Wei Xu

Lingfeng Sun, Haichao Zhang, Wei Xu et al. · berkeley

The purpose of multi-task reinforcement learning (MTRL) is to train a single policy that can be applied to a set of different tasks. Sharing parameters allows us to take advantage of the similarities among tasks. However, the gaps between contents and difficulties of different tasks bring us challenges on both which tasks should share the parameters and what parameters should be shared, as well as the optimization challenges due to parameter sharing. In this work, we introduce a parameter-compositional approach (PaCo) as an attempt to address these challenges. In this framework, a policy subspace represented by a set of parameters is learned. Policies for all the single tasks lie in this subspace and can be composed by interpolating with the learned set. It allows not only flexible parameter sharing but also a natural way to improve training. We demonstrate the state-of-the-art performance on Meta-World benchmarks, verifying the effectiveness of the proposed approach.

Lingfeng Sun, Haichao Zhang, Wei Xu et al. · berkeley

In this work, we investigate the potential of improving multi-task training and also leveraging it for transferring in the reinforcement learning setting. We identify several challenges towards this goal and propose a transferring approach with a parameter-compositional formulation. We investigate ways to improve the training of multi-task reinforcement learning which serves as the foundation for transferring. Then we conduct a number of transferring experiments on various manipulation tasks. Experimental results demonstrate that the proposed approach can have improved performance in the multi-task training stage, and further show effective transferring in terms of both sample efficiency and performance.

Dingkang Liang, Jiahao Xie, Zhikang Zou et al.

Supervised crowd counting relies heavily on costly manual labeling, which is difficult and expensive, especially in dense scenes. To alleviate the problem, we propose a novel unsupervised framework for crowd counting, named CrowdCLIP. The core idea is built on two observations: 1) the recent contrastive pre-trained vision-language model (CLIP) has presented impressive performance on various downstream tasks; 2) there is a natural mapping between crowd patches and count text. To the best of our knowledge, CrowdCLIP is the first to investigate the vision language knowledge to solve the counting problem. Specifically, in the training stage, we exploit the multi-modal ranking loss by constructing ranking text prompts to match the size-sorted crowd patches to guide the image encoder learning. In the testing stage, to deal with the diversity of image patches, we propose a simple yet effective progressive filtering strategy to first select the highly potential crowd patches and then map them into the language space with various counting intervals. Extensive experiments on five challenging datasets demonstrate that the proposed CrowdCLIP achieves superior performance compared to previous unsupervised state-of-the-art counting methods. Notably, CrowdCLIP even surpasses some popular fully-supervised methods under the cross-dataset setting. The source code will be available at https://github.com/dk-liang/CrowdCLIP.

Yining Jiang, Yunxin Xu, Wenyun Xu et al.

Result reproduction of computer networking research is challenging as the scarcity of open-source implementations and the complexity of heterogeneous system architectures. Even though Large Language Models have demonstrated potential in code generation, existing code generation frameworks often fail to address the long-context constraints and intricate logical dependencies, which are vital in reproducing network systems from academic papers. Thus, we introduce RepLLM, an end-to-end multi-agent framework designed to automate code reproduction from paper content. RepLLM features a collaborative architecture comprising four specialized agents -- Content Parsing, Architecture Design, Code Generation, and Audit&Repair, which are coordinated through Shared Memory mechanism to ensure global context consistency. With the enhancement of Structured Chain-of-Thought LLM reasoning and a sandbox-isolated static-dynamic debugging methodology, our framework effectively resolves semantic discrepancies and runtime errors, thereby improving reliable reproductions. Extensive evaluations on representative papers in top conferences demonstrate that RepLLM outperforms state-of-the-art system-level LLM frameworks in generating compile-ready and logically correct systems. Our results show that, with the aid of RepLLM, we can reproduce 95% of the original benchmarks within approximately two hours while reducing token consumption by up to 10% compared with state-of-the-art baselines.

Renjie Xie, Wei Xu, Jiabao Yu et al.

Deep learning (DL) applied to a device's radio-frequency fingerprint~(RFF) has attracted significant attention in physical-layer authentication due to its extraordinary classification performance. Conventional DL-RFF techniques are trained by adopting maximum likelihood estimation~(MLE). Although their discriminability has recently been extended to unknown devices in open-set scenarios, they still tend to overfit the channel statistics embedded in the training dataset. This restricts their practical applications as it is challenging to collect sufficient training data capturing the characteristics of all possible wireless channel environments. To address this challenge, we propose a DL framework of disentangled representation~(DR) learning that first learns to factor the signals into a device-relevant component and a device-irrelevant component via adversarial learning. Then, it shuffles these two parts within a dataset for implicit data augmentation, which imposes a strong regularization on RFF extractor learning to avoid the possible overfitting of device-irrelevant channel statistics, without collecting additional data from unknown channels. Experiments validate that the proposed approach, referred to as DR-based RFF, outperforms conventional methods in terms of generalizability to unknown devices even under unknown complicated propagation environments, e.g., dispersive multipath fading channels, even though all the training data are collected in a simple environment with dominated direct line-of-sight~(LoS) propagation paths.

Jiahao Xie, Wei Xu, Dingkang Liang et al.

Crowd counting is a challenging task due to the heavy occlusions, scales, and density variations. Existing methods handle these challenges effectively while ignoring low-resolution (LR) circumstances. The LR circumstances weaken the counting performance deeply for two crucial reasons: 1) limited detail information; 2) overlapping head regions accumulate in density maps and result in extreme ground-truth values. An intuitive solution is to employ super-resolution (SR) pre-processes for the input LR images. However, it complicates the inference steps and thus limits application potentials when requiring real-time. We propose a more elegant method termed Multi-Scale Super-Resolution Module (MSSRM). It guides the network to estimate the lost de tails and enhances the detailed information in the feature space. Noteworthy that the MSSRM is plug-in plug-out and deals with the LR problems with no inference cost. As the proposed method requires SR labels, we further propose a Super-Resolution Crowd Counting dataset (SR-Crowd). Extensive experiments on three datasets demonstrate the superiority of our method. The code will be available at https://github.com/PRIS-CV/MSSRM.git.

Yihan Hu, Siqi Chai, Zhening Yang et al.

As autonomous driving systems being deployed to millions of vehicles, there is a pressing need of improving the system's scalability, safety and reducing the engineering cost. A realistic, scalable, and practical simulator of the driving world is highly desired. In this paper, we present an efficient solution based on generative models which learns the dynamics of the driving scenes. With this model, we can not only simulate the diverse futures of a given driving scenario but also generate a variety of driving scenarios conditioned on various prompts. Our innovative design allows the model to operate in both full-Autoregressive and partial-Autoregressive modes, significantly improving inference and training speed without sacrificing generative capability. This efficiency makes it ideal for being used as an online reactive environment for reinforcement learning, an evaluator for planning policies, and a high-fidelity simulator for testing. We evaluated our model against two real-world datasets: the Waymo motion dataset and the nuPlan dataset. On the simulation realism and scene generation benchmark, our model achieves the state-of-the-art performance. And in the planning benchmarks, our planner outperforms the prior arts. We conclude that the proposed generative model may serve as a foundation for a variety of motion planning tasks, including data generation, simulation, planning, and online training. Source code is public at https://github.com/HorizonRobotics/GUMP/

Yurong Qian, Jindan Xu, Shuhan Zhu et al.

In this paper, we consider a multiuser mobile edge computing (MEC) system, where a mixed-integer offloading strategy is used to assist the resource assignment for task offloading. Although the conventional branch and bound (BnB) approach can be applied to solve this problem, a huge burden of computational complexity arises which limits the application of BnB. To address this issue, we propose an intelligent BnB (IBnB) approach which applies deep learning (DL) to learn the pruning strategy of the BnB approach. By using this learning scheme, the structure of the BnB approach ensures near-optimal performance and meanwhile DL-based pruning strategy significantly reduces the complexity. Numerical results verify that the proposed IBnB approach achieves optimal performance with complexity reduced by over 80%.

Hao Fu, Guotai Wang, Wenhui Lei et al.

Accurate segmentation of Anatomical brain Barriers to Cancer spread (ABCs) plays an important role for automatic delineation of Clinical Target Volume (CTV) of brain tumors in radiotherapy. Despite that variants of U-Net are state-of-the-art segmentation models, they have limited performance when dealing with ABCs structures with various shapes and sizes, especially thin structures (e.g., the falx cerebri) that span only few slices. To deal with this problem, we propose a High and Multi-Resolution Network (HMRNet) that consists of a multi-scale feature learning branch and a high-resolution branch, which can maintain the high-resolution contextual information and extract more robust representations of anatomical structures with various scales. We further design a Bidirectional Feature Calibration (BFC) block to enable the two branches to generate spatial attention maps for mutual feature calibration. Considering the different sizes and positions of ABCs structures, our network was applied after a rough localization of each structure to obtain fine segmentation results. Experiments on the MICCAI 2020 ABCs challenge dataset showed that: 1) Our proposed two-stage segmentation strategy largely outperformed methods segmenting all the structures in just one stage; 2) The proposed HMRNet with two branches can maintain high-resolution representations and is effective to improve the performance on thin structures; 3) The proposed BFC block outperformed existing attention methods using monodirectional feature calibration. Our method won the second place of ABCs 2020 challenge and has a potential for more accurate and reasonable delineation of CTV of brain tumors.

Huahui Yi, Ziyuan Qin, Qicheng Lao et al.

Inevitable domain and task discrepancies in real-world scenarios can impair the generalization performance of the pre-trained deep models for medical data. Therefore, we audaciously propose that we should build a general-purpose medical AI system that can be seamlessly adapted to downstream domains/tasks. Since the domain/task adaption procedures usually involve additional labeling work for the target data, designing a data-efficient adaption algorithm is desired to save the cost of transferring the learned knowledge. Our recent work found that vision-language models (VLMs) are efficient learners with extraordinary cross-domain ability. Therefore, in this work, we further explore the possibility of leveraging pre-trained VLMs as medical foundation models for building general-purpose medical AI, where we thoroughly investigate three machine-learning paradigms, i.e., domain/task-specialized learning, joint learning, and continual learning, for training the VLMs and evaluate their generalization performance on cross-domain and cross-task test sets. To alleviate the catastrophic forgetting during sequential training, we employ rehearsal learning and receive a sharp boost in terms of generalization capability. In a nutshell, our empirical evidence suggests that continual learning may be a practical and efficient learning paradigm for the medical foundation model. And we hope researchers can use our empirical evidence as basement to further explore the path toward medical foundation model.

Yao Tang, Guangxu Zhu, Wei Xu et al.

Federated edge learning (FEEL) enables privacy-preserving model training through periodic communication between edge devices and the server. Unmanned Aerial Vehicle (UAV)-mounted edge devices are particularly advantageous for FEEL due to their flexibility and mobility in efficient data collection. In UAV-assisted FEEL, sensing, computation, and communication are coupled and compete for limited onboard resources, and UAV deployment also affects sensing and communication performance. Therefore, the joint design of UAV deployment and resource allocation is crucial to achieving the optimal training performance. In this paper, we address the problem of joint UAV deployment design and resource allocation for FEEL via a concrete case study of human motion recognition based on wireless sensing. We first analyze the impact of UAV deployment on the sensing quality and identify a threshold value for the sensing elevation angle that guarantees a satisfactory quality of data samples. Due to the non-ideal sensing channels, we consider the probabilistic sensing model, where the successful sensing probability of each UAV is determined by its position. Then, we derive the upper bound of the FEEL training loss as a function of the sensing probability. Theoretical results suggest that the convergence rate can be improved if UAVs have a uniform successful sensing probability. Based on this analysis, we formulate a training time minimization problem by jointly optimizing UAV deployment, integrated sensing, computation, and communication (ISCC) resources under a desirable optimality gap constraint. To solve this challenging mixed-integer non-convex problem, we apply the alternating optimization technique, and propose the bandwidth, batch size, and position optimization (BBPO) scheme to optimize these three decision variables alternately.

Jiacheng Wang, Ping Liu, Jingen Liu et al.

Virtual try-on of eyeglasses involves placing eyeglasses of different shapes and styles onto a face image without physically trying them on. While existing methods have shown impressive results, the variety of eyeglasses styles is limited and the interactions are not always intuitive or efficient. To address these limitations, we propose a Text-guided Eyeglasses Manipulation method that allows for control of the eyeglasses shape and style based on a binary mask and text, respectively. Specifically, we introduce a mask encoder to extract mask conditions and a modulation module that enables simultaneous injection of text and mask conditions. This design allows for fine-grained control of the eyeglasses' appearance based on both textual descriptions and spatial constraints. Our approach includes a disentangled mapper and a decoupling strategy that preserves irrelevant areas, resulting in better local editing. We employ a two-stage training scheme to handle the different convergence speeds of the various modality conditions, successfully controlling both the shape and style of eyeglasses. Extensive comparison experiments and ablation analyses demonstrate the effectiveness of our approach in achieving diverse eyeglasses styles while preserving irrelevant areas.

Jiayuan Chen, Xiang Zhang, Yinfei Xu et al.

Recently, graph-based models designed for downstream tasks have significantly advanced research on graph neural networks (GNNs). GNN baselines based on neural message-passing mechanisms such as GCN and GAT perform worse as the network deepens. Therefore, numerous GNN variants have been proposed to tackle this performance degradation problem, including many deep GNNs. However, a unified framework is still lacking to connect these existing models and interpret their effectiveness at a high level. In this work, we focus on deep GNNs and propose a novel view for understanding them. We establish a theoretical framework via inference on a probabilistic graphical model. Given the fixed point equation (FPE) derived from the variational inference on the Markov random fields, the deep GNNs, including JKNet, GCNII, DGCN, and the classical GNNs, such as GCN, GAT, and APPNP, can be regarded as different approximations of the FPE. Moreover, given this framework, more accurate approximations of FPE are brought, guiding us to design a more powerful GNN: coupling graph neural network (CoGNet). Extensive experiments are carried out on citation networks and natural language processing downstream tasks. The results demonstrate that the CoGNet outperforms the SOTA models.

Zhenyu Tao, Wei Xu, Xiaohu You

The proliferation of diverse wireless services in 5G and beyond has led to the emergence of network slicing technologies. Among these, admission control plays a crucial role in achieving service-oriented optimization goals through the selective acceptance of service requests. Although deep reinforcement learning (DRL) forms the foundation in many admission control approaches thanks to its effectiveness and flexibility, initial instability with excessive convergence delay of DRL models hinders their deployment in real-world networks. We propose a digital twin (DT) accelerated DRL solution to address this issue. Specifically, we first formulate the admission decision-making process as a semi-Markov decision process, which is subsequently simplified into an equivalent discrete-time Markov decision process to facilitate the implementation of DRL methods. A neural network-based DT is established with a customized output layer for queuing systems, trained through supervised learning, and then employed to assist the training phase of the DRL model. Extensive simulations show that the DT-accelerated DRL improves resource utilization by over 40% compared to the directly trained state-of-the-art dueling deep Q-learning model. This improvement is achieved while preserving the model's capability to optimize the long-term rewards of the admission process.

Ziqing Yin, Renjie Xie, Wei Xu et al.

Deep learning (DL)-based channel state information (CSI) feedback methods compressed the CSI matrix by exploiting its delay and angle features straightforwardly, while the measure in terms of information contained in the CSI matrix has rarely been considered. Based on this observation, we introduce self-information as an informative CSI representation from the perspective of information theory, which reflects the amount of information of the original CSI matrix in an explicit way. Then, a novel DL-based network is proposed for temporal CSI compression in the self-information domain, namely SD-CsiNet. The proposed SD-CsiNet projects the raw CSI onto a self-information matrix in the newly-defined self-information domain, extracts both temporal and spatial features of the self-information matrix, and then couples these two features for effective compression. Experimental results verify the effectiveness of the proposed SD-CsiNet by exploiting the self-information of CSI. Particularly for compression ratios 1/8 and 1/16, the SD-CsiNet respectively achieves 7.17 dB and 3.68 dB performance gains compared to state-of-the-art methods.

Wei Xu

The field of user experience (UX) based on the design philosophy of "user-centered design" is moving towards the intelligence era. Still, the existing UX paradigm mainly aims at non-intelligent systems and lacks a systematic approach to UX for intelligent systems. Throughout the development of UX, the UX paradigm shows the evolution characteristics of the cross-technology era. At present, the intelligence era has put forward new demands on the UX paradigm. For this reason, this paper proposes a "UX 3.0" paradigm framework and the corresponding UX methodology system in the intelligence era. The "UX 3.0" paradigm framework includes five categories of UX methods: ecological experience, innovation-enabled experience, AI-enabled experience, human-AI interaction-based experience, and human-AI collaboration-based experience methods, each providing corresponding multiple UX paradigmatic orientations. The proposal of the "UX 3.0" paradigm helps improve the existing UX methods and provides methodological support for the research and applications of UX in developing intelligent systems. Finally, this paper looks forward to future research and applications of the "UX 3.0" paradigm.

Han Zhang, Yuqi Li, Shun Zheng et al.

Accurately predicting the lifetime of battery cells in early cycles holds tremendous value for battery research and development as well as numerous downstream applications. This task is rather challenging because diverse conditions, such as electrode materials, operating conditions, and working environments, collectively determine complex capacity-degradation behaviors. However, current prediction methods are developed and validated under limited aging conditions, resulting in questionable adaptability to varied aging conditions and an inability to fully benefit from historical data collected under different conditions. Here we introduce a universal deep learning approach that is capable of accommodating various aging conditions and facilitating effective learning under low-resource conditions by leveraging data from rich conditions. Our key finding is that incorporating inter-cell feature differences, rather than solely considering single-cell characteristics, significantly increases the accuracy of battery lifetime prediction and its cross-condition robustness. Accordingly, we develop a holistic learning framework accommodating both single-cell and inter-cell modeling. A comprehensive benchmark is built for evaluation, encompassing 401 battery cells utilizing 5 prevalent electrode materials across 168 cycling conditions. We demonstrate remarkable capabilities in learning across diverse aging conditions, exclusively achieving 10% prediction error using the first 100 cycles, and in facilitating low-resource learning, almost halving the error of single-cell modeling in many cases. More broadly, by breaking the learning boundaries among different aging conditions, our approach could significantly accelerate the development and optimization of lithium-ion batteries.

Zhenyu Tao, Wei Xu, Xiaohu You

User mobility modeling serves a crucial role in analysis and optimization of contemporary wireless networks. Typical stochastic mobility models, e.g., random waypoint model and Gauss Markov model, can hardly capture the distribution characteristics of users within real-world areas. State-of-the-art trace-based mobility models and existing learning-based trajectory generation methods, however, are frequently constrained by the inaccessibility of substantial real trajectories due to privacy concerns. In this paper, we harness the intrinsic correlation between street maps and trajectories and develop a novel zero-shot trajectory generation method, named Map2Traj, by exploiting the diffusion model. We incorporate street maps as a condition to consistently pilot the denoising process and train our model on diverse sets of real trajectories from various regions in Xi'an, China, and their corresponding street maps. With solely the street map of an unobserved area, Map2Traj generates synthetic trajectories that not only closely resemble the real-world mobility pattern but also offer comparable efficacy. Extensive experiments validate the efficacy of our proposed method on zero-shot trajectory generation tasks in terms of both trajectory and distribution similarities. In addition, a case study of employing Map2Traj in wireless network optimization is presented to validate its efficacy for downstream applications.

Lixiang Xu, Qingzhe Cui, Richang Hong et al.

In recent years, the results of view-based 3D shape recognition methods have saturated, and models with excellent performance cannot be deployed on memory-limited devices due to their huge size of parameters. To address this problem, we introduce a compression method based on knowledge distillation for this field, which largely reduces the number of parameters while preserving model performance as much as possible. Specifically, to enhance the capabilities of smaller models, we design a high-performing large model called Group Multi-view Vision Transformer (GMViT). In GMViT, the view-level ViT first establishes relationships between view-level features. Additionally, to capture deeper features, we employ the grouping module to enhance view-level features into group-level features. Finally, the group-level ViT aggregates group-level features into complete, well-formed 3D shape descriptors. Notably, in both ViTs, we introduce spatial encoding of camera coordinates as innovative position embeddings. Furthermore, we propose two compressed versions based on GMViT, namely GMViT-simple and GMViT-mini. To enhance the training effectiveness of the small models, we introduce a knowledge distillation method throughout the GMViT process, where the key outputs of each GMViT component serve as distillation targets. Extensive experiments demonstrate the efficacy of the proposed method. The large model GMViT achieves excellent 3D classification and retrieval results on the benchmark datasets ModelNet, ShapeNetCore55, and MCB. The smaller models, GMViT-simple and GMViT-mini, reduce the parameter size by 8 and 17.6 times, respectively, and improve shape recognition speed by 1.5 times on average, while preserving at least 90% of the classification and retrieval performance. The code is available at https://github.com/bigdata-graph/GMViT.

Han Zhang, Zihao Zhang, Wenhao Zheng et al.

A major challenge in image segmentation is classifying object boundaries. Recent efforts propose to refine the segmentation result with boundary masks. However, models are still prone to misclassifying boundary pixels even when they correctly capture the object contours. In such cases, even a perfect boundary map is unhelpful for segmentation refinement. In this paper, we argue that assigning proper prior weights to error-prone pixels such as object boundaries can significantly improve the segmentation quality. Specifically, we present the \textit{pixel null model} (PNM), a prior model that weights each pixel according to its probability of being correctly classified by a random segmenter. Empirical analysis shows that PNM captures the misclassification distribution of different state-of-the-art (SOTA) segmenters. Extensive experiments on semantic, instance, and panoptic segmentation tasks over three datasets (Cityscapes, ADE20K, MS COCO) confirm that PNM consistently improves the segmentation quality of most SOTA methods (including the vision transformers) and outperforms boundary-based methods by a large margin. We also observe that the widely-used mean IoU (mIoU) metric is insensitive to boundaries of different sharpness. As a byproduct, we propose a new metric, \textit{PNM IoU}, which perceives the boundary sharpness and better reflects the model segmentation performance in error-prone regions.

Hang Fan, Yu Shi, Zongliang Fu et al.

High-quality wind power forecasting is crucial for the operation of modern power grids. However, prevailing data-driven paradigms either train a site-specific model which cannot generalize to other locations or rely on fine-tuning of general-purpose time series foundation models which are difficult to incorporate domain-specific data in the energy sector. This paper introduces WindFM, a lightweight and generative Foundation Model designed specifically for probabilistic wind power forecasting. WindFM employs a discretize-and-generate framework. A specialized time-series tokenizer first converts continuous multivariate observations into discrete, hierarchical tokens. Subsequently, a decoder-only Transformer learns a universal representation of wind generation dynamics by autoregressively pre-training on these token sequences. Using the comprehensive WIND Toolkit dataset comprising approximately 150 billion time steps from more than 126,000 sites, WindFM develops a foundational understanding of the complex interplay between atmospheric conditions and power output. Extensive experiments demonstrate that our compact 8.1M parameter model achieves state-of-the-art zero-shot performance on both deterministic and probabilistic tasks, outperforming specialized models and larger foundation models without any fine-tuning. In particular, WindFM exhibits strong adaptiveness under out-of-distribution data from a different continent, demonstrating the robustness and transferability of its learned representations. Our pre-trained model is publicly available at https://github.com/shiyu-coder/WindFM.

Haonan Yu, Wei Xu, Haichao Zhang

We consider the safe reinforcement learning (RL) problem of maximizing utility with extremely low constraint violation rates. Assuming no prior knowledge or pre-training of the environment safety model given a task, an agent has to learn, via exploration, which states and actions are safe. A popular approach in this line of research is to combine a model-free RL algorithm with the Lagrangian method to adjust the weight of the constraint reward relative to the utility reward dynamically. It relies on a single policy to handle the conflict between utility and constraint rewards, which is often challenging. We present SEditor, a two-policy approach that learns a safety editor policy transforming potentially unsafe actions proposed by a utility maximizer policy into safe ones. The safety editor is trained to maximize the constraint reward while minimizing a hinge loss of the utility state-action values before and after an action is edited. SEditor extends existing safety layer designs that assume simplified safety models, to general safe RL scenarios where the safety model can in theory be arbitrarily complex. As a first-order method, it is easy to implement and efficient for both inference and training. On 12 Safety Gym tasks and 2 safe racing tasks, SEditor obtains much a higher overall safety-weighted-utility (SWU) score than the baselines, and demonstrates outstanding utility performance with constraint violation rates as low as once per 2k time steps, even in obstacle-dense environments. On some tasks, this low violation rate is up to 200 times lower than that of an unconstrained RL method with similar utility performance. Code is available at https://github.com/hnyu/seditor.

Yijia Shao, Yucheng Jiang, Theodore A. Kanell et al. · stanford

We study how to apply large language models to write grounded and organized long-form articles from scratch, with comparable breadth and depth to Wikipedia pages. This underexplored problem poses new challenges at the pre-writing stage, including how to research the topic and prepare an outline prior to writing. We propose STORM, a writing system for the Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking. STORM models the pre-writing stage by (1) discovering diverse perspectives in researching the given topic, (2) simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources, (3) curating the collected information to create an outline. For evaluation, we curate FreshWiki, a dataset of recent high-quality Wikipedia articles, and formulate outline assessments to evaluate the pre-writing stage. We further gather feedback from experienced Wikipedia editors. Compared to articles generated by an outline-driven retrieval-augmented baseline, more of STORM's articles are deemed to be organized (by a 25% absolute increase) and broad in coverage (by 10%). The expert feedback also helps identify new challenges for generating grounded long articles, such as source bias transfer and over-association of unrelated facts.

Wei Xu, Yi Wan

The attention mechanism has gained significant recognition in the field of computer vision due to its ability to effectively enhance the performance of deep neural networks. However, existing methods often struggle to effectively utilize spatial information or, if they do, they come at the cost of reducing channel dimensions or increasing the complexity of neural networks. In order to address these limitations, this paper introduces an Efficient Local Attention (ELA) method that achieves substantial performance improvements with a simple structure. By analyzing the limitations of the Coordinate Attention method, we identify the lack of generalization ability in Batch Normalization, the adverse effects of dimension reduction on channel attention, and the complexity of attention generation process. To overcome these challenges, we propose the incorporation of 1D convolution and Group Normalization feature enhancement techniques. This approach enables accurate localization of regions of interest by efficiently encoding two 1D positional feature maps without the need for dimension reduction, while allowing for a lightweight implementation. We carefully design three hyperparameters in ELA, resulting in four different versions: ELA-T, ELA-B, ELA-S, and ELA-L, to cater to the specific requirements of different visual tasks such as image classification, object detection and sementic segmentation. ELA can be seamlessly integrated into deep CNN networks such as ResNet, MobileNet, and DeepLab. Extensive evaluations on the ImageNet, MSCOCO, and Pascal VOC datasets demonstrate the superiority of the proposed ELA module over current state-of-the-art methods in all three aforementioned visual tasks.

Duong Minh Le, Yang Chen, Alan Ritter et al.

Zero-shot cross-lingual transfer utilizing multilingual LLMs has become a popular learning paradigm for low-resource languages with no labeled training data. However, for NLP tasks that involve fine-grained predictions on words and phrases, the performance of zero-shot cross-lingual transfer learning lags far behind supervised fine-tuning methods. Therefore, it is common to exploit translation and label projection to further improve the performance by (1) translating training data that is available in a high-resource language (e.g., English) together with the gold labels into low-resource languages, and/or (2) translating test data in low-resource languages to a high-source language to run inference on, then projecting the predicted span-level labels back onto the original test data. However, state-of-the-art marker-based label projection methods suffer from translation quality degradation due to the extra label markers injected in the input to the translation model. In this work, we explore a new direction that leverages constrained decoding for label projection to overcome the aforementioned issues. Our new method not only can preserve the quality of translated texts but also has the versatility of being applicable to both translating training and translating test data strategies. This versatility is crucial as our experiments reveal that translating test data can lead to a considerable boost in performance compared to translating only training data. We evaluate on two cross-lingual transfer tasks, namely Named Entity Recognition and Event Argument Extraction, spanning 20 languages. The results demonstrate that our approach outperforms the state-of-the-art marker-based method by a large margin and also shows better performance than other label projection methods that rely on external word alignment.

Zhaohui Yang, Wei Xu, Le Liang et al.

Combining wireless communication with large artificial intelligence (AI) models can open up a myriad of novel application scenarios. In sixth generation (6G) networks, ubiquitous communication and computing resources allow large AI models to serve democratic large AI models-related services to enable real-time applications like autonomous vehicles, smart cities, and Internet of Things (IoT) ecosystems. However, the security considerations and sustainable communication resources limit the deployment of large AI models over distributed wireless networks. This paper provides a comprehensive overview of privacy, security, and trustworthy for distributed wireless large AI model (WLAM). In particular, a detailed privacy and security are analysis for distributed WLAM is fist revealed. The classifications and theoretical findings about privacy and security in distributed WLAM are discussed. Then the trustworthy and ethics for implementing distributed WLAM are described. Finally, the comprehensive applications of distributed WLAM are presented in the context of electromagnetic signal processing.

Xiaofeng Wu, Karl Stratos, Wei Xu

The glyphic writing system of Chinese incorporates information-rich visual features in each character, such as radicals that provide hints about meaning or pronunciation. However, there has been no investigation into whether contemporary Large Language Models (LLMs) and Vision-Language Models (VLMs) can harness these sub-character features in Chinese through prompting. In this study, we establish a benchmark to evaluate LLMs' and VLMs' understanding of visual elements in Chinese characters, including radicals, composition structures, strokes, and stroke counts. Our results reveal that models surprisingly exhibit some, but still limited, knowledge of the visual information, regardless of whether images of characters are provided. To incite models' ability to use radicals, we further experiment with incorporating radicals into the prompts for Chinese language processing (CLP) tasks. We observe consistent improvement in Part-Of-Speech tagging when providing additional information about radicals, suggesting the potential to enhance CLP by integrating sub-character information.

Wei Xu, An Liu, Yiting Zhang et al.

Efficient learning and model compression algorithm for deep neural network (DNN) is a key workhorse behind the rise of deep learning (DL). In this work, we propose a message passing based Bayesian deep learning algorithm called EM-TDAMP to avoid the drawbacks of traditional stochastic gradient descent (SGD) based learning algorithms and regularization-based model compression methods. Specifically, we formulate the problem of DNN learning and compression as a sparse Bayesian inference problem, in which group sparse prior is employed to achieve structured model compression. Then, we propose an expectation maximization (EM) framework to estimate posterior distributions for parameters (E-step) and update hyperparameters (M-step), where the E-step is realized by a newly proposed turbo deep approximate message passing (TDAMP) algorithm. We further extend the EM-TDAMP and propose a novel Bayesian federated learning framework, in which and the clients perform TDAMP to efficiently calculate the local posterior distributions based on the local data, and the central server first aggregates the local posterior distributions to update the global posterior distributions and then update hyperparameters based on EM to accelerate convergence. We detail the application of EM-TDAMP to Boston housing price prediction and handwriting recognition, and present extensive numerical results to demonstrate the advantages of EM-TDAMP.

Jiacheng Wang, Ping Liu, Wei Xu

Existing text-to-image editing methods tend to excel either in rigid or non-rigid editing but encounter challenges when combining both, resulting in misaligned outputs with the provided text prompts. In addition, integrating reference images for control remains challenging. To address these issues, we present a versatile image editing framework capable of executing both rigid and non-rigid edits, guided by either textual prompts or reference images. We leverage a dual-path injection scheme to handle diverse editing scenarios and introduce an integrated self-attention mechanism for fusion of appearance and structural information. To mitigate potential visual artifacts, we further employ latent fusion techniques to adjust intermediate latents. Compared to previous work, our approach represents a significant advance in achieving precise and versatile image editing. Comprehensive experiments validate the efficacy of our method, showcasing competitive or superior results in text-based editing and appearance transfer tasks, encompassing both rigid and non-rigid settings.

Zhenyu Tao, Wei Xu, Xiaohu You

Digital twin (DT)-driven deep reinforcement learning (DRL) has emerged as a promising paradigm for wireless network optimization, offering safe and efficient training environment for policy exploration. However, in theory existing methods cannot always guarantee real-world performance of DT-trained policies before actual deployment, due to the absence of a universal metric for assessing DT's ability to support reliable DRL training transferrable to physical networks. In this paper, we propose the DT bisimulation metric (DT-BSM), a novel metric based on the Wasserstein distance, to quantify the discrepancy between Markov decision processes (MDPs) in both the DT and the corresponding real-world wireless network environment. We prove that for any DT-trained policy, the sub-optimality of its performance (regret) in the real-world deployment is bounded by a weighted sum of the DT-BSM and its sub-optimality within the MDP in the DT. Then, a modified DT-BSM based on the total variation distance is also introduced to avoid the prohibitive calculation complexity of Wasserstein distance for large-scale wireless network scenarios. Further, to tackle the challenge of obtaining accurate transition probabilities of the MDP in real world for the DT-BSM calculation, we propose an empirical DT-BSM method based on statistical sampling. We prove that the empirical DT-BSM always converges to the desired theoretical one, and quantitatively establish the relationship between the required sample size and the target level of approximation accuracy. Numerical experiments validate this first theoretical finding on the provable and calculable performance bounds for DT-driven DRL.

Huahui Yi, Ziyuan Qin, Wei Xu et al.

Large pre-trained vision-language models have shown great prominence in transferring pre-acquired knowledge to various domains and downstream tasks with appropriate prompting or tuning. Existing prevalent tuning methods can be generally categorized into three genres: 1) prompt engineering by creating suitable prompt texts, which is time-consuming and requires domain expertise; 2) or simply fine-tuning the whole model, which is extremely inefficient; 3) prompt tuning through parameterized prompt embeddings with the text encoder. Nevertheless, all methods rely on the text encoder for bridging the modality gap between vision and language. In this work, we question the necessity of the cumbersome text encoder for a more lightweight and efficient tuning paradigm as well as more representative prompt embeddings closer to the image representations. To achieve this, we propose a Concept Embedding Search (ConES) approach by optimizing prompt embeddings -- without the need of the text encoder -- to capture the 'concept' of the image modality through a variety of task objectives. By dropping the text encoder, we are able to significantly speed up the learning process, \eg, from about an hour to just ten minutes in our experiments for personalized text-to-image generation without impairing the generation quality. Moreover, our proposed approach is orthogonal to current existing tuning methods since the searched concept embeddings can be further utilized in the next stage of fine-tuning the pre-trained large models for boosting performance. Extensive experiments show that our approach can beat the prompt tuning and textual inversion methods in a variety of downstream tasks including objection detection, instance segmentation, and image generation. Our approach also shows better generalization capability for unseen concepts in specialized domains, such as the medical domain.

Wei Xu

In this chapter, we review and discuss the transformation of AI technology in HCI/UX work and assess how AI technology will change how we do the work. We first discuss how AI can be used to enhance the result of user research and design evaluation. We then discuss how AI technology can be used to enhance HCI/UX design. Finally, we discuss how AI-enabled capabilities can improve UX when users interact with computing systems, applications, and services.

Dingkang Liang, Wei Xu, Xiang Bai

Crowd localization, predicting head positions, is a more practical and high-level task than simply counting. Existing methods employ pseudo-bounding boxes or pre-designed localization maps, relying on complex post-processing to obtain the head positions. In this paper, we propose an elegant, end-to-end Crowd Localization Transformer named CLTR that solves the task in the regression-based paradigm. The proposed method views the crowd localization as a direct set prediction problem, taking extracted features and trainable embeddings as input of the transformer-decoder. To reduce the ambiguous points and generate more reasonable matching results, we introduce a KMO-based Hungarian matcher, which adopts the nearby context as the auxiliary matching cost. Extensive experiments conducted on five datasets in various data settings show the effectiveness of our method. In particular, the proposed method achieves the best localization performance on the NWPU-Crowd, UCF-QNRF, and ShanghaiTech Part A datasets.

Qinxun Bai, Steven Rosenberg, Wei Xu

Natural gradients have been widely studied from both theoretical and empirical perspectives, and it is commonly believed that natural gradients have advantages over standard (Euclidean) gradients in capturing the intrinsic geometric structure of the underlying function space and being invariant under reparameterization. However, for function optimization, a fundamental theoretical issue regarding the existence of natural gradients on the function space remains underexplored. We address this issue by providing a geometric perspective and mathematical framework for studying both natural gradient and standard gradient that is more complete than existing studies. The key tool that unifies natural gradient and standard gradient is a generalized form of the Neural Tangent Kernel (NTK), which we name the Generalized Tangent Kernel (GTK). Using a novel orthonormality property of GTK, we show that for a fixed parameterization, GTK determines a Riemannian metric on the entire function space which makes the standard gradient as "natural" as the natural gradient in capturing the intrinsic structure of the parameterized function space. Many aspects of this approach relate to RKHS theory. For the practical side of this theory paper, we showcase that our framework motivates new solutions to the non-immersion/degenerate case of natural gradient and leads to new families of natural/standard gradient descent methods.

Haonan Yu, Haichao Zhang, Wei Xu

Maximum entropy (MaxEnt) RL maximizes a combination of the original task reward and an entropy reward. It is believed that the regularization imposed by entropy, on both policy improvement and policy evaluation, together contributes to good exploration, training convergence, and robustness of learned policies. This paper takes a closer look at entropy as an intrinsic reward, by conducting various ablation studies on soft actor-critic (SAC), a popular representative of MaxEnt RL. Our findings reveal that in general, entropy rewards should be applied with caution to policy evaluation. On one hand, the entropy reward, like any other intrinsic reward, could obscure the main task reward if it is not properly managed. We identify some failure cases of the entropy reward especially in episodic Markov decision processes (MDPs), where it could cause the policy to be overly optimistic or pessimistic. On the other hand, our large-scale empirical study shows that using entropy regularization alone in policy improvement, leads to comparable or even better performance and robustness than using it in both policy improvement and policy evaluation. Based on these observations, we recommend either normalizing the entropy reward to a zero mean (SACZero), or simply removing it from policy evaluation (SACLite) for better practical results.

Wei Xu, Dongjiao He, Yixi Cai et al.

This paper presents a generic motion model to capture mobile robots' dynamic behaviors (translation and rotation). The model is based on statistical models driven by white random processes and is formulated into a full state estimation algorithm based on the error-state extended Kalman filtering framework (ESEKF). Major benefits of this method are its versatility, being applicable to different robotic systems without accurately modeling the robots' specific dynamics, and ability to estimate the robot's (angular) acceleration, jerk, or higher-order dynamic states with low delay. Mathematical analysis with numerical simulations are presented to show the properties of the statistical model-based estimation framework and to reveal its connection to existing low-pass filters. Furthermore, a new paradigm is developed for robots observability analysis by developing Lie derivatives and associated partial differentiation directly on manifolds. It is shown that this new paradigm is much simpler and more natural than existing methods based on quaternion parameterizations. It is also scalable to high dimensional systems. A novel \textbf{\textit{thin}} set concept is introduced to characterize the unobservable subset of the system states, providing the theoretical foundation to observability analysis of robotic systems operating on manifolds and in high dimension. Finally, extensive experiments including full state estimation and extrinsic calibration (both POS-IMU and IMU-IMU) on a quadrotor UAV, a handheld platform and a ground vehicle are conducted. Comparisons with existing methods show that the proposed method can effectively estimate all extrinsic parameters, the robot's translation/angular acceleration and other state variables (e.g., position, velocity, attitude) of high accuracy and low delay.

Peng Yin, Yuqing He, Lingyun Xu et al.

Loop Closure Detection (LCD) is the essential module in the simultaneous localization and mapping (SLAM) task. In the current appearance-based SLAM methods, the visual inputs are usually affected by illumination, appearance and viewpoints changes. Comparing to the visual inputs, with the active property, light detection and ranging (LiDAR) based point-cloud inputs are invariant to the illumination and appearance changes. In this paper, we extract 3D voxel maps and 2D top view maps from LiDAR inputs, and the former could capture the local geometry into a simplified 3D voxel format, the later could capture the local road structure into a 2D image format. However, the most challenge problem is to obtain efficient features from 3D and 2D maps to against the viewpoints difference. In this paper, we proposed a synchronous adversarial feature learning method for the LCD task, which could learn the higher level abstract features from different domains without any label data. To the best of our knowledge, this work is the first to extract multi-domain adversarial features for the LCD task in real time. To investigate the performance, we test the proposed method on the KITTI odometry dataset. The extensive experiments results show that, the proposed method could largely improve LCD accuracy even under huge viewpoints differences.

Haonan Yu, Haichao Zhang, Wei Xu

We build a virtual agent for learning language in a 2D maze-like world. The agent sees images of the surrounding environment, listens to a virtual teacher, and takes actions to receive rewards. It interactively learns the teacher's language from scratch based on two language use cases: sentence-directed navigation and question answering. It learns simultaneously the visual representations of the world, the language, and the action control. By disentangling language grounding from other computational routines and sharing a concept detection function between language grounding and prediction, the agent reliably interpolates and extrapolates to interpret sentences that contain new word combinations or new words missing from training sentences. The new words are transferred from the answers of language prediction. Such a language ability is trained and evaluated on a population of over 1.6 million distinct sentences consisting of 119 object words, 8 color words, 9 spatial-relation words, and 50 grammatical words. The proposed model significantly outperforms five comparison methods for interpreting zero-shot sentences. In addition, we demonstrate human-interpretable intermediate outputs of the model in the appendix.

Haichao Zhang, Haonan Yu, Wei Xu

One of the long-term goals of artificial intelligence is to build an agent that can communicate intelligently with human in natural language. Most existing work on natural language learning relies heavily on training over a pre-collected dataset with annotated labels, leading to an agent that essentially captures the statistics of the fixed external training data. As the training data is essentially a static snapshot representation of the knowledge from the annotator, the agent trained this way is limited in adaptiveness and generalization of its behavior. Moreover, this is very different from the language learning process of humans, where language is acquired during communication by taking speaking action and learning from the consequences of speaking action in an interactive manner. This paper presents an interactive setting for grounded natural language learning, where an agent learns natural language by interacting with a teacher and learning from feedback, thus learning and improving language skills while taking part in the conversation. To achieve this goal, we propose a model which incorporates both imitation and reinforcement by leveraging jointly sentence and reward feedbacks from the teacher. Experiments are conducted to validate the effectiveness of the proposed approach.