Author name cluster

Lu Yang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

AAAI Conference 2026 Conference Paper

Exploring Position Encoding Mechanism in Diffusion U-Net for Training-free High-resolution Image Generation

Feng Zhou
Pu Cao
Yiyang Ma
Lu Yang
Yonghao Dang
Jianqin Yin

Denoising higher-resolution latents using a pre-trained U-Net often results in repetitive and disordered image patterns. In this work, we are motivated to reveal the intrinsic cause of such pattern disruption in high-resolution image generation. Through theoretical analysis and empirical studies, we reveal that the pre-trained U-Net fails to provide sufficient positional information for tokens at high-resolution. Specifically, 1) zero-padding serves as a critical mechanism for position encoding but lacks robustness across varying resolutions; and 2) tokens located farther from the feature map boundaries have increasing difficulty acquiring positional awareness, leading to pattern disruptions. Inspired by these findings, we propose a novel training-free approach for high-resolution generation, introducing a Progressive Boundary Complement (PBC) method. It creates dynamic virtual image boundaries inside the feature map to supplement position information at high resolution, enabling high-quality and rich-content high-resolution image synthesis. Extensive experiments show that our method significantly improves high-resolution image synthesis in terms of visual quality and content richness, achieving state-of-the-art performance.

PDF Details DOI

AAAI Conference 2025 Conference Paper

A Compact Implicit Neural Representation for Efficient Storage of Massive 4D Functional Magnetic Resonance Imaging

Ruoran Li
Runzhao Yang
Wenxin Xiang
Yuxiao Cheng
Tingxiong Xiao
Lu Yang
Jinli Suo

Functional Magnetic Resonance Imaging (fMRI) data is a widely used kind of four-dimensional biomedical data, which requires effective compression. However, fMRI compressing poses unique challenges due to its intricate temporal dynamics, low signal-to-noise ratio, and complicated underlying redundancies. This paper reports a novel compression paradigm specifically tailored for fMRI data based on Implicit Neural Representation (INR). The proposed approach focuses on removing the various redundancies among the time series by employing several methods, including (i) conducting spatial correlation modeling for intra-region dynamics, (ii) decomposing reusable neuronal activation patterns, and (iii) using proper initialization together with nonlinear fusion to describe the inter-region similarity. This scheme appropriately incorporates the unique features of fMRI data, and experimental results on publicly available datasets demonstrate the effectiveness of the proposed method, surpassing state-of-the-art algorithms in both conventional image quality evaluation metrics and fMRI downstream tasks. This work in this paper paves the way for sharing massive fMRI data at low bandwidth and high fidelity.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

How Far Are We from Optimal Reasoning Efficiency?

Jiaxuan Gao
Shu Yan
Qixin Tan
Lu Yang
Shusheng Xu
Wei Fu
Zhiyu Mei
Kaifeng Lyu

Large Reasoning Models (LRMs) demonstrate remarkable problem-solving capabilities through extended Chain-of-Thought (CoT) reasoning but often produce excessively verbose and redundant reasoning traces. This inefficiency incurs high inference costs and limits practical deployment. While existing fine-tuning methods aim to improve reasoning efficiency, assessing their efficiency gains remains challenging due to inconsistent evaluations. In this work, we introduce the reasoning efficiency frontiers, empirical upper bounds derived from fine-tuning a base LRM (DeepSeek-R1-Distill-Qwen-1. 5B/7B) across diverse approaches and training configurations. Based on these frontiers, we propose the Reasoning Efficiency Gap (REG), a unified metric quantifying deviations of any fine-tuned LRMs from these frontiers. Systematic evaluation on challenging mathematical benchmarks, AMC23, AIME24, and AIME25, reveals significant gaps in current methods: they either sacrifice accuracy for short length or use excessive tokens to achieve sub-optimal accuracies despite high overall accuracy. To reduce the efficiency gap, we propose REO-RL, a Reinforcement Learning algorithm that optimizes reasoning efficiency by targeting a sparse set of token budgets. Leveraging numerical integration over strategically selected budgets, REO-RL approximates the full efficiency objective with low error using a small set of token budgets. Experiments show that, compared to vanilla RL with outcome reward, REO-RL reduces the reasoning efficiency gap by 74. 5\% and 64. 2\% in the 1. 5B and 7B settings. The 7B LRM fine-tuned with REO-RL achieves reasoning conciseness surpassing frontier LRMs like Qwen3 and Claude Sonnet 3. 7. Ablation studies confirm the efficacy of our token budget strategy and highlight REO-RL’s flexibility across design choices. This work establishes a systematic framework for evaluating and optimizing reasoning efficiency in LRMs. We will release the related code, data, and models to support future research on efficient reasoning in LRMs.

PDF Details

ICRA Conference 2025 Conference Paper

LLGS: Unsupervised Gaussian Splatting for Image Enhancement and Reconstruction in Pure Dark Environment

Haoran Wang
Jingwei Huang
Lu Yang
Tianchen Deng
Gaojing Zhang
Mingrui Li

D Gaussian Splatting has shown remarkable capabilities in novel view rendering tasks and exhibits significant potential for multi-view optimization. However, the original 3D Gaussian Splatting lacks color representation for inputs in lowlight environments. Simply using enhanced images as inputs would lead to issues with multi-view consistency, and current single-view enhancement systems rely on pre-trained data, lacking scene generalization. These problems limit the application of 3D Gaussian Splatting in low-light conditions in the field of robotics, including high-fidelity modeling and feature matching. To address these challenges, we propose an unsupervised multiview stereoscopic system based on Gaussian Splatting, called Low-Light Gaussian Splatting (LLGS). This system aims to enhance images in low-light environments while reconstructing the scene. Our method introduces a decomposable Gaussian representation called M-Color, which separately characterizes color information for targeted enhancement. Furthermore, we propose an unsupervised optimization method with zeroknowledge priors, using direction-based enhancement to ensure multi-view consistency. Experiments conducted on real-world datasets demonstrate that our system outperforms state-of-theart methods in both low-light enhancement and 3D Gaussian Splatting.

Details

AAAI Conference 2023 Conference Paper

Stop-Gradient Softmax Loss for Deep Metric Learning

Lu Yang
Peng Wang
Yanning Zhang

Deep metric learning aims to learn a feature space that models the similarity between images, and feature normalization is a critical step for boosting performance. However directly optimizing L2-normalized softmax loss cause the network to fail to converge. Therefore some SOTA approaches appends a scale layer after the inner product to relieve the convergence problem, but it incurs a new problem that it's difficult to learn the best scaling parameters. In this letter, we look into the characteristic of softmax-based approaches and propose a novel learning objective function Stop-Gradient Softmax Loss (SGSL) to solve the convergence problem in softmax-based deep metric learning with L2-normalization. In addition, we found a useful trick named Remove the last BN-ReLU (RBR). It removes the last BN-ReLU in the backbone to reduce the learning burden of the model. Experimental results on four fine-grained image retrieval benchmarks show that our proposed approach outperforms most existing approaches, i.e., our approach achieves 75.9% on CUB-200-2011, 94.7% on CARS196 and 83.1% on SOP which outperforms other approaches at least 1.7%, 2.9% and 1.7% on Recall@1.

PDF Details DOI

EAAI Journal 2019 Journal Article

Mining of skyline patterns by considering both frequent and utility constraints

Jerry Chun-Wei Lin
Lu Yang
Philippe Fournier-Viger
Tzung-Pei Hong

Association-rule mining (ARM) or frequent itemset mining (FIM) is the most fundamental task in knowledge discovery, which is used to find the occurrence frequency of the item/sets in transactional database. The other factors such as weight, interestingness or unit profit of the items are not considered whether in ARM or FIM. To reveal more information, high-utility itemset mining (HUIM) was designed to consider both quantity and unit profit of items to discover the high-utility itemsets (HUIs). Several algorithms for FIM or HUIM were extensively studied but fewer works concern both frequency and utility together to provide better solutions in decision-making. In the past, the SKYMINE algorithm was designed to find the skyline frequent-utility patterns (SFUPs). A SFUP is a non-dominated pattern, in which each solution dominates the others by considering the aspects of frequency and utility. The SKYMINE algorithm needs, however, amounts of computation to level-wisely discover the SFUPs. In this paper, an efficient utility-list structure is used instead of the UP-tree structure used in SKYMINE to mine the SFUPs. Two algorithms are respectively designed by using the depth-first search (called SKYFUP-D) and breath-first search (SKYFUP-B) to mine the SFUPs. An efficient structure is also designed to record the maximal utility of the potential itemsets, thus reducing the computations for finding the SFUPs in the search space. Extensive experiments are conducted on several real-world and simulated datasets and the results indicate that the designed two algorithms have better performance than that of the state-of-the-art SKYMINE algorithm in terms of runtime, memory usage, search space size and the scalability.

Details DOI

EAAI Journal 2016 Journal Article

Mining high-utility itemsets based on particle swarm optimization

Jerry Chun-Wei Lin
Lu Yang
Philippe Fournier-Viger
Jimmy Ming-Thai Wu
Tzung-Pei Hong
Leon Shyue-Liang Wang
Justin Zhan

High-utility itemset mining (HUIM) is a critical issue in recent years since it can be used to reveal the profitable products by considering both the quantity and profit factors instead of frequent itemset mining (FIM) or association-rule mining (ARM). Several algorithms have been presented to mine high-utility itemsets (HUIs) and most of the designed algorithms have to handle the exponential search space for discovering HUIs when the number of distinct items and the size of database are very large. In the past, a heuristic HUPEumu-GRAM algorithm was proposed to mine HUIs based on genetic algorithm (GA). For the evolutionary computation (EC) techniques of particle swarm optimization (PSO), it only requires fewer parameters compared to the GA-based approach. Since the traditional PSO mechanism is used to handle the continuous problem, in this paper, the discrete PSO is adopted to encode the particles as the binary variables. An efficient PSO-based algorithm namely HUIM-BPSOsig is proposed to efficiently find HUIs. It first sets the number of discovered high-transaction-weighted utilization 1-itemsets (1-HTWUIs) as the size of a particle based on transaction-weighted utility (TWU) model, which can greatly reduce the combinational problem in evolution process. The sigmoid function is adopted in the updating process of the particles of the designed HUIM-BPSOsig algorithm. Substantial experiments on real-life datasets show that the proposed algorithm has better results compared to the state-of-the-art GA-based algorithm.

Details DOI

TCS Journal 1990 Journal Article

The parallel numerical method of mechanical theorem proving

Jingzhong Zhang
Lu Yang
Mike Deng

In this paper, we present results of the work which allow us to prove geometry theorems by the parallel numerical method based on the multi-instance numerical verification of algebraic identity. The algebraic principle of the parallel numerical method is discussed and illustrated intuitively; the advantages of our method are given. It is acceptable on the complexity of both memory and time. It can be used to prove non-trivial geometric theorems by microcomputer, even by hand. We give some examples proved by parallel numerical method, including certain new unexpected results.

Details DOI