Build Loop

Automate online system operation tasks with Bian Que's flexible agentic framework.

A framework for efficient edge-cloud visual data processing using progressive semantic communication for VLMs.

A framework for efficiently distilling high-performance diffusion language models into smaller, faster models.

OMEGA automates ML algorithm generation and optimization using LLMs to outperform common benchmarks.

DepthPilot is an interpretable framework for generating realistic and clinically aligned colonoscopy videos, enabling better surgical navigation and diagnosis.

SecMate is a multi-agent virtual customer assistant that enhances cybersecurity troubleshooting through device, user, and service specificity.

LATTICE benchmarks the decision support utility of crypto agents in user-facing scenarios.

DSIPA is a training-free framework for detecting LLM-generated content using sentiment distribution analysis.

A domain-adaptive pipeline for improving multilingual crisis communication through fine-tuning language models.

Domain-adapted small language models improve clinical triage accuracy and efficiency.

An end-to-end pipeline for synthetic industrial defect generation and detection to overcome data scarcity.

A transformer-based framework for offline safe reinforcement learning that uses Lyapunov-guided imagination for test-time adaptation without retraining.

XDFT is a self-evolving agent that automates diagnosis of DFT-experiment band-gap mismatches.

CheXthought provides a detailed multimodal dataset for enhancing AI-driven chest X-ray interpretation and clinical decision-making.

An atomic-quality probe and hybrid selector system for managing skill updates in compositional robot policies, improving reliability and reducing costs.

Detecting LLM alignment faking by analyzing tool selection, with a new dataset and evaluation of frontier models.

ClawGym is a scalable framework for developing environment-grounded Claw-style agents with a specific focus on task synthesis and evaluation.

SeeCo is a plug-and-play framework that recalibrates open-vocabulary models on-the-fly for improved semantic segmentation in remote sensing images.

A fine-tuning framework that precisely targets and corrects language confusion in LLMs at the token level, improving multilingual consistency without sacrificing general capabilities.

A unified 4D world model for robots that synthesizes high-fidelity video and 3D reconstructions while enabling real-time action execution.

HalluCiteChecker is a lightweight, offline toolkit for detecting and verifying hallucinated citations in scientific papers, reducing reviewer workload.

A training-free framework that dynamically routes large reasoning models to different scaling strategies based on output disagreement, improving accuracy and reducing cost.

FutureWorld is a live environment for training predictive agents using real-world outcome rewards.

MedSynapse-V bridges visual perception and clinical intuition for enhanced medical diagnosis.

StratMem-Bench evaluates strategic memory use in virtual character conversations for enhanced realism.

A Human-in-the-Loop framework for automating secondary-level mathematics assessment using multiple LLMs.

A tree-structured prompting framework for LLMs to generate sports game reports from tables, improving comprehension and efficiency.

Naamah is a large-scale, high-quality synthetic Sanskrit NER dataset created using DBpedia seeding and LLM generation, addressing the scarcity of annotated resources for classical Sanskrit literature.

An annotation tool for long-horizon robotic action segmentation that synchronizes multi-modal data and streamlines the annotation process.

A novel optimization framework for diffusion models that enhances perceptual quality without sacrificing generative fidelity.

STLGT is a scalable, trace-based linear graph transformer for accurate and efficient tail-latency prediction in microservice systems, enabling proactive SLO management.

QYOLO is a lightweight object detection framework that uses quantum-inspired channel mixing to achieve significant parameter and GFLOPs reduction with minimal accuracy loss.

A retrieval framework that adaptively injects evidence during multi-step reasoning for large language models, improving accuracy and efficiency.

A multi-modal transformer architecture for discrete celestial orientation, achieving high accuracy and efficiency for autonomous spacecraft navigation.

A framework for multi-agent policy simulation that uses architectural heterogeneity and coherence validation to prevent artificial consensus and improve diverse decision-making.

An uncertainty-aware reward framework for reinforcement learning that mitigates reward hacking and improves alignment.

A framework to enhance refusal capabilities of Vision-Language Models, improving their trustworthiness in specialized domains.

DreamProver is an agentic framework that evolves reusable lemma libraries for formal theorem proving.

A dynamic multimodal recommendation framework that adapts to evolving user interests over time.

An agentic system for evaluating the AI-readiness of scientific data across governance, quality, compatibility, and adaptability.

A novel decoding algorithm that accelerates LLM-based attribute value extraction by up to 13.8X through hyper-parallel processing, significantly reducing inference costs.

A pipeline for converting imperative programs into typed, attributed graphs using neural and structural methods for verification artefact reuse.

Reveals a stable, content-invariant distributional attractor in LLM response positions under sandbagging, offering a black-box signature for this behavior.

A framework for engineering measurable trust in clinical AI through evidence, supervision, and staged autonomy, moving beyond black-box confidence.

A toolkit to detect spurious correlations in speech datasets, preventing overestimation of performance in critical applications.

Efficient methods for utilizing text-only data to improve encoder-dominated speech recognition models, showing larger encoders with smaller decoders perform comparably.

An iterative logic tensor network that empirically demonstrates reasoning is a distinct capability from symbol grounding for compositional generalization in neural networks.

A framework for enhancing creative writing quality through calibrated surprise using information theory.

This study analyzes learned helplessness in math tutoring using the Apriori algorithm to identify behavioral patterns.

A training-free framework for open-vocabulary change detection that uses cross-temporal memory reasoning and global-local adaptive rectification to identify semantic changes in bi-temporal images.

This research frames language diffusion models as associative memories, identifying a sharp transition between memorization and generalization using conditional entropy.

A theoretical framework for automated relational reasoning integrated with ANNs, achieving high IQ test scores.

A training-free framework for small object detection that uses spatial attention entropy to adaptively focus computation on high-saliency, high-uncertainty regions, improving performance with marginal latency overhead.

A training-free method for discovering minimal neural network architectures by progressively reducing random topologies, outperforming pruning baselines with significant parameter reduction and faster execution.

A framework for efficient LLM inference on edge devices by intelligently offloading KV caches to NVMe storage, reducing latency and improving SSD utilization.

Benchmarking the safety of LLMs for robotic health attendant control using a dataset of harmful instructions, revealing significant safety concerns.

A quantum-based image steganography framework that ensures secure payload recovery through multi-factor context binding.

DIRECT introduces a novel mechanism for causal learning in neural assemblies, enhancing the understanding of causal influence.

Improving text style transfer in graphic designs by developing new methods for word alignment in machine translation.

A neuro-symbolic framework for interactive agents that improves compositional generalization by grounding actions with a causal program graph and inductive logic programming.

A system for political question evasion detection using a multi-head RoBERTa with chunking for long contexts.

MetaSR enhances generative super-resolution by dynamically utilizing content-adaptive metadata.

An AI-native TDD framework that operationalizes classical TDD principles for reliable LLM-assisted development.

A behavioral firewall for structured-workflow AI agents to enhance security against tool-call anomalies.

This research explores culturally specific Generative AI risks for youth in non-Western contexts, providing design implications for inclusive parental controls.

This paper proposes a hierarchical decision-making framework for UAV search-and-rescue missions combining rule-based coaching with online reinforcement learning.

A multi-stage framework for bi-atrial segmentation from 3D MRI using V-Net models.

The ST-PT framework explores the mathematical equivalence of Transformers and probabilistic models for time series.

This paper explores how generative AI subtly influences control and agency in recruiting workflows, leading to deskilling despite perceived efficiency gains.

A benchmarking framework for evaluating complex multimodal document processing pipelines in enterprise AI.

This paper explores how Large Language Models respond to legal arguments, investigating their persuadability and implications for legal decision-making.

Exploring foundational physics with implications for AI and information processing.

Qvine introduces a vine-structured ansatz for quantum circuits to efficiently load high-dimensional distributions, addressing challenges in quantum machine learning and finance.

This paper reviews advancements in mm-wave and sub-THz oscillators for future communication technologies.