arXiv subjects

Cheng Cao

Publications and source records attributed to Cheng Cao.

13 recordsLinked to original sources

arXiv2026

Omni123: Exploring 3D Native Foundation Models with Limited 3D Data by Unifying Text to 2D and 3D Generation

Recent multimodal large language models have achieved strong performance in unified text and image understanding and generation, yet extending such native capability to 3D remains challenging due to limited data. Compared to abundant 2D imagery, high-quality 3D assets are scarce, making 3D synthesis under-constrained. Existing methods often rely on indirect pipelines that edit in 2D and lift results into 3D via optimization, sacrificing geometric consistency. We present Omni123, a 3D-native foundation model that unifies text-to-2D and text-to-3D generation within a single autoregressive framework. Our key insight is that cross-modal consistency between images and 3D can serve as an implicit structural constraint. By representing text, images, and 3D as discrete tokens in a shared sequence space, the model leverages abundant 2D data as a geometric prior to improve 3D representations. We introduce an interleaved X-to-X training paradigm that coordinates diverse cross-modal tasks over heterogeneous paired datasets without requiring fully aligned text-image-3D triplets. By traversing semantic-visual-geometric cycles (e.g., text to image to 3D to image) within autoregressive sequences, the model jointly enforces semantic alignment, appearance fidelity, and multi-view geometric consistency. Experiments show that Omni123 significantly improves text-guided 3D generation and editing, demonstrating a scalable path toward multimodal 3D world models.

cs.CV↗

arXiv2026

EvoEGF-Mol: Evolving Exponential Geodesic Flow for Structure-based Drug Design

Structure-Based Drug Design (SBDD) aims to discover bioactive ligands. Conventional approaches construct probability paths separately in Euclidean and probabilistic spaces for continuous atomic coordinates and discrete chemical categories, leading to a mismatch with the underlying statistical manifolds. We address this issue by representing molecules using composite exponential-family distributions, where coordinates and categories are represented within a unified natural parameter space to evolve synchronously along exponential geodesics under the Fisher-Rao metric. To avoid the instantaneous trajectory collapse induced by geodesics directly targeting Dirac distributions, we propose Evolving Exponential Geodesic Flow for SBDD (EvoEGF-Mol), which replaces static Dirac targets with dynamically concentrating distributions and is trained with a progressive-parameter-refinement architecture. Our model approaches a reference-level PoseBusters passing rate (93.4%) on CrossDock, demonstrating remarkable geometric precision and interaction fidelity, while achieving superior performance over baseline methods on real-world MolGenBench tasks for bioactive scaffold recovery. Code is available at https://github.com/BLEACH366/EvoEGF-Mol.

cs.LG↗

arXiv2026

Dynamic Role Assignment for Multi-Agent Debate

Multi-agent large language model (LLM) and vision-language model (VLM) debate systems employ specialized roles for complex problem-solving, yet model specializations are not leveraged to decide which model should fill which role. We propose dynamic role assignment, a framework that runs a Meta-Debate to select suitable agents before the actual debate. The meta-debate has two stages: (1) proposal, where candidates provide role-tailored arguments, and (2) peer review, where proposals are scored with data and role-specific criteria to choose the best agent for each position. We evaluate our method on LLM problem solving benchmarks. Applied on top of existing debate systems, our approach consistently outperforms uniform assignments (filling all roles with the same model) by up to 74.8% and random assignments (assigning models to roles without considering their suitability) by up to 29.7%, depending on the task and the specific assignment. This work establishes a new paradigm for multi-agent system design, shifting from static agent deployment to dynamic and capability-aware selection.

cs.CL↗

arXiv2026

Fact-Checking with Large Language Models via Probabilistic Certainty and Consistency

Large language models (LLMs) are increasingly used in applications requiring factual accuracy, yet their outputs often contain hallucinated responses. While fact-checking can mitigate these errors, existing methods typically retrieve external evidence indiscriminately, overlooking the model's internal knowledge and potentially introducing irrelevant noise. Moreover, current systems lack targeted mechanisms to resolve specific uncertainties in the model's reasoning. Inspired by how humans fact-check, we argue that LLMs should adaptively decide whether to rely on internal knowledge or initiate retrieval based on their confidence in a given claim. We introduce Probabilistic Certainty and Consistency (PCC), a framework that estimates factual confidence by jointly modeling an LLM's probabilistic certainty and reasoning consistency. These confidence signals enable an adaptive verification strategy: the model answers directly when confident, triggers targeted retrieval when uncertain or inconsistent, and escalates to deep search when ambiguity is high. Our confidence-guided routing mechanism ensures that retrieval is invoked only when necessary, improving both efficiency and reliability. Extensive experiments across three challenging benchmarks show that PCC achieves better uncertainty quantification than verbalized confidence and consistently outperforms strong LLM-based fact-checking baselines. Furthermore, we demonstrate that PCC generalizes well across various LLMs.

cs.CL↗

arXiv2024

An Efficient Two-Dimensional Functional Mixed-Effect Model Framework for Repeatedly Measured Functional Data

With the rapid development of wearable device technologies, accelerometers can record minute-by-minute physical activity for consecutive days, which provides important insight into a dynamic association between the intensity of physical activity and mental health outcomes for large-scale population studies. Using Shanghai school adolescent cohort we estimate the effect of health assessment results on physical activity profiles recorded by accelerometers throughout a week, which is recognized as repeatedly measured functional data. To achieve this goal, we propose an innovative two-dimensional functional mixed-effect model (2dFMM) for the specialized data, which smoothly varies over longitudinal day observations with covariate-dependent mean and covariance functions. The modeling framework characterizes the longitudinal and functional structures while incorporating two-dimensional fixed effects for covariates of interest. We also develop a fast three-stage estimation procedure to provide accurate fixed-effect inference for model interpretability and improve computational efficiency when encountering large datasets. We find strong evidence of intraday and interday varying significant associations between physical activity and mental health assessments among our cohort population, which shed light on possible intervention strategies targeting daily physical activity patterns to improve school adolescent mental health. Our method is also used in environmental data to illustrate the wide applicability. Supplementary materials for this article are available online.

stat.ME↗

arXiv2023

Efficient Toxic Content Detection by Bootstrapping and Distilling Large Language Models

Toxic content detection is crucial for online services to remove inappropriate content that violates community standards. To automate the detection process, prior works have proposed varieties of machine learning (ML) approaches to train Language Models (LMs) for toxic content detection. However, both their accuracy and transferability across datasets are limited. Recently, Large Language Models (LLMs) have shown promise in toxic content detection due to their superior zero-shot and few-shot in-context learning ability as well as broad transferability on ML tasks. However, efficiently designing prompts for LLMs remains challenging. Moreover, the high run-time cost of LLMs may hinder their deployments in production. To address these challenges, in this work, we propose BD-LLM, a novel and efficient approach to Bootstrapping and Distilling LLMs for toxic content detection. Specifically, we design a novel prompting method named Decision-Tree-of-Thought (DToT) to bootstrap LLMs' detection performance and extract high-quality rationales. DToT can automatically select more fine-grained context to re-prompt LLMs when their responses lack confidence. Additionally, we use the rationales extracted via DToT to fine-tune student LMs. Our experimental results on various datasets demonstrate that DToT can improve the accuracy of LLMs by up to 4.6%. Furthermore, student LMs fine-tuned with rationales extracted via DToT outperform baselines on all datasets with up to 16.9\% accuracy improvement, while being more than 60x smaller than conventional LLMs. Finally, we observe that student LMs fine-tuned with rationales exhibit better cross-dataset transferability.

cs.CL↗

arXiv2023

Human Transcription Quality Improvement

High quality transcription data is crucial for training automatic speech recognition (ASR) systems. However, the existing industry-level data collection pipelines are expensive to researchers, while the quality of crowdsourced transcription is low. In this paper, we propose a reliable method to collect speech transcriptions. We introduce two mechanisms to improve transcription quality: confidence estimation based reprocessing at labeling stage, and automatic word error correction at post-labeling stage. We collect and release LibriCrowd - a large-scale crowdsourced dataset of audio transcriptions on 100 hours of English speech. Experiment shows the Transcription WER is reduced by over 50%. We further investigate the impact of transcription error on ASR model performance and found a strong correlation. The transcription quality improvement provides over 10% relative WER reduction for ASR models. We release the dataset and code to benefit the research community.

cs.CL↗

arXiv2023

HTEC: Human Transcription Error Correction

High-quality human transcription is essential for training and improving Automatic Speech Recognition (ASR) models. Recent study~\cite{libricrowd} has found that every 1% worse transcription Word Error Rate (WER) increases approximately 2% ASR WER by using the transcriptions to train ASR models. Transcription errors are inevitable for even highly-trained annotators. However, few studies have explored human transcription correction. Error correction methods for other problems, such as ASR error correction and grammatical error correction, do not perform sufficiently for this problem. Therefore, we propose HTEC for Human Transcription Error Correction. HTEC consists of two stages: Trans-Checker, an error detection model that predicts and masks erroneous words, and Trans-Filler, a sequence-to-sequence generative model that fills masked positions. We propose a holistic list of correction operations, including four novel operations handling deletion errors. We further propose a variant of embeddings that incorporates phoneme information into the input of the transformer. HTEC outperforms other methods by a large margin and surpasses human annotators by 2.2% to 4.5% in WER. Finally, we deployed HTEC to assist human annotators and showed HTEC is particularly effective as a co-pilot, which improves transcription quality by 15.1% without sacrificing transcription velocity.

eess.AS↗

arXiv2023

Functional Adaptive Double-Sparsity Estimator for Functional Linear Regression Model with Multiple Functional Covariates

Sensor devices have been increasingly used in engineering and health studies recently, and the captured multi-dimensional activity and vital sign signals can be studied in association with health outcomes to inform public health. The common approach is the scalar-on-function regression model, in which health outcomes are the scalar responses while high-dimensional sensor signals are the functional covariates, but how to effectively interpret results becomes difficult. In this study, we propose a new Functional Adaptive Double-Sparsity (FadDoS) estimator based on functional regularization of sparse group lasso with multiple functional predictors, which can achieve global sparsity via functional variable selection and local sparsity via zero-subinterval identification within coefficient functions. We prove that the FadDoS estimator converges at a bounded rate and satisfies the oracle property under mild conditions. Extensive simulation studies confirm the theoretical properties and exhibit excellent performances compared to existing approaches. Application to a Kinect sensor study that utilized an advanced motion sensing device tracking human multiple joint movements and conducted among community-dwelling elderly demonstrates how the FadDoS estimator can effectively characterize the detailed association between joint movements and physical health assessments. The proposed method is not only effective in Kinect sensor analysis but also applicable to broader fields, where multi-dimensional sensor signals are collected simultaneously, to expand the use of sensor devices in health studies and facilitate sensor data analysis.

stat.ME↗

arXiv2023

Evidence of a hydrated mineral enriched in water and ammonium molecules in the Chang'e-5 lunar sample

The presence and distribution of water on the Moon are fundamental to our understanding of the Earth-Moon system. Despite extensive research and remote detection, the origin and chemical form of lunar water (H2O) have remained elusive. In this study, we present the discovery of a hydrated mineral, (NH4)MgCl3*6H2O, in lunar soil samples returned by the Chang'e-5 mission, containing approximately 41 wt% H2O. The mineral's structure and composition closely resemble novograblenovite, a terrestrial fumarole mineral formed through the reaction of hot basalt with water-rich volcanic gases, and carnallite, an earth evaporite mineral. We rule out terrestrial contamination or rocket exhaust as the origin of this hydrate, based on its chemical and isotopic compositions and formation conditions. The presence of ammonium indicates a more complex lunar degassing history and highlights its potential as a resource for lunar habitation. Our findings also suggest that water molecules can persist in sunlit areas of the Moon as hydrated salt, providing crucial constraints to the fugacity of water and ammonia vapor in lunar volcanic gases.

astro-ph.EP↗

arXiv2022

Two-Dimensional Electron Gas with High Mobility Forming at BaO/SrTiO3 Interface

Two-dimensional electron gas (2DEG) formed at the interface between two insulating oxides offers an opportunity for fundamental research and device applications. Binary alkaline earth metal oxides possess compatible lattice constants with both silicon and perovskite oxides, exhibiting an enormous potential to bridging those two materials classes for multifunctionalities. Here we report the formation of 2DEG at the interface between the rock-salt BaO and SrTiO3. The highest electron mobility reaches 69000 cm^2 V.S^-1 at 2 K, leading to the typical Shubniko de Haas (SdH) oscillations under the high magnetic fields. The presence of SdH oscillations at different field-angles reveals a quasi-two-dimensional character of the Fermi surface. The first-principles calculations suggest that the effective charge transfer from the BaO to Ti 3dxy orbital at the interfaces is responsible to the observed high carrier mobility. Our results demonstrate that the BaO/STO heterointerface is a platform for exploring the correlated quantum phases, opening a door to the low-power and mesoscopic electronic devices.

cond-mat.mtrl-sci↗

arXiv2020

Miscibility gap and possible intrinsic Griffiths phase in Sr(Fe1-xMnx)2As2 crystals grown by transitional metal arsenide flux

The crystal structure, magnetic, electronic, and thermal properties of Mn-doped SrFe2As2 crystals have been systematically investigated. A miscibility gap is found in the system from x = 0.4362(4) to x = 0.9612(9). For x < 0.2055(2), the single crystals holding tetragonal structure (space group I4/mmm (no. 139)) have a continuously enlarged lattice parameter c, followed by a phase separation with crystals holding both tetragonal and trigonal structures up to x = 0.4362(4). Beyond the miscibility gap, the crystals with x > 0.9612(9) hold the trigonal structure (space group P-3m1 (no. 164)). Moreover, the spin density wave ordering for x = 0 to x = 0.0973(1) is suppressed, followed by an abnormal and broadened increase of the ordering temperature for x = 0.0973(1) to x = 0.2055(2). Eliminating the possibility of real-space phase separation of Mn and Fe based on the results of X-ray diffraction, this novel phenomenon can be attributed to a possible intrinsic Griffiths phase. No any superconducting signals are observed down to 2 K in the whole composition range with 0 < x < 1. A phase diagram with multi-critical points of Mn-doped SrFe2As2 system is established accordingly.

cond-mat.supr-con↗

arXiv2018

Collaborative Alerts Ranking for Anomaly Detection

Given a large number of low-level heterogeneous categorical alerts from an anomaly detection system, how to characterize complex relationships between different alerts, filter out false positives, and deliver trustworthy rankings and suggestions to end users? This problem is motivated by and generalized from applications in enterprise security and attack scenario reconstruction. While existing techniques focus on either reconstructing abnormal scenarios or filtering out false positive alerts, it can be more advantageous to consider the two perspectives simultaneously in order to improve detection accuracy and better understand anomaly behaviors. In this paper, we propose CAR, a collaborative alerts ranking framework that exploits both temporal and content correlations from heterogeneous categorical alerts. CAR first builds a tree-based model to capture both short-term correlations and long-term dependencies in each alert sequence, which identifies abnormal action sequences. Then, an embedding-based model is employed to learn the content correlations between alerts via their heterogeneous categorical attributes. Finally, by incorporating both temporal and content dependencies into one optimization framework, CAR ranks both alerts and their corresponding alert patterns. Our experiments, using real-world enterprise monitoring data and real attacks launched by professional hackers, show that CAR can accurately identify true positive alerts and successfully reconstruct attack scenarios at the same time.

cs.CR↗