Author name cluster

Lu Wei

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.

5 papers

1 author row

EAAI Journal 2026 Journal Article

Physics-aware learning for detecting robust universal perturbation attacks in wind power forecasting

Liliang Wang
Zheng Qian
Jiaqi Ruan
Lu Wei
Gaoqi Liang

Wind power forecasting (WPF) models can be vulnerable to adversarial attacks, where minor perturbations may introduce large forecast errors. However, existing attack methods are typically customized for specific models, temporal instances, and wind farms, often suboptimal because they require excessive computational resources that are incompatible with real-time dispatch requirements. Moreover, effective defense mechanisms for addressing such adversarial attacks in WPF models remain scarce. This paper addresses these challenges through dual contributions: algorithmic innovations in artificial intelligence (AI) security and engineering applications for critical infrastructure protection. Algorithmically, we first develop a universal perturbation (UP) framework trained offline to target multiple WPF models across diverse spatiotemporal contexts. Building on this, we propose the robust universal perturbation (RUP) method, which uses weighted density ensemble learning to aggregate UPs generated under various conditions—such as different train–validation splits, initialization parameters, and model architectures—achieving superior transferability and robustness. Additionally, we introduce the physics-aware learning model (PALM), the first framework to leverage physical constraints for detecting adversarial attacks in WPF systems by quantifying deviations from established physical principles, marking a departure from conventional data-driven approaches. From an engineering perspective, RUP closely approaches state-of-the-art attack performance while reducing real-time computational overhead by 98% through offline preprocessing, facilitating practical deployment. PALM ensures operational integrity with 100% precision and recall, validated using operational wind farm datasets. These advances bridge AI security theory with critical infrastructure requirements, exposing vulnerabilities in safety-critical energy systems and delivering implementable solutions for robust power grid AI deployment.

Details DOI

TMLR Journal 2026 Journal Article

Retrieval as a Decision: Training-Free Adaptive Gating for Efficient RAG

Yufeng Wang
Lu Wei
Haibin Ling

Retrieval-Augmented Generation (RAG) improves factuality but retrieving for every query often hurts quality while inflating tokens and latency. We propose Training-free Adaptive Retrieval Gating (\textbf{TARG}), a single-shot policy that decides when to retrieve using only a short, no-context draft from the base model. From the draft’s prefix logits, TARG computes lightweight uncertainty scores—mean token entropy, a margin signal derived from the top-1/top-2 logit gap via a monotone link, or small-$N$ variance across a handful of stochastic prefixes—and triggers retrieval only when the score exceeds a threshold. The gate is model-agnostic, adds only tens to hundreds of draft tokens, and requires no additional training or auxiliary heads. On NQ-Open, TriviaQA, and PopQA, TARG consistently pushes the accuracy–efficiency frontier: compared with Always-RAG\footnote{\textsc{Always-RAG}: retrieve for every query; \textsc{Never-RAG}: never retrieve.}, TARG matches or improves EM/F1 while reducing retrieval by 70–90\% and cutting end-to-end latency, and it remains close to Never-RAG in overhead. A central empirical finding is that under modern instruction-tuned LLMs the margin signal is a robust default (entropy compresses as backbones sharpen), with small-$N$ variance offering a conservative, budget-first alternative. We provide ablations over gate type and prefix length and use a $\Delta$-latency view to make budget trade-offs explicit.

PDF Details

EAAI Journal 2025 Journal Article

Food multi-factor decoupling recognition based on progressive spatial-frequency attention distillation learning

Minkang Chai
Lu Wei
Zheng Qian
Ran Zhang
Ye Zhu
Baoqing Zhou

With the widespread application of image recognition technology in daily life, food image recognition faces challenges such as diverse categories and complex forms. Particularly when dealing with subtle differences between similar food items, imbalanced categories, feature ambiguities, and classification confusion caused by the coupling of multiple factors in food representation, existing models still have room for improvement in their recognition accuracy and generalization ability. Therefore, constructing a recognition model that can precisely differentiate food categories while effectively addressing the complexities of coupled factors has become a key issue in this field. In response to these challenges, we propose the innovative Progressive Spatial-Frequency Distillation Network (PSFDNet). By utilizing a unique multidimensional progressive learning strategy combined with an adaptive spatial-frequency attention mechanism, the model significantly enhances its feature extraction and discrimination capabilities within complex food structures. Additionally, we introduce the food correlation evaluation loss to decouple the mutual interference among food features effectively, thereby improving the accuracy and robustness of food image recognition. Extensive experiments verified the outstanding performance of PSFDNet across datasets, demonstrating a notable increase of 0. 87% in the Top-1 recognition accuracy and a 50% increase in inference speed. Particularly in recognizing food images characterized by highly coupled features and extremely imbalanced categories, PSFDNet exhibited significant performance advantages over other methods.

Details DOI

NeurIPS Conference 2024 Conference Paper

Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism

Ronast Subedi
Lu Wei
Wenhan Gao
Shayok Chakraborty
Yi Liu

Molecular learning is pivotal in many real-world applications, such as drug discovery. Supervised learning requires heavy human annotation, which is particularly challenging for molecular data, e. g. , the commonly used density functional theory (DFT) is highly computationally expensive. Active learning (AL) automatically queries labels for most informative samples, thereby remarkably alleviating the annotation hurdle. In this paper, we present a principled AL paradigm for molecular learning, where we treat molecules as 3D molecular graphs. Specifically, we propose a new diversity sampling method to eliminate mutual redundancy built on distributions of 3D geometries. We first propose a set of new 3D graph isometries for 3D graph isomorphism analysis. Our method is provably at least as expressive as the Geometric Weisfeiler-Lehman (GWL) test. The moments of the distributions of the associated geometries are then extracted for efficient diversity computing. To ensure our AL paradigm selects samples with maximal uncertainties, we carefully design a Bayesian geometric graph neural network to compute uncertainties specifically for 3D molecular graphs. We pose active sampling as a quadratic programming (QP) problem using the proposed components. Experimental results demonstrate the effectiveness of our AL paradigm, as well as the proposed diversity and uncertainty methods.

PDF Details DOI

NeurIPS Conference 2023 Conference Paper

CARE-MI: Chinese Benchmark for Misinformation Evaluation in Maternity and Infant Care

Tong Xiang
Liangzhi Li
Wangyue Li
Mingbai Bai
Lu Wei
Bowen Wang
Noa Garcia

The recent advances in natural language processing (NLP), have led to a new trend of applying large language models (LLMs) to real-world scenarios. While the latest LLMs are astonishingly fluent when interacting with humans, they suffer from the misinformation problem by unintentionally generating factually false statements. This can lead to harmful consequences, especially when produced within sensitive contexts, such as healthcare. Yet few previous works have focused on evaluating misinformation in the long-form (LF) generation of LLMs, especially for knowledge-intensive topics. Moreover, although LLMs have been shown to perform well in different languages, misinformation evaluation has been mostly conducted in English. To this end, we present a benchmark, CARE-MI, for evaluating LLM misinformation in: 1) a sensitive topic, specifically the maternity and infant care domain; and 2) a language other than English, namely Chinese. Most importantly, we provide an innovative paradigm for building LF generation evaluation benchmarks that can be transferred to other knowledge-intensive domains and low-resourced languages. Our proposed benchmark fills the gap between the extensive usage of LLMs and the lack of datasets for assessing the misinformation generated by these models. It contains 1, 612 expert-checked questions, accompanied with human-selected references. Using our benchmark, we conduct extensive experiments and found that current Chinese LLMs are far from perfect in the topic of maternity and infant care. In an effort to minimize the reliance on human resources for performance evaluation, we offer off-the-shelf judgment models for automatically assessing the LF output of LLMs given benchmark questions. Moreover, we compare potential solutions for LF generation evaluation and provide insights for building better automated metrics.

PDF Details