground and aerial vehicles to navigate uncertain road networks, demonstrated on urban environments.

A novel metric and toolset to identify and prioritize data errors that most impact machine learning model performance, enabling targeted data cleaning.

The first large-scale Dutch medical language corpus, comprising 35 billion tokens, is now available on Hugging Face for NLP development.

A framework that uses structured specifications and optimization to design networked systems, outperforming LLMs in constraint satisfaction and explainability.

Introduces HotComment, a multimodal benchmark and StyleCmt model for evaluating online comment popularity by considering content quality, trends, and user behavior.

Investigating linguistic biases in LLM recommendations across different English dialects and Hindi-English code-switching.

LLMs can dynamically adjust work environments to enhance focus, well-being, and engagement in hybrid settings, while raising ethical considerations.

A recursive, sparse mixture-of-experts framework integrated into diffusion models to enhance structured reasoning and text following in image generation.

A genetic programming algorithm for multi-task reinforcement learning in continuous control environments with interpretable decision flows.

SkillGuard-Robust is a system for auditing untrusted agent skills, improving security review by combining evidence extraction, semantic verification, and consistency adjudication.

A graph-guided loss function that improves fine-tuning of language models by incorporating global semantic structure for more discriminative embeddings.

Improves zero-shot offline reinforcement learning by extracting task vectors directly from the offline dataset for more principled task sampling.

Evaluating LLMs' understanding of embodied cognition and cultural variation using cross-linguistic demonstratives, revealing English-centric biases.

This paper compares traditional and LLM-based methods for estimating recipe nutrients, highlighting a trade-off between accuracy and computational cost.

A hybrid framework combines machine learning with Answer Set Programming for context-aware phishing website classification, allowing for efficient knowledge updates.

Analyzes weak-to-strong AI alignment failures using a bias-variance perspective to identify risks and improve model robustness.

A novel loss function continuum that mitigates cold-start stalling in reasoning models by dynamically adjusting supervision commitment.

An AI-driven framework for offloading tasks in mobile edge computing networks to optimize energy and computing resources.

Canonical knowledge distillation significantly improves semantic segmentation performance, achieving state-of-the-art results with smaller models.

An empirical study investigating the scaling behavior and geometric conditions for in-context learning in Transformers.

A robust reinforcement learning approach for city-scale EV ride-hailing that guarantees feasibility of dispatch, repositioning, and charging decisions.

Demonstrates how large language models can be manipulated to spread misinformation by prioritizing fringe scientific material, posing risks to public understanding.

A dependency-driven prompt pipeline for generating coherent and structurally consistent RPG content using LLMs.

A Q-learning framework for reinforcement learning that prevents unsafe state visitation during training by leveraging a behavior policy supported on a safe set.

A threat-oriented digital twinning methodology for evaluating the cybersecurity of autonomous platforms, adaptable for UAV and space systems.

ValueAlpha is a stress-testing protocol for LLM-judged investment rationales, ensuring claims are stable and agreed upon before returns are observable.

A faceted model for transparent attribution of AI-assisted text production, detailing the form, generation, and evaluation of AI interventions.

A medoid prototype alignment framework for detecting unknown attacks in industrial control systems across different plants.

An automated adversarial collaboration framework uses LLMs and program synthesis to adjudicate between competing cognitive science theories.

Explains how overparameterization in neural networks introduces symmetries that improve optimization and make global minima more reachable.

A new method for attributing failures in vision-language navigation agents to specific capabilities, improving debugging and agent development.

An empirical study of how university students integrate AI into diverse academic writing workflows, identifying three distinct usage configurations.

A theoretical framework for bandit problems with smooth payoffs on graphs, applicable to content-based recommendation.

A framework for evaluating factual correctness in procedural video captioning, separating conceptual and contextual facts.

A novel method for unsupervised domain adaptation in information retrieval that improves document sampling by considering model uncertainty.

A transformer-based meta-agent that learns to model and exploit opponents in imperfect-information games.

GPT-based tools show promise for assisting in spreadsheet model building but remain unreliable for professional use due to inconsistencies and workflow challenges.

This paper evaluates agentic AI failures in astrophysical workflows, highlighting silent incorrect computations as a critical risk and releasing an evaluation framework.

This research investigates positional bias in small LLMs as a failure mode for detecting deliberate underperformance, suggesting positional distribution shifts as a more effective signature than below-chance accuracy.

Reveals a hidden coupling between gradient directions and linear centroids in neural networks, which is obscured by standard optimizers like AdamW.

A functional geometric algebra framework that offers a mathematically superior foundation for natural language semantics, enhancing compositionality and interpretability.

Propose a TEE-based architecture for auditable AI-assisted grant evaluation without exposing the model.

This research identifies conditional misalignment in language models, where common interventions can hide emergent misbehavior that reappears in specific contexts.

This paper proposes a framework for accountable knowledge distillation, focusing on preserving teacher model capabilities beyond simple task scores to ensure reliability in deployed systems.

This paper theoretically distinguishes three models of human feedback for large language models to improve annotation strategies.

A framework for learning language model policies that manage risk through explicit intervention actions, focusing on epistemic conduct.

Develops algorithms for optimal audit policies in principal-agent games to mitigate fraud in resource allocation.

An effective field theory approach to derive quasiparticle band narrowing from frozen core dynamics, resolving discrepancies in electronic band structure calculations.

A comprehensive framework for inferring learner cognitive and affective states from observable nonverbal cues, based on a systematic review of existing research.

Investigates the persistence of gradient alignment in multi-step settings for subliminal learning, with implications for mitigation methods.

This paper explores value-sensitive AI system designs for prayer, emphasizing the preservation of user agency and authenticity in spiritual contexts.

Theoretical analysis of convergence and halting conditions in Recurrent Graph Neural Networks.

Theoretical introduction to the verification of neural networks, covering various architectures and verification techniques.

This paper argues for a co-design agenda between AI tools and MBSE models to enable AI participation beyond simple prompt-based interaction.