Selected work

Representative Papers

Quantum Circuits in Diffusion Models: A Fair-Comparison Study and a Mechanistic Analysis of Angle-Embedding Failures

Jul 10, 2026

This study investigates the efficacy of variational quantum circuits—such as EfficientSU2—in diffusion generative models, with a focus on the failure mechanism of angle embedding under unbounded score-matching objectives due to phase aliasing. To enable a fair evaluation, the authors propose a protocol that integrates quantum modules into DDPM and latent diffusion frameworks via squeeze-and-excitation structures, and conduct systematic assessments using NCSN-based score models, FID metrics, and multiple subsampling significance tests. Experiments show that quantum modules achieve average FID scores comparable to classical baselines despite using 4.5–9 times fewer parameters. Furthermore, applying a π·tanh(·) bounded input transformation effectively mitigates phase aliasing and significantly improves performance; however, the anticipated parameter efficiency advantage of quantum circuits is not realized.

0 citationsRead paper

PETRA: Pretrained Evolutionary Transformer for SARS-CoV-2 Mutation Prediction

Nov 06, 2025

SARS-CoV-2’s continuous evolution and immune-escape mutations pose significant challenges to vaccine design and public health response. To address key bottlenecks—including high noise in raw RNA sequences and spatiotemporally biased global sequencing data—this work introduces the first pre-trained Transformer model that takes phylogenetic tree evolutionary trajectories as input, abandoning direct modeling of raw sequences. Instead, it encodes branch paths and evolutionary distances, and incorporates a weighted loss framework to mitigate sampling bias. By integrating evolutionary biological priors with deep learning, our method achieves weighted recall scores of 9.45% (nucleotide-level) and 17.10% (spike protein amino acid-level) for mutation prediction—substantially outperforming baseline approaches. Furthermore, it enables real-time tracking of dominant variants and delivers an interpretable, deployable computational tool for prospective vaccine strain updates.

0 citationsRead paper

UDA: Unsupervised Debiasing Alignment for Pair-wise LLM-as-a-Judge

Aug 13, 2025

Pairwise evaluation of large language models (LLMs) suffers from preference bias, causing inter-judge ranking inconsistency. To address this, we propose an unsupervised debiased alignment framework that empirically identifies heterogeneous bias across model evaluations for the first time and introduces a consensus alignment mechanism minimizing Elo trajectory dispersion—enabling dynamic, annotation-free calibration of multi-judge scores. Our method employs a compact neural network to adaptively modulate the Elo K-factor and win-probability estimation, reducing inter-judge score standard deviation by up to 63.4% and improving average correlation with human judgments by 24.7%. It notably enhances low-performing judges’ reliability and promotes evaluation fairness. The core innovation lies in formalizing judge consistency as a trajectory alignment problem and establishing a theoretically grounded, unsupervised debiasing paradigm.

0 citationsRead paper

Recent publications

Latest Papers

Quantum Circuits in Diffusion Models: A Fair-Comparison Study and a Mechanistic Analysis of Angle-Embedding Failures

Jul 10, 2026

This study investigates the efficacy of variational quantum circuits—such as EfficientSU2—in diffusion generative models, with a focus on the failure mechanism of angle embedding under unbounded score-matching objectives due to phase aliasing. To enable a fair evaluation, the authors propose a protocol that integrates quantum modules into DDPM and latent diffusion frameworks via squeeze-and-excitation structures, and conduct systematic assessments using NCSN-based score models, FID metrics, and multiple subsampling significance tests. Experiments show that quantum modules achieve average FID scores comparable to classical baselines despite using 4.5–9 times fewer parameters. Furthermore, applying a π·tanh(·) bounded input transformation effectively mitigates phase aliasing and significantly improves performance; however, the anticipated parameter efficiency advantage of quantum circuits is not realized.

0 citationsRead paper

PETRA: Pretrained Evolutionary Transformer for SARS-CoV-2 Mutation Prediction

Nov 06, 2025

SARS-CoV-2’s continuous evolution and immune-escape mutations pose significant challenges to vaccine design and public health response. To address key bottlenecks—including high noise in raw RNA sequences and spatiotemporally biased global sequencing data—this work introduces the first pre-trained Transformer model that takes phylogenetic tree evolutionary trajectories as input, abandoning direct modeling of raw sequences. Instead, it encodes branch paths and evolutionary distances, and incorporates a weighted loss framework to mitigate sampling bias. By integrating evolutionary biological priors with deep learning, our method achieves weighted recall scores of 9.45% (nucleotide-level) and 17.10% (spike protein amino acid-level) for mutation prediction—substantially outperforming baseline approaches. Furthermore, it enables real-time tracking of dominant variants and delivers an interpretable, deployable computational tool for prospective vaccine strain updates.

0 citationsRead paper

UDA: Unsupervised Debiasing Alignment for Pair-wise LLM-as-a-Judge

Aug 13, 2025

Pairwise evaluation of large language models (LLMs) suffers from preference bias, causing inter-judge ranking inconsistency. To address this, we propose an unsupervised debiased alignment framework that empirically identifies heterogeneous bias across model evaluations for the first time and introduces a consensus alignment mechanism minimizing Elo trajectory dispersion—enabling dynamic, annotation-free calibration of multi-judge scores. Our method employs a compact neural network to adaptively modulate the Elo K-factor and win-probability estimation, reducing inter-judge score standard deviation by up to 63.4% and improving average correlation with human judgments by 24.7%. It notably enhances low-performing judges’ reliability and promotes evaluation fairness. The core innovation lies in formalizing judge consistency as a trajectory alignment problem and establishing a theoretically grounded, unsupervised debiasing paradigm.

0 citationsRead paper