EVODiff: Entropy-aware Variance Optimized Diffusion Inference, Shigui Li, Wei Chen, Delu Zeng* [Bib]

NeurIPS 2025 | Paper | Code | News🎉

• Proposes EVODiff, a fast inference method for diffusion models that optimizes conditional entropy during denoising.
• Generates higher-quality images with fewer steps (e.g., 25% fewer steps on ImageNet-256) while significantly reducing artifacts.

Entropy-informed weighting channel normalizing flow for deep generative models, Wei Chen#, Shian Du#, Shigui Li#, Delu Zeng*, John Paisley [Bib]

Pattern Recognition (PR) 2025 | Paper | Code

• Proposes EIW-Flow, which adaptively assigns channel-wise weights and shuffles latent variables in normalizing flows.
• Achieves state-of-the-art density estimation on CIFAR-10, CelebA, and ImageNet with negligible extra cost.

ReciprocalLA-LLIE: Low-light image enhancement with luminance-aware reciprocal diffusion process, Zhiqi Lin, Wei Chen, Jian Xu, Delu Zeng*, Min Chen [Bib]

Neurocomputing 2025 | Paper

• Proposes a reciprocal diffusion process within DDPM that iteratively enhances low-light images.
• Introduces a Luminance Adjustment Block for robust global brightness control, recovering details in dark regions.

To-Flow: Efficient Continuous Normalizing Flows with Temporal Optimization Adjoint with Moving Speed, Shian Du#, Yihong Luo#, Wei Chen#, Jian Xu, Delu Zeng* [Bib]

CVPR 2022 | Paper | Code

• Proposes To-Flow, which optimizes the evolutionary time of neural ODEs via coordinate descent to speed up continuous normalizing flow training.
• Accelerates training by ~20% without sacrificing generation quality, and is compatible with existing regularization methods.

Towards Disentangled Preference Optimization Dynamics: Suppress the Loser, Preserve the Winner, Wei Chen, Yubing Wu, Junmei Yang, Delu Zeng*, Qibin Zhao, John Paisley, Min Chen, Zhou Wang [Bib]

ICML 2026 | Paper | Code

• Reveals that diverse preference optimization objectives share the same update direction and differ only in two scalar weights, and identifies a simple condition (disentanglement band) for the desired learning pathway.
• Proposes a plug-and-play reward calibration method that rebalances updates to stay within this band, improving LLM alignment without modifying the base objective.

One-Step Score-Based Density Ratio Estimation, Wei Chen, Qibin Zhao, John Paisley, Junmei Yang, Delu Zeng* [Bib]

arXiv 2026 | Paper | Code

• Proposes OS-DRE, a solver-free framework that decomposes the time score into spatial and temporal parts, with the temporal part solved analytically.
• Enables accurate density ratio estimation with only one function evaluation, combining the speed of classical methods with the accuracy of score-based approaches.

A Minimum Variance Path Principle for Accurate and Stable Score-Based Density Ratio Estimation, Wei Chen, Jiacheng Li, Shigui Li, Zhiqi Lin, Junmei Yang, John Paisley, Delu Zeng* [Bib]

ICLR 2026 | Paper | Code

• Resolves the path-dependence paradox in score-based methods by identifying the overlooked path variance term in training objectives.
• Derives a closed-form variance expression and learns optimal, data-adaptive interpolation paths automatically, achieving state-of-the-art DRE accuracy.

Diffusion Secant Alignment for Score-Based Density Ratio Estimation, Wei Chen, Shigui Li, Jiacheng Li, Jian Xu, Zhiqi Lin, Junmei Yang, Delu Zeng*, John Paisley, Qibin Zhao [Bib]

arXiv 2025 | Paper | Code

• Replaces high-variance tangent-based learning targets with their interval integral (secant), which is provably lower-variance and smoother for neural networks to learn.
• Achieves accurate density ratio estimation with fewer function evaluations, and handles large distribution discrepancies more robustly than prior methods.

Dequantified Diffusion-Schrödinger Bridge for Density Ratio Estimation, Wei Chen, Shigui Li, Jiacheng Li, Junmei Yang, John Paisley, Delu Zeng* [Bib]

ICML 2025 | Paper | Code

• Proposes D3RE, a unified framework that handles the density-chasm and support-chasm problems where existing methods fail.
• Combines diffusion bridges with optimal transport to expand support coverage and stabilize score learning, enabling robust estimation even for very different distributions.

DeepAR-Attention probabilistic prediction for stock price series, Jiacheng Li, Wei Chen, Zhiheng Zhou, Junmei Yang, Delu Zeng* [Bib]

Neural Computing and Applications 2024 | Paper

• Proposes DeepAR-Attention, combining DeepAR with attention mechanisms for probabilistic stock price forecasting.
• Captures complex temporal dependencies and provides uncertainty-aware predictions for financial time series.

Neural ordinary differential equation networks for fintech applications using IoT, Jiacheng Li, Wei Chen, Yican Liu, Junmei Yang, Delu Zeng*, Zhiheng Zhou [Bib]

IEEE Internet of Things Journal (IoTJ) 2024 | Paper

• Develops neural ODE network approaches that model continuous-time dynamics for fintech applications in IoT.
• Captures irregularly sampled financial data more naturally than discrete-time models.

Integrating Ordinary Differential Equations with Sparse Attention for Power Load Forecasting, Jiacheng Li, Wei Chen, Yican Liu, Junmei Yang, Zhiheng Zhou, Delu Zeng* [Bib]

IEEE Trans on Instrumentation and Measurement (T-IM) 2025 | Paper

• Proposes EvolvInformer, which integrates neural ODE solvers with sparse attention for long-sequence power load forecasting.
• Reduces forecasting error by 29.7% (MSE) while maintaining logarithmic memory complexity.

Generative Self-Supervised Time-Series Forecasting Leveraging Wavelet Diffusion, Jiacheng Li, Wei Chen, Yican Liu, Junmei Yang, Zhiheng Zhou, Delu Zeng* [Bib]

IEEE Trans on Instrumentation and Measurement (T-IM) 2025 | Paper | News🎉

• Proposes TimeWaveDiff, a lightweight self-supervised framework that combines wavelet decomposition with diffusion modeling for time series.
• Captures multi-scale periodic patterns and achieves superior long-term forecasting accuracy with significantly lower computational cost.

Diffinformer: Diffusion Informer model for long sequence time-series forecasting, Jiacheng Li, Wei Chen, Yican Liu, Junmei Yang, Zhiheng Zhou, Delu Zeng* [Bib]

Expert Systems with Applications (ESWA) 2025 | Paper

• Proposes Diffinformer, which combines conditional diffusion models with Informer’s efficient sparse attention for long-sequence time series forecasting.
• Achieves consistent improvements over existing methods across five large-scale real-world datasets.