Author name cluster

Zili Wang

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.

17 papers

2 author rows

EAAI Journal 2025 Journal Article

A physical-modulated framework for process optimization and shape inference of industrial metal tube

Le Wang
Zili Wang
Shuyou Zhang
Jianrong Tan
Yaochen Lin
Yongzhe Xiang

Accurate forming shape prediction and process optimization are crucial for ensuring the quality of tubular components throughout both the design and iteration phases. However, the nonlinear multi-physics coupling between plastic deformation and process attributes presents significant complexity. Although the two tasks are inherently interdependent, they are often treated as separate paradigms in industrial applications. This lack of a synergistic approach impedes the establishment of an efficient closed-loop manufacturing process. We propose a physical-modulated dual-branch prediction framework, called Forming Process to Three-Dimension (FP-3D). It operates under a unified feature scale, which interactively maps from the process attributes to the three-dimensional (3D) tube shape. It bridges branches by extracting structurally embedded geometric latent features as a reliable intermediate representation. The process optimization branch contrasts shape features to learn latent disparity in pairs. It alleviates limitations posed by sample quantity and encourages the model to learn process attribute adjustments as a historically measured shape deformed to target one. A physical-increment-modulated (PIM) layer is proposed to facilitate the accelerated learning of physical increments that are sensitive to process attributes. In the shape inference branch, we propose radially transferring features toward the implicit skeleton, which enables physical information to intervene in the latent space for controllable shape generation. Under the sole condition of process parameters, FP-3D allows the conditional generation of point-wise features to decode refined 3D shapes. The extensive experiments conducted on diverse tube and benchmark datasets demonstrate that FP-3D exhibits state-of-the-art performance.

Details DOI

AAAI Conference 2025 Conference Paper

AIM: Let Any Multimodal Large Language Models Embrace Efficient In-Context Learning

Jun Gao
Qian Qiao
Tianxiang Wu
Zili Wang
Ziqiang Cao
Wenjie Li

In-context learning (ICL) advances Large Language Models (LLMs) exhibiting emergent ability on downstream tasks without updating billions of parameters. However, in the area of multimodal Large Language Models (MLLMs), two problems hinder the application of multimodal ICL: (1) Most primary MLLMs are only trained on single-image datasets, making them unable to read extra multimodal demonstrations. (2) With the demonstrations increasing, thousands of visual tokens highly challenge hardware and degrade ICL performance. During preliminary explorations, we discovered that the inner LLM focuses more on the linguistic modality within multimodal demonstrations during generation. Therefore, we propose a general and lightweight framework AIM to tackle the mentioned problems through Aggregating Image information of Multimodal demonstrations to the latent space of the corresponding textual labels. After aggregation, AIM substitutes each demonstration with generated fused virtual tokens whose length is reduced to the same as its texts. Except for shortening input length, AIM further upgrades MLLMs pre-trained on image-text pairs to support multimodal ICL, as images from demonstrations are disregarded. Furthermore, benefiting from aggregating different demonstrations independently, AIM configures Demonstration Bank (DB) to avoid repeated aggregation, which significantly boosts model efficiency. We build AIM upon QWen-VL and LLaVA-Next, and AIM is comprehensively evaluated on image caption, VQA, and hateful speech detection. Outstanding results reveal that AIM provides an efficient and effective solution in upgrading MLLMs for multimodal ICL.

PDF Details DOI

ICRA Conference 2025 Conference Paper

BoxMap: Efficient Structural Mapping and Navigation

Zili Wang
Christopher Allum
Sean B. Andersson
Roberto Tron

While humans can successfully navigate using abstractions, ignoring details that are irrelevant to the task at hand, most of the existing approaches in robotics require detailed environment representations which consume a significant amount of sensing, computing, and storage; these issues become particularly important in resource-constrained settings with limited power budgets. Deep learning methods can learn from prior experience to abstract knowledge from novel environments, and use it to more efficiently execute tasks such as frontier exploration, object search, or scene understanding. We propose BoxMap, a Detection-Transformer-based architecture that takes advantage of the structure of the sensed partial environment to update a topological graph of the environment as a set of semantic entities (rooms and doors) and their relations (connectivity). The predictions from low-level measurements can be leveraged to achieve high-level goals with lower computational costs than methods based on detailed representations. As an example application, we consider a robot equipped with a 2-D laser scanner tasked with exploring a residential building. Our BoxMap representation scales quadratically with the number of rooms (with a small constant), resulting in significant savings over a full geometric map. Moreover, our high-level topological representation results in 30. 9 % shorter trajectories in the exploration task with respect to a standard method. Code is available at: bit. ly/3F6w2Yl.

Details

EAAI Journal 2025 Journal Article

Diameter-adjustable mandrel for thin-wall tube bending and its domain knowledge-integrated optimization design framework

Zili Wang
Jie Li
Xiaojian Liu
Shuyou Zhang
Yaochen Lin
Jianrong Tan

In response to the growing demand for small-batch bending tube production, traditional bending dies require separate customization for each tube size, resulting in extended design cycles and high costs. To meet bending requirements for tubes of different diameters using a single mandrel, a novel adjustable diameter mechanism (DAM) and its optimization design method are proposed. Initially, the DAM based on a planetary bevel gear-screw transmission set is developed for bending tubes of varying diameters. Subsequently, a domain knowledge-integrated optimization design framework is introduced. To reduce the cost of acquiring training samples for training surrogate models, a monotonicity-constrained neural network based on cascade boosting architecture (CB-MCNN) is introduced that enhances prediction accuracy while maintaining monotonicity. To improve the optimization speed and quality of Evolutionary Algorithms (EAs), a domain knowledge-guided EA (DK-EA) method is proposed, incorporating domain knowledge into the population initialization phase. The results indicate that: (1) CB-MCNN outperforms traditional methods and shows excellent performance on small-sample datasets. (2) DK-EA accelerates optimization processes and produces better outcomes. As a result, the domain knowledge-integrated optimization design framework enables the DAM to achieve a wider diameter variation range and enhanced reliability. The optimized DAM demonstrates the capability to bend tubes with diameters of 46–60 mm.

Details DOI

ICLR Conference 2025 Conference Paper

Layerwise Recurrent Router for Mixture-of-Experts

Zihan Qiu
Zeyu Huang
Shuang Cheng
Yizhi Zhou
Zili Wang
Ivan Titov 0001
Jie Fu 0001

The scaling of large language models (LLMs) has revolutionized their capabilities in various tasks, yet this growth must be matched with efficient computational strategies. The Mixture-of-Experts (MoE) architecture stands out for its ability to scale model size without significantly increasing training costs. Despite their advantages, current MoE models often display parameter inefficiency. For instance, a pre-trained MoE-based LLM with 52 billion parameters might perform comparably to a standard model with 6.7 billion. Being a crucial part of MoE, current routers in different layers independently assign tokens without leveraging historical routing information, potentially leading to suboptimal token-expert combinations and the parameter inefficiency problem. To alleviate this issue, we introduce the Layerwise Recurrent Router for Mixture-of-Experts (RMoE). RMoE leverages a Gated Recurrent Unit (GRU) to establish dependencies between routing decisions across consecutive layers. Such layerwise recurrence can be efficiently parallelly computed for input tokens and introduces negotiable costs. Our extensive empirical evaluations demonstrate that RMoE-based language models consistently outperform a spectrum of baseline models. Furthermore, RMoE integrates a novel computation stage orthogonal to existing methods, allowing seamless compatibility with other MoE architectures. Our analyses attribute RMoE's gains to its effective cross-layer information sharing, which also improves expert selection and diversity.

Details

NeurIPS Conference 2025 Conference Paper

MVU-Eval: Towards Multi-Video Understanding Evaluation for Multimodal LLMs

Tianhao Peng
Haochen Wang
Yuanxing Zhang
Noah Wang
Zili Wang
Ge Zhang
Jian Yang
Shihao Li

The advent of Multimodal Large Language Models (MLLMs) has expanded AI capabilities to visual modalities, yet existing evaluation benchmarks remain limited to single-video understanding, overlooking the critical need for multi-video understanding in real-world scenarios (e. g. , sports analytics and autonomous driving). To address this significant gap, we introduce MVU-Eval, the first comprehensive benchmark for evaluating M ulti- V ideo U nderstanding for MLLMs. Specifically, our MVU-Eval mainly assesses eight core competencies through 1, 824 meticulously curated question-answer pairs spanning 4, 959 videos from diverse domains, addressing both fundamental perception tasks and high-order reasoning tasks. These capabilities are rigorously aligned with real-world applications such as multi-sensor synthesis in autonomous systems and cross-angle sports analytics. Through extensive evaluation of state-of-the-art open-source and closed-source models, we reveal significant performance discrepancies and limitations in current MLLMs' ability to perform understanding across multiple videos. The benchmark will be made publicly available to foster future research.

PDF Details

ICLR Conference 2025 Conference Paper

Post-hoc Reward Calibration: A Case Study on Length Bias

Zeyu Huang
Zihan Qiu
Zili Wang
Edoardo M. Ponti
Ivan Titov 0001

Reinforcement Learning from Human Feedback aligns the outputs of Large Language Models with human values and preferences. Central to this process is the reward model (RM), which translates human feedback into training signals for optimising LLM behaviour. However, RMs can develop biases by exploiting spurious correlations in their training data, such as favouring outputs based on length or style rather than true quality. These biases can lead to incorrect output rankings, sub-optimal model evaluations, and the amplification of undesirable behaviours in LLMs alignment. This paper addresses the challenge of correcting such biases without additional data and training, introducing the concept of Post-hoc Reward Calibration. We first propose to use local average reward to estimate the bias term and, thus, remove it to approximate the underlying true reward. We then extend the approach to a more general and robust form with the Locally Weighted Regression. Focusing on the prevalent length bias, we validate our proposed approaches across three experimental settings, demonstrating consistent improvements: (1) a 3.11 average performance gain across 33 reward models on the RewardBench dataset; (2) improved agreement of RM produced rankings with GPT-4 evaluations and human preferences based on the AlpacaEval benchmark; and (3) improved Length-Controlled win rate (Dubois et al., 2024) of the RLHF process in multiple LLM–RM combinations. According to our experiments, our method is computationally efficient and generalisable to other types of bias and RMs, offering a scalable and robust solution for mitigating biases in LLM alignment and evaluation.

Details

NeurIPS Conference 2025 Conference Paper

Predictable Scale (Part II) --- Farseer: A Refined Scaling Law in LLMs

Houyi Li
Wenzhen Zheng
Qiufeng Wang
Zhenyu Ding
Haoying Wang
Zili Wang
Shijie Xuyang
Ning Ding

Training Large Language Models (LLMs) is prohibitively expensive, creating a critical scaling gap where insights from small-scale experiments often fail to transfer to resource-intensive production systems, thereby hindering efficient innovation. To bridge this, we introduce Farseer, a novel and refined scaling law offering enhanced predictive accuracy across scales. By systematically constructing a model loss surface $L(N, D)$, Farseer achieves a significantly better fit to empirical data than prior laws (e. g. , \Chinchilla's law). Our methodology yields accurate, robust, and highly generalizable predictions, demonstrating excellent extrapolation capabilities, outperforming Chinchilla's law, whose extrapolation error is 433\% higher. This allows for the reliable evaluation of competing training strategies across all $(N, D)$ settings, enabling conclusions from small-scale ablation studies to be confidently extrapolated to predict large-scale performance. Furthermore, Farseer provides new insights into optimal compute allocation, better reflecting the nuanced demands of modern LLM training. To validate our approach, we trained an extensive suite of approximately 1, 000 LLMs across diverse scales and configurations, consuming roughly 3 million NVIDIA H100 GPU hours. To foster further research, we are comprehensively open-sourcing all code, data, results (https: //github. com/Farseer-Scaling-Law/Farseer), all training logs (https: //wandb. ai/billzid/Farseer? nw=nwuserbillzid), all models used in scaling law fitting (https: //huggingface. co/Farseer-Scaling-Law).

PDF Details

NeurIPS Conference 2025 Conference Paper

SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines

Xeron Du
Yifan Yao
Kaijing Ma
Bingli Wang
Tianyu Zheng
Minghao Liu
Yiming Liang
Xiaolong Jin

Large language models (LLMs) have demonstrated remarkable proficiency in mainstream academic disciplines such as mathematics, physics, and computer science. However, human knowledge encompasses over 200 specialized disciplines, far exceeding the scope of existing benchmarks. The capabilities of LLMs in many of these specialized fields-particularly in light industry, agriculture, and service-oriented disciplines-remain inadequately evaluated. To address this gap, we present SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. Our benchmark employs a novel Human-LLM collaborative filtering mechanism to eliminate trivial or ambiguous questions through iterative refinement based on both LLM responses and expert feedback. Our experimental results reveal significant room for improvement in the performance of current state-of-the-art LLMs across diverse knowledge domains (e. g. , the reasoning-focused model Gemini-2. 5-Pro achieved the highest accuracy of 63. 56% on SuperGPQA), highlighting the considerable gap between current model capabilities and artificial general intelligence. Additionally, we present comprehensive insights from our management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.

PDF Details

EAAI Journal 2024 Journal Article

A transferred hybrid surrogate model integrating Gaussian membership virtual sample generation for small sample prediction: Applications in metal tube bending

Zili Wang
Yuchen Gu
Shuyou Zhang
Xiaojian Liu
Yaochen Lin
Jianrong Tan

The high-performance virtual sample generation (VSG) method has been extensively introduced to solve the problem of small sample sizes. Data distribution information is a key element of current VSG methods at the data-driven level. Herein, we propose an improved VSG method with a Gaussian distribution and explore the relationship between the Gaussian function and data expansion. To obtain more feasible virtual samples, information expanded based on the Gaussian membership function (GMIE) was established. For further improvement, a hybrid surrogate model based on transfer (THSM) is proposed, which differs from the general hybrid surrogate model (HSM) methods that only mix single models. Using a prevailing evaluation method, our proposed method, which combines a Gaussian membership function with a hybrid surrogate model, outperforms other competing approaches in 12 numerical cases owing to its feasibility and efficacy. Additionally, the proposed approach is applied to a metal tube rotary draw bending (RDB) prediction problem to illustrate its ability to support complex engineering designs.

Details DOI

ICLR Conference 2024 Conference Paper

MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training

Yizhi Li
Ruibin Yuan
Ge Zhang 0009
Yinghao Ma
Xingran Chen
Hanzhi Yin
Chenghao Xiao
Chenghua Lin 0002

Self-supervised learning (SSL) has recently emerged as a promising paradigm for training generalisable models on large-scale data in the fields of vision, text, and speech. Although SSL has been proven effective in speech and audio, its application to music audio has yet to be thoroughly explored. This is partially due to the distinctive challenges associated with modelling musical knowledge, particularly tonal and pitched characteristics of music. To address this research gap, we propose an acoustic **M**usic und**ER**standing model with large-scale self-supervised **T**raining (**MERT**), which incorporates teacher models to provide pseudo labels in the masked language modelling (MLM) style acoustic pre-training. In our exploration, we identified an effective combination of teacher models, which outperforms conventional speech and audio approaches in terms of performance. This combination includes an acoustic teacher based on Residual Vector Quantization - Variational AutoEncoder (RVQ-VAE) and a musical teacher based on the Constant-Q Transform (CQT). Furthermore, we explore a wide range of settings to overcome the instability in acoustic language model pre-training, which allows our designed paradigm to scale from 95M to 330M parameters. Experimental results indicate that our model can generalise and perform well on 14 music understanding tasks and attain state-of-the-art (SOTA) overall scores.

Details

ICLR Conference 2024 Conference Paper

MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework

Sirui Hong
Mingchen Zhuge
Jonathan Chen
Xiawu Zheng
Yuheng Cheng
Jinlin Wang
Ceyao Zhang
Zili Wang

Recently, remarkable progress has been made on automated problem solving through societies of agents based on large language models (LLMs). Previous LLM-based multi-agent systems can already solve simple dialogue tasks. More complex tasks, however, face challenges through logic inconsistencies due to cascading hallucinations caused by naively chaining LLMs. Here we introduce MetaGPT, an innovative meta-programming framework incorporating efficient human workflows into LLM-based multi-agent collaborations. MetaGPT encodes Standardized Operating Procedures (SOPs) into prompt sequences for more streamlined workflows, thus allowing agents with human-like domain expertise to verify intermediate results and reduce errors. MetaGPT utilizes an assembly line paradigm to assign diverse roles to various agents, efficiently breaking down complex tasks into subtasks involving many agents working together. On collaborative software engineering benchmarks, MetaGPT generates more coherent solutions than previous chat-based multi-agent systems.

Details

AAAI Conference 2024 Conference Paper

QuerySum: A Multi-Document Query-Focused Summarization Dataset Augmented with Similar Query Clusters

Yushan Liu
Zili Wang
Ruifeng Yuan

Query-focused summarization (QFS) aims to summarize the source document(s) with regard to a specific aspect of information given in a query. It plays an important role in presenting users with a concise answer summary from a set of query-relevant documents retrieved by the information retrieval system. Nonetheless, the QFS research has long been hampered by the lack of adequate datasets in terms of both quality and quantity. In this paper, we introduce a large-scale multi-document query-focused summarization dataset, called QuerySum, which contains 27,041 data samples covering diverse topics and its quality is guaranteed through human verification. Unlike some previous QFS datasets constructed directly from the question answering datasets, 74% queries in our dataset are the challenging non-factoid What-, Why-, and How- questions. More importantly, we also provide a set of similar queries together with the corresponding summaries pairs for each query as the retrieved context, presenting a new feature of QuerySum. We aim to encourage research efforts in query intention understanding in the context of QFS. Leveraging QuerySum's depth, we propose a model for query-aware multi-document summarization and set a new QFS benchmark.

PDF Details DOI

AAAI Conference 2024 Conference Paper

Xiezhi: An Ever-Updating Benchmark for Holistic Domain Knowledge Evaluation

Zhouhong Gu
Xiaoxuan Zhu
Haoning Ye
Lin Zhang
Jianchen Wang
Yixin Zhu
Sihang Jiang
Zhuozhi Xiong

New Natural Langauge Process~(NLP) benchmarks are urgently needed to align with the rapid development of large language models (LLMs). We present Xiezhi, the most comprehensive evaluation suite designed to assess holistic domain knowledge.Xiezhi comprises multiple-choice questions across 516 diverse disciplines ranging from 13 different subjects with 249,587 questions and accompanied by Xiezhi-Specialty with 14,041 questions and Xiezhi-Interdiscipline with 10,746 questions. We conduct evaluation of the 47 cutting-edge LLMs on Xiezhi. Results indicate that LLMs exceed average performance of humans in science, engineering, agronomy, medicine, and art, but fall short in economics, jurisprudence, pedagogy, literature, history, and management. All the evaluation code and data are open sourced in https://github.com/MikeGu721/XiezhiBenchmark

PDF Details DOI

IROS Conference 2023 Conference Paper

Do More with Less: Single-Model, Multi-Goal Architectures for Resource-Constrained Robots

Zili Wang
Drew Threatt
Sean B. Andersson
Roberto Tron

Deep learning methods are widely used in robotic applications. By learning from prior experience, the robot can abstract knowledge of the environment, and use this knowledge to accomplish different goals, such as object search, frontier exploration, or scene understanding, with a smaller amount of resources than might be needed without that knowledge. Most existing methods typically require a significant amount of sensing, which in turn has significant costs in terms of power consumption for acquisition and processing, and typically focus on models that are tuned for each specific goal, leading to the need to train, store and run each one separately. These issues are particularly important in a resource-constrained setting, such as with small-scale robots or during long-duration missions. We propose a single, multi-task deep learning architecture that takes advantage of the structure of the partial environment to predict different abstractions of the environment (thus reducing the need for rich sensing), and to leverage these predictions to simultaneously achieve different high-level goals (thus sharing computation between goals). As an example application of the proposed architecture, we consider the specific example of a robot equipped with a 2-D laser scanner and an object detector, tasked with searching for an object (such as an exit) in a residential building while constructing a topological map that can be used for future missions. The prior knowledge of the environment is encoded using a U-Net deep network architecture. In this context, our work leads to an object search algorithm that is complete, and that outperforms a more traditional frontier-based approach. The topological map we produce uses scene trees to qualitatively represent the environment as a graph at a fraction of the cost of existing SLAM-based solutions. Our results demonstrate that it is possible to extract multi-task semantic information that is useful for navigation and mapping directly from bare-bone, non-semantic measurements.

Details

AAAI Conference 2023 Conference Paper

Preserve Context Information for Extract-Generate Long-Input Summarization Framework

Ruifeng Yuan
Zili Wang
Ziqiang Cao
Wenjie Li

The Extract-generate framework has been a classic approach for text summarization. As pretrained language models struggling with long-input summarization for their high memory cost, extract-generate framework regains researchers' interests. However, the cost of its effectiveness in dealing with long-input summarization is the loss of context information. In this paper, we present a context-aware extract-generate framework (CAEG) for long-input text summarization. It focuses on preserving both local and global context information in an extract-generate framework with little cost, and can be applied to most of existing extract-generate summarization models. CAEG generates a set of context-related text spans called context prompts for each text snippet and use them to transfer the context information from the extractor and generator. To find such context prompts, we propose to capture the context information based on the interpretation of the extractor, where the text spans having the highest contribution to the extraction decision are considered as containing the richest context information. We evaluate our approach on both long-document and long-dialogue summarization datasets: arXiv and QMSum. The experiment results show that CAEG achieves the-state-of-art result on QMSum and outperforms other extract-generate based models in arXiv.

PDF Details DOI

EAAI Journal 2022 Journal Article

Toward axial accuracy prediction and optimization of metal tube bending forming: A novel GRU-integrated Pb-NSGA-III optimization framework

Chang Sun
Zili Wang
Shuyou Zhang
Xiaojian Liu
Le Wang
Jianrong Tan

Springback is particularly common in metal tube bending, which extremely affects the metal tube axial accuracy. At present, the springback mechanism still remains unclear due to the complex plastic deformation characteristics of metal materials. It is difficult to obtain the accurate axial springback information before bending forming, not to mention formulating a reasonable processing plan to compensate springback. To alleviate this difficulty, a novel optimization framework is constructed which takes the radius changes series (RCS) as the new axial accuracy evaluation index for the first time. The optimization framework contains a GRU-based deep learning network as the prediction module to predict the springback more reliably. Subsequently, NSGA-III has been improved by the proposed guiding factor (GF) and dynamic reference points (DRF) algorithm, i. e, priority-based NSGA-III (Pb-NSGA-III), which can more efficiently deal with the objectives with different priorities when generating the processing plan. With the help of the finite element (FE) and bending experiment, the springback dataset construction and the accuracy verification can be achieved. The results show that the framework can achieve high-precision and robust prediction of the tube axial accuracy. Compared with the existing commonly used multi-objective algorithms, the proposed Pb-NSGA-III shows its superiority in engineering application.

Details DOI