arXiv subjects

Junyi Fan

Publications and source records attributed to Junyi Fan.

At least 19 recordsLinked to original sources

arXiv2026

CANN Bench: Benchmarking Agent Generated Kernels against Real NPU and Algorithmic Limits

AI agents are now capable of writing, compiling, and iteratively optimizing low-level operator kernels on different hardware platforms. Existing benchmarks, however, focus almost exclusively on CUDA and Triton, leaving hardware ecosystems with less-exposed programming models without a common evaluation baseline. We present CANN Bench, an open benchmark for AI-generated operator code on Huawei's Ascend NPU. The current release covers 53 operators and 1060 test cases organized into four difficulty tiers -- from simple elementwise primitives to MoE dispatch and FlashAttention kernels -- spanning FP16, BF16, FP32, and INT8 precision formats. Evaluation adopts a \textbf{three-dimensional weighted composite score} that treats compilation, functional correctness, and performance as independent axes, providing a principled reward signal for kernel-generation agents. Performance is graded against an out-of-the-box PyTorch-on-Ascend baseline and an analytical per-case Hardware-Anchored Performance (HAP) limit on real NPU hardware, ensuring scores reflect genuine optimization headroom rather than measurement artifacts. The evaluation harness is designed to resist reward hacking from the ground up. CANN Bench is versioned within the official CANN repository and is designed for long-term community co-construction, providing the Ascend ecosystem with a quantitative, reproducible, and sustainably maintained yardstick for AI operator-authoring capability.

cs.AI↗

arXiv2026

PrefSQA: Pairwise Preference Prediction for Speech Quality Assessment and the Critical Role of High Quality Datasets

Mean opinion scores (MOS) are widely used for speech quality assessment, yet scalar labels are sensitive to rater variability and listening test differences. This introduces labeling noise, which limits the reliability of MOS prediction. Preference prediction reduces this variability as listeners compare signals directly, producing cleaner labels. We study MOS-free preference prediction and propose PrefSQA, which incorporates uncertainty-aware logits, an impairment attention head, and a module based on non-matching-reference comparisons. We use and refine five datasets, including MOS-derived and low-noise simulated sets with matching and non-matching content, experiment with human preference sets, and test on unseen data. Experiments show small improvements on MOS-derived data, while other sets reveal clear improvement over the baselines, highlighting the value of high-quality preference data and demonstrating the effectiveness of the proposed method.

cs.SD↗

arXiv2026

Multi-Scale Dequant: Eliminating Dequantization Bottleneck via Activation Decomposition for Efficient LLM Inference

Quantization is essential for efficient large language model (LLM) inference, yet the dequantization step-converting low-bit weights back to high-precision for matrix multiplication has become a critical bottleneck on modern AI accelerators. On architectures with decoupled compute units (e.g., Ascend NPUs), dequantization operations can consume more cycles than the matrix multiplication itself, leaving the high-throughput tensor cores underutilized. This paper presents Multi-Scale Dequant (MSD), a quantization framework that removes weight/KV dequantization from the GEMM critical path. Instead of lifting low-bit weights to BF16 precision, MSD decomposes high-precision BF16 activations into multiple low-precision components, each of which can be multiplied directly with quantized weights via native hardware-accelerated GEMM. This approach shifts the computational paradigm from precision conversion to multi-scale approximation, avoiding INT8-to-BF16 weight conversion before GEMM. We instantiate MSD for two weight formats and derive tight error bounds for each. For INT8 weights (W4A16), two-pass INT8 decomposition achieves near 16 effective bits. For MXFP4 weights (W4A16), two-pass MXFP4 decomposition yields near 6.6 effective bits with error bound 1/64 per block surpassing single-pass MXFP8(5.24 bits) while maintaining the same effective GEMM compute time. We further derive closed-form latency and HBM traffic models showing that MSD avoids the Vector-Cube pipeline stall caused by dequantization and reduces KV cache HBM traffic by up to 2.5 times in attention. Numerical simulations on matrix multiplication and Flash Attention kernels confirm that MSD does not degrade accuracy compared to dequantization baselines, and in many settings achieves lower L2 error.

stat.ML↗

arXiv2026

Which Voices Move Markets? Speaker Identity and the Cross-Section of Post-Earnings Returns

We utilize FinBERT, a domain-specific transformer model, to parse 6.5 million sentences from 16,428 S&P 500 quarterly earnings call transcripts (2015-2025) and demonstrate that post-earnings stock returns are not equally affected by all speakers in a conference call. Our section-weighted sentiment, with empirically derived speaker weights (Analyst 49%, CFO 30%, Executive 16%, Other 5%), achieves an out-of-sample Spearman IC of 0.142 versus 0.115 in-sample, generates monthly long-short alpha of 2.03% unexplained by the Fama-French five-factor model (t = 6.49), and remains significant after controlling for standardized unexpected earnings (SUE). FinBERT section-weighted sentiment entirely subsumes the Loughran-McDonald dictionary approach (FinBERT t = 5.90; LM t = 0.86 in the combined specification). Signal decay analysis and cumulative abnormal return charts confirm gradual price adjustment consistent with sluggish assimilation of soft information. All results undergo rigorous out-of-sample validation with an explicit temporal split, yielding improved rather than deteriorated predictive power.

q-fin.TR↗

arXiv2025

A Cross-Chain Event-Driven Data Infrastructure for Aave Protocol Analytics and Applications

Decentralized lending protocols, exemplified by Aave V3, have transformed financial intermediation by enabling permissionless, multi-chain borrowing and lending without intermediaries. Despite managing over $10 billion in total value locked, empirical research remains severely constrained by the lack of standardized, cross-chain event-level datasets. This paper introduces the first comprehensive, event-driven data infrastructure for Aave V3 spanning six major EVM-compatible chains (Ethereum, Arbitrum, Optimism, Polygon, Avalanche, and Base) from respective deployment blocks through October 2025. We collect and fully decode eight core event types -- Supply, Borrow, Withdraw, Repay, LiquidationCall, FlashLoan, ReserveDataUpdated, and MintedToTreasury -- producing over 50 million structured records enriched with block metadata and USD valuations. Using an open-source Python pipeline with dynamic batch sizing and automatic sharding (each file less than or equal to 1 million rows), we ensure strict chronological ordering and full reproducibility. The resulting publicly available dataset enables granular analysis of capital flows, interest rate dynamics, liquidation cascades, and cross-chain user behavior, providing a foundational resource for future studies on decentralized lending markets and systemic risk.

cs.DB↗

arXiv2025

FlowPortal: Residual-Corrected Flow for Training-Free Video Relighting and Background Replacement

Video relighting with background replacement is a challenging task critical for applications in film production and creative media. Existing methods struggle to balance temporal consistency, spatial fidelity, and illumination naturalness. To address these issues, we introduce FlowPortal, a novel training-free flow-based video relighting framework. Our core innovation is a Residual-Corrected Flow mechanism that transforms a standard flow-based model into an editing model, guaranteeing perfect reconstruction when input conditions are identical and enabling faithful relighting when they differ, resulting in high structural consistency. This is further enhanced by a Decoupled Condition Design for precise lighting control and a High-Frequency Transfer mechanism for detail preservation. Additionally, a masking strategy isolates foreground relighting from background pure generation process. Experiments demonstrate that FlowPortal achieves superior performance in temporal coherence, structural preservation, and lighting realism, while maintaining high efficiency. Project Page: https://gaowenshuo.github.io/FlowPortalProject/.

cs.CV↗

arXiv2025

From the perspective of perceptual speech quality: The robustness of frequency bands to noise

Speech quality is one of the main foci of speech-related research, where it is frequently studied with speech intelligibility, another essential measurement. Band-level perceptual speech intelligibility, however, has been studied frequently, whereas speech quality has not been thoroughly analyzed. In this paper, a Multiple Stimuli With Hidden Reference and Anchor (MUSHRA) inspired approach was proposed to study the individual robustness of frequency bands to noise with perceptual speech quality as the measure. Speech signals were filtered into thirty-two frequency bands with compromising real-world noise employed at different signal-to-noise ratios. Robustness to noise indices of individual frequency bands was calculated based on the human-rated perceptual quality scores assigned to the reconstructed noisy speech signals. Trends in the results suggest the mid-frequency region appeared less robust to noise in terms of perceptual speech quality. These findings suggest future research aiming at improving speech quality should pay more attention to the mid-frequency region of the speech signals accordingly.

eess.AS↗

arXiv2025

Aligning Language Models with Clinical Expertise: DPO for Heart Failure Nursing Documentation in Critical Care

Nursing documentation in intensive care units (ICUs) provides essential clinical intelligence but often suffers from inconsistent terminology, informal styles, and lack of standardization, challenges that are particularly critical in heart failure care. This study applies Direct Preference Optimization (DPO) to adapt Mistral-7B, a locally deployable language model, using 8,838 heart failure nursing notes from the MIMIC-III database and 21,210 preference pairs derived from expert-verified GPT outputs, model generations, and original notes. Evaluation across BLEU, ROUGE, BERTScore, Perplexity, and expert qualitative assessments demonstrates that DPO markedly enhances documentation quality. Specifically, BLEU increased by 84% (0.173 to 0.318), BERTScore improved by 7.6% (0.828 to 0.891), and expert ratings rose across accuracy (+14.4 points), completeness (+14.5 points), logical consistency (+14.1 points), readability (+11.1 points), and structural clarity (+6.0 points). These results indicate that DPO can align lightweight clinical language models with expert standards, supporting privacy-preserving, AI-assisted documentation within electronic health record systems to reduce administrative burden and improve ICU patient safety.

cs.CL↗

arXiv2025

Interpretable Machine Learning Model for Early Prediction of Acute Kidney Injury in Critically Ill Patients with Cirrhosis: A Retrospective Study

Background: Cirrhosis is a progressive liver disease with high mortality and frequent complications, notably acute kidney injury (AKI), which occurs in up to 50% of hospitalized patients and worsens outcomes. AKI stems from complex hemodynamic, inflammatory, and metabolic changes, making early detection essential. Many predictive tools lack accuracy, interpretability, and alignment with intensive care unit (ICU) workflows. This study developed an interpretable machine learning model for early AKI prediction in critically ill patients with cirrhosis. Methods: We conducted a retrospective analysis of the MIMIC-IV v2.2 database, identifying 1240 adult ICU patients with cirrhosis and excluding those with ICU stays under 48 hours or missing key data. Laboratory and physiological variables from the first 48 hours were extracted. The pipeline included preprocessing, missingness filtering, LASSO feature selection, and SMOTE class balancing. Six algorithms-LightGBM, CatBoost, XGBoost, logistic regression, naive Bayes, and neural networks-were trained and evaluated using AUROC, accuracy, F1-score, sensitivity, specificity, and predictive values. Results: LightGBM achieved the best performance (AUROC 0.808, 95% CI 0.741-0.856; accuracy 0.704; NPV 0.911). Key predictors included prolonged partial thromboplastin time, absence of outside-facility 20G placement, low pH, and altered pO2, consistent with known cirrhosis-AKI mechanisms and suggesting actionable targets. Conclusion: The LightGBM-based model enables accurate early AKI risk stratification in ICU patients with cirrhosis using routine clinical variables. Its high negative predictive value supports safe de-escalation for low-risk patients, and interpretability fosters clinician trust and targeted prevention. External validation and integration into electronic health record systems are warranted.

cs.LG↗

arXiv2025

Prediction of Significant Creatinine Elevation in First ICU Stays with Vancomycin Use: A retrospective study through Catboost

Background: Vancomycin, a key antibiotic for severe Gram-positive infections in ICUs, poses a high nephrotoxicity risk. Early prediction of kidney injury in critically ill patients is challenging. This study aimed to develop a machine learning model to predict vancomycin-related creatinine elevation using routine ICU data. Methods: We analyzed 10,288 ICU patients (aged 18-80) from the MIMIC-IV database who received vancomycin. Kidney injury was defined by KDIGO criteria (creatinine rise >=0.3 mg/dL within 48h or >=50% within 7d). Features were selected via SelectKBest (top 30) and Random Forest ranking (final 15). Six algorithms were tested with 5-fold cross-validation. Interpretability was evaluated using SHAP, Accumulated Local Effects (ALE), and Bayesian posterior sampling. Results: Of 10,288 patients, 2,903 (28.2%) developed creatinine elevation. CatBoost performed best (AUROC 0.818 [95% CI: 0.801-0.834], sensitivity 0.800, specificity 0.681, negative predictive value 0.900). Key predictors were phosphate, total bilirubin, magnesium, Charlson index, and APSIII. SHAP confirmed phosphate as a major risk factor. ALE showed dose-response patterns. Bayesian analysis estimated mean risk 60.5% (95% credible interval: 16.8-89.4%) in high-risk cases. Conclusions: This machine learning model predicts vancomycin-associated creatinine elevation from routine ICU data with strong accuracy and interpretability, enabling early risk detection and supporting timely interventions in critical care.

cs.LG↗

arXiv2025

Early Mortality Prediction in ICU Patients with Hypertensive Kidney Disease Using Interpretable Machine Learning

Background: Hypertensive kidney disease (HKD) patients in intensive care units (ICUs) face high short-term mortality, but tailored risk prediction tools are lacking. Early identification of high-risk individuals is crucial for clinical decision-making. Methods: We developed a machine learning framework to predict 30-day in-hospital mortality among ICU patients with HKD using early clinical data from the MIMIC-IV v2.2 database. A cohort of 1,366 adults was curated with strict criteria, excluding malignancy cases. Eighteen clinical features-including vital signs, labs, comorbidities, and therapies-were selected via random forest importance and mutual information filtering. Several models were trained and compared with stratified five-fold cross-validation; CatBoost demonstrated the best performance. Results: CatBoost achieved an AUROC of 0.88 on the independent test set, with sensitivity of 0.811 and specificity of 0.798. SHAP values and Accumulated Local Effects (ALE) plots showed the model relied on meaningful predictors such as altered consciousness, vasopressor use, and coagulation status. Additionally, the DREAM algorithm was integrated to estimate patient-specific posterior risk distributions, allowing clinicians to assess both predicted mortality and its uncertainty. Conclusions: We present an interpretable machine learning pipeline for early, real-time risk assessment in ICU patients with HKD. By combining high predictive performance with uncertainty quantification, our model supports individualized triage and transparent clinical decisions. This approach shows promise for clinical deployment and merits external validation in broader critical care populations.

cs.LG↗

arXiv2025

JSQA: Speech Quality Assessment with Perceptually-Inspired Contrastive Pretraining Based on JND Audio Pairs

Speech quality assessment (SQA) is often used to learn a mapping from a high-dimensional input space to a scalar that represents the mean opinion score (MOS) of the perceptual speech quality. Learning such a mapping is challenging for many reasons, but largely because MOS exhibits high levels of inherent variance due to perceptual and experimental-design differences. Many solutions have been proposed, but many approaches do not properly incorporate perceptual factors into their learning algorithms (beyond the MOS label), which could lead to unsatisfactory results. To this end, we propose JSQA, a two-stage framework that pretrains an audio encoder using perceptually-guided contrastive learning on just noticeable difference (JND) pairs, followed by fine-tuning for MOS prediction. We first generate pairs of audio data within JND levels, which are then used to pretrain an encoder to leverage perceptual quality similarity information and map it into an embedding space. The JND pairs come from clean LibriSpeech utterances that are mixed with background noise from CHiME-3, at different signal-to-noise ratios (SNRs). The encoder is later fine-tuned with audio samples from the NISQA dataset for MOS prediction. Experimental results suggest that perceptually-inspired contrastive pretraining significantly improves the model performance evaluated by various metrics when compared against the same network trained from scratch without pretraining. These findings suggest that incorporating perceptual factors into pretraining greatly contributes to the improvement in performance for SQA.

eess.AS↗

arXiv2025

Interpretable Machine Learning Model for Early Prediction of 30-Day Mortality in ICU Patients With Coexisting Hypertension and Atrial Fibrillation: A Retrospective Cohort Study

Hypertension and atrial fibrillation (AF) often coexist in critically ill patients, significantly increasing mortality rates in the ICU. Early identification of high-risk individuals is crucial for targeted interventions. However, limited research has focused on short-term mortality prediction for this subgroup. This study analyzed 1,301 adult ICU patients with hypertension and AF from the MIMIC-IV database. Data including chart events, laboratory results, procedures, medications, and demographic information from the first 24 hours of ICU admission were extracted. After quality control, missing data imputation, and feature selection, 17 clinically relevant variables were retained. The cohort was split into training (70%) and test (30%) sets, with outcome-weighted training applied to address class imbalance. The CatBoost model, along with five baseline models (LightGBM, XGBoost, logistic regression, Naive Bayes, and neural networks), was evaluated using five-fold cross-validation, with AUROC as the primary performance metric. Model interpretability was assessed using SHAP, ALE, and DREAM analyses. The CatBoost model showed strong performance with an AUROC of 0.889 (95% CI: 0.840-0.924), accuracy of 0.831, and sensitivity of 0.837. Key predictors identified by SHAP and other methods included the Richmond-RAS Scale, pO2, CefePIME, and Invasive Ventilation, demonstrating the model's robustness and clinical applicability. This model shows strong performance and interpretability in early mortality prediction, enabling early intervention and personalized care decisions. Future work will involve multi-center validation and extending the approach to other diseases.

stat.AP↗

arXiv2025

Predicting Short-Term Mortality in Elderly ICU Patients with Diabetes and Heart Failure: A Distributional Inference Framework

Elderly ICU patients with coexisting diabetes mellitus and heart failure experience markedly elevated short-term mortality, yet few predictive models are tailored to this high-risk group. Diabetes mellitus affects nearly 30% of U.S. adults over 65, and significantly increases the risk of heart failure. When combined, these conditions worsen frailty, renal dysfunction, and hospitalization risk, leading to one-year mortality rates of up to 40%. Despite their clinical burden and complexity, no established models address individualized mortality prediction in elderly ICU patients with both diabetes mellitus and heart failure. We developed and validated a probabilistic mortality prediction framework using the MIMIC-IV database, targeting 65-90-year-old patients with both diabetes mellitus and heart failure. Using a two-stage feature selection pipeline and a cohort of 1,478 patients, we identified 19 clinically significant variables that reflect physiology, comorbidities, and intensity of treatment. Among six ML models benchmarked, CatBoost achieved the highest test AUROC (0.863), balancing performance and interpretability. To enhance clinical relevance, we employed the DREAM algorithm to generate posterior mortality risk distributions rather than point estimates, enabling assessment of both risk magnitude and uncertainty. This distribution-aware approach facilitates individualized triage in complex ICU settings. Interpretability was further supported via ablation and ALE analysis, highlighting key predictors such as APS III, oxygen flow, GCS eye, and Braden Mobility. Our model enables transparent, personalized, and uncertainty-informed decision support for high-risk ICU populations.

stat.AP↗

arXiv2025

Clinically Interpretable Mortality Prediction for ICU Patients with Diabetes and Atrial Fibrillation: A Machine Learning Approach

Background: Patients with both diabetes mellitus (DM) and atrial fibrillation (AF) face elevated mortality in intensive care units (ICUs), yet models targeting this high-risk group remain limited. Objective: To develop an interpretable machine learning (ML) model predicting 28-day mortality in ICU patients with concurrent DM and AF using early-phase clinical data. Methods: A retrospective cohort of 1,535 adult ICU patients with DM and AF was extracted from the MIMIC-IV database. Data preprocessing involved median/mode imputation, z-score normalization, and early temporal feature engineering. A two-step feature selection pipeline-univariate filtering (ANOVA F-test) and Random Forest-based multivariate ranking-yielded 19 interpretable features. Seven ML models were trained with stratified 5-fold cross-validation and SMOTE oversampling. Interpretability was assessed via ablation and Accumulated Local Effects (ALE) analysis. Results: Logistic regression achieved the best performance (AUROC: 0.825; 95% CI: 0.779-0.867), surpassing more complex models. Key predictors included RAS, age, bilirubin, and extubation. ALE plots showed intuitive, non-linear effects such as age-related risk acceleration and bilirubin thresholds. Conclusion: This interpretable ML model offers accurate risk prediction and clinical insights for early ICU triage in patients with DM and AF.

cs.LG↗

arXiv2025

Predicting Postoperative Stroke in Elderly SICU Patients: An Interpretable Machine Learning Model Using MIMIC Data

Postoperative stroke remains a critical complication in elderly surgical intensive care unit (SICU) patients, contributing to prolonged hospitalization, elevated healthcare costs, and increased mortality. Accurate early risk stratification is essential to enable timely intervention and improve clinical outcomes. We constructed a combined cohort of 19,085 elderly SICU admissions from the MIMIC-III and MIMIC-IV databases and developed an interpretable machine learning (ML) framework to predict in-hospital stroke using clinical data from the first 24 hours of Intensive Care Unit (ICU) stay. The preprocessing pipeline included removal of high-missingness features, iterative Singular Value Decomposition (SVD) imputation, z-score normalization, one-hot encoding, and class imbalance correction via the Adaptive Synthetic Sampling (ADASYN) algorithm. A two-stage feature selection process-combining Recursive Feature Elimination with Cross-Validation (RFECV) and SHapley Additive exPlanations (SHAP)-reduced the initial 80 variables to 20 clinically informative predictors. Among eight ML models evaluated, CatBoost achieved the best performance with an AUROC of 0.8868 (95% CI: 0.8802--0.8937). SHAP analysis and ablation studies identified prior cerebrovascular disease, serum creatinine, and systolic blood pressure as the most influential risk factors. Our results highlight the potential of interpretable ML approaches to support early detection of postoperative stroke and inform decision-making in perioperative critical care.

q-bio.QM↗

arXiv2025

Development of Interactive Nomograms for Predicting Short-Term Survival in ICU Patients with Aplastic Anemia

Aplastic anemia is a rare, life-threatening hematologic disorder characterized by pancytopenia and bone marrow failure. ICU admission in these patients often signals critical complications or disease progression, making early risk assessment crucial for clinical decision-making and resource allocation. In this study, we used the MIMIC-IV database to identify ICU patients diagnosed with aplastic anemia and extracted clinical features from five domains: demographics, synthetic indicators, laboratory results, comorbidities, and medications. Over 400 variables were reduced to seven key predictors through machine learning-based feature selection. Logistic regression and Cox regression models were constructed to predict 7-, 14-, and 28-day mortality, and their performance was evaluated using AUROC. External validation was conducted using the eICU Collaborative Research Database to assess model generalizability. Among 1,662 included patients, the logistic regression model demonstrated superior performance, with AUROC values of 0.8227, 0.8311, and 0.8298 for 7-, 14-, and 28-day mortality, respectively, compared to the Cox model. External validation yielded AUROCs of 0.7391, 0.7119, and 0.7093. Interactive nomograms were developed based on the logistic regression model to visually estimate individual patient risk. In conclusion, we identified a concise set of seven predictors, led by APS III, to build validated and generalizable nomograms that accurately estimate short-term mortality in ICU patients with aplastic anemia. These tools may aid clinicians in personalized risk stratification and decision-making at the point of care.

cs.LG↗

arXiv2025

Predicting ICU Readmission in Acute Pancreatitis Patients Using a Machine Learning-Based Model with Enhanced Clinical Interpretability

Acute pancreatitis (AP) is a common and potentially life-threatening gastrointestinal disease that imposes a significant burden on healthcare systems. ICU readmissions among AP patients are common, especially in severe cases, with rates exceeding 40%. Identifying high-risk patients for readmission is crucial for improving outcomes. This study used the MIMIC-III database to identify ICU admissions for AP based on diagnostic codes. We applied a preprocessing pipeline including missing data imputation, correlation analysis, and hybrid feature selection. Recursive Feature Elimination with Cross-Validation (RFECV) and LASSO regression, supported by expert review, reduced over 50 variables to 20 key predictors, covering demographics, comorbidities, lab tests, and interventions. To address class imbalance, we used the Synthetic Minority Over-sampling Technique (SMOTE) in a five-fold cross-validation framework. We developed and optimized six machine learning models-Logistic Regression, k-Nearest Neighbors, Naive Bayes, Random Forest, LightGBM, and XGBoost-using grid search. Model performance was evaluated with AUROC, accuracy, F1 score, sensitivity, specificity, PPV, and NPV. XGBoost performed best, with an AUROC of 0.862 (95% CI: 0.800-0.920) and accuracy of 0.889 (95% CI: 0.858-0.923) on the test set. An ablation study showed that removing any feature decreased performance. SHAP analysis identified platelet count, age, and SpO2 as key predictors of readmission. This study shows that ensemble learning, informed feature selection, and handling class imbalance can improve ICU readmission prediction in AP patients, supporting targeted post-discharge interventions.

stat.AP↗