Hong Mei

Yuqi Zhu, Jia Li, Ge Li et al.

Recently, Large Language Models (LLMs) have shown impressive abilities in code generation. However, existing LLMs' decoding strategies are designed for Natural Language (NL) generation, overlooking the differences between NL and programming languages (PL). Due to this oversight, a better decoding strategy for code generation remains an open question. In this paper, we conduct the first systematic study to explore a decoding strategy specialized in code generation. With an analysis of loss distributions of code tokens, we find that code tokens can be divided into two categories: challenging tokens that are difficult to predict and confident tokens that can be easily inferred. Among them, the challenging tokens mainly appear at the beginning of a code block. Inspired by the above findings, we propose a simple yet effective method: Adaptive Temperature (AdapT) sampling, which dynamically adjusts the temperature coefficient when decoding different tokens. We apply a larger temperature when sampling for challenging tokens, allowing LLMs to explore diverse choices. We employ a smaller temperature for confident tokens avoiding the influence of tail randomness noises. We apply AdapT sampling to LLMs with different sizes and conduct evaluations on two popular datasets. Results show that AdapT sampling significantly outperforms state-of-the-art decoding strategy.

Zihao Zheng, Zhihao Mao, Sicheng Tian et al.

Vision-Language-Action (VLA) Models have become the mainstream solution for robot control, but suffer from slow inference speeds. Speculative Decoding (SD) is a promising acceleration method which can be divided into two categories: drafter-based SD and retrieval-based SD. Existing methods fail to analyze the advantages and disadvantages of these two types of SD in VLA models, leading to their sole application or optimization. In this paper, we analyze the trajectory patterns of robots controlled by the VLA model and derive a key insight: the two types of SD should be used in a hybrid manner. However, achieving hybrid SD in VLA models poses several challenges: (1) draft rejection and persistent errors in retrieval-based SD; (2) difficulty in determining the hybrid boundary. To address these, we propose the HeiSD framework. We propose a retrieval-based SD optimization method in HeiSD,which contains a verify-skip mechanism and a sequence-wise relaxed acceptance strategy. Moreover, we proposed a kinematic-based fused metric in HeiSD to automatically determine the hybrid boundary. Experimental results demonstrate that HeiSD attains a speedup of up to 2.45x in simulation benchmarks and 2.06x~2.41x in real-world scenarios, while sustaining a high task success rate.

Zihao Zheng, Zhihao Mao, Maoliang Li et al.

Vision-Language-Action (VLA) models build a token-domain robot control paradigm, yet suffer from low speed. Speculative Decoding (SD) is an optimization strategy that can boost inference speed. Two key issues emerge when integrating VLA and SD: first, SD relies on re-inference to address token errors, which is computationally expensive; second, to mitigate token errors, the acceptance threshold in SD requires careful adjustment. Existing works fail to address the above two issues effectively. Meanwhile, as the bridge between AI and the physical world, existing embodied intelligence has overlooked the application of robotic kinematics. To address these issues, we innovatively combine token-domain VLA models with kinematic-domain prediction for SD, proposing a kinematic-rectified SD framework named KERV. We employ a kinematics-based Kalman Filter to predict actions and compensate for SD errors, avoiding costly re-inference. Moreover, we design a kinematics-based adjustment strategy to dynamically rectify the acceptance threshold, addressing the difficulty of threshold determination. Experimental results across diverse tasks and environments demonstrate that KERV achieves 27%~37% acceleration with nearly no Success Rate loss.

Yihong Dong, Ge Li, Xue Jiang et al. · pku

Periodicity, as one of the most important basic characteristics, lays the foundation for facilitating structured knowledge acquisition and systematic cognitive processes within human learning paradigms. However, the potential flaws of periodicity modeling in Transformer affect the learning efficiency and establishment of underlying principles from data for large language models (LLMs) built upon it. In this paper, we demonstrate that integrating effective periodicity modeling can improve the learning efficiency and performance of LLMs. We introduce FANformer, which adapts Fourier Analysis Network (FAN) into attention mechanism to achieve efficient periodicity modeling, by modifying the feature projection process of attention mechanism. Extensive experimental results on language modeling show that FANformer consistently outperforms Transformer when scaling up model size and training tokens, underscoring its superior learning efficiency. Our pretrained FANformer-1B exhibits marked improvements on downstream tasks compared to open-source LLMs with similar model parameters or training tokens. Moreover, we reveal that FANformer exhibits superior ability to learn and apply rules for reasoning compared to Transformer. The results position FANformer as an effective and promising architecture for advancing LLMs.

Yang Lin, Xinyu Ma, Xu Chu et al.

Parameter-efficient fine-tuning methods, represented by LoRA, play an essential role in adapting large-scale pre-trained models to downstream tasks. However, fine-tuning LoRA-series models also faces the risk of overfitting on the training dataset, and yet there's still a lack of theoretical guidance and practical mechanism to control overfitting on LoRA-based PEFT methods. In this paper, we propose a LoRA Dropout mechanism for the LoRA-based methods by introducing random noises to the learnable low-rank matrices and increasing parameter sparsity. We then demonstrate the theoretical mechanism of our LoRA Dropout mechanism from the perspective of sparsity regularization by providing a generalization error bound under this framework. Theoretical results show that appropriate sparsity would help tighten the gap between empirical and generalization risks and thereby control overfitting. Furthermore, based on the LoRA Dropout framework, we introduce a test-time ensemble strategy and provide theoretical evidence demonstrating that the ensemble method can further compress the error bound, and lead to better performance during inference time. Extensive experiments on various NLP tasks provide practical validations of the effectiveness of our LoRA Dropout framework in improving model accuracy and calibration.

Yang Yao, Xin Wang, Zeyang Zhang et al. · tsinghua

Large language models (LLMs) have achieved great success in many fields, and recent works have studied exploring LLMs for graph discriminative tasks such as node classification. However, the abilities of LLMs for graph generation remain unexplored in the literature. Graph generation requires the LLM to generate graphs with given properties, which has valuable real-world applications such as drug discovery, while tends to be more challenging. In this paper, we propose LLM4GraphGen to explore the ability of LLMs for graph generation with systematical task designs and extensive experiments. Specifically, we propose several tasks tailored with comprehensive experiments to address key questions regarding LLMs' understanding of different graph structure rules, their ability to capture structural type distributions, and their utilization of domain knowledge for property-based graph generation. Our evaluations demonstrate that LLMs, particularly GPT-4, exhibit preliminary abilities in graph generation tasks, including rule-based and distribution-based generation. We also observe that popular prompting methods, such as few-shot and chain-of-thought prompting, do not consistently enhance performance. Besides, LLMs show potential in generating molecules with specific properties. These findings may serve as foundations for designing good LLMs based models for graph generation and provide valuable insights and further research.

Zihao Zheng, Hangyu Cao, Sicheng Tian et al.

Vision-Language-Action (VLA) models are dominant in embodied intelligence but are constrained by inference overheads. While model quantization alleviates these bottlenecks for edge deployment, static quantization approaches remain suboptimal for VLAs due to two critical challenges: (1) Temporal-dynamic sensitivity, where fixed precision wastes resources by ignoring stage-varying error tolerances; and (2) Real-time allocation, where identifying real-time sensitivity to guide bit allocation remains unsolved. To address these challenges, we propose DyQ-VLA, a dynamic quantization framework for VLAs. Specifically, a sensitivity-aware switching strategy leverages real-time kinematic proxies to trigger the bit-width switch, while a kinematic-guided module dynamically allocates the optimal bit-width. Experiments show that DyQ-VLA requires only 30.9% of the original memory footprint while maintaining 99.5% of its original performance, achieving 1.49x simulation and up to 1.43x real-world speedups.

Zihao Zheng, Zhihao Mao, Xingyue Zhou et al.

Vision-and-Language Navigation (VLN) increasingly relies on large vision-language models, but their inference cost conflicts with real-time deployment. Token caching is a promising training-free strategy that avoids redundant computation by reusing stable visual tokens across frames. However, existing methods assume a static camera and fixed semantic focus, assumptions that VLN fundamentally violates. We identify two failure modes: (1) visual dynamics, where viewpoint shift displaces token positions across frames, causing position-wise matching to pair misaligned content; (2) semantic dynamics, where token relevance shifts across task stages as navigation progresses, making cached states stale. We propose VLN-Cache, a visual-dynamic-aware and semantic-dynamic-aware caching framework that introduces view-aligned remapping to recover geometric correspondences and a task-relevance saliency filter to veto reuse at semantic transitions. A layer-adaptive entropy policy further balances the per-layer reuse budget. Experiments on the R2R-CE simulation benchmark show up to 1.52x speedup while maintaining competitive navigation success rates.

Lecheng Wang, Xianjie Shi, Ge Li et al. · pku

Auto-regressive language models (LMs) have been widely used to generate data in data-scarce domains to train new LMs, compensating for the scarcity of real-world data. Previous work experimentally found that LMs collapse when trained on recursively generated data. This paper presents a theoretical proof: once a corpus (such as a subset of the World Wide Web) begins to incorporate generated data and no new real-world data is added to the corpus, then no matter how small the amount of data each LM generates and contributes to the corpus, LM collapse is inevitable after sufficient time. This finding suggests that attempts to mitigate collapse by limiting the quantity of synthetic data in the corpus are fundamentally insufficient. Instead, avoiding collapse hinges on ensuring the quality of synthetic data.

Xianglin Yang, Yun Lin, Ruofan Liu et al.

Understanding how the predictions of deep learning models are formed during the training process is crucial to improve model performance and fix model defects, especially when we need to investigate nontrivial training strategies such as active learning, and track the root cause of unexpected training results such as performance degeneration. In this work, we propose a time-travelling visual solution DeepVisualInsight (DVI), aiming to manifest the spatio-temporal causality while training a deep learning image classifier. The spatio-temporal causality demonstrates how the gradient-descent algorithm and various training data sampling techniques can influence and reshape the layout of learnt input representation and the classification boundaries in consecutive epochs. Such causality allows us to observe and analyze the whole learning process in the visible low dimensional space. Technically, we propose four spatial and temporal properties and design our visualization solution to satisfy them. These properties preserve the most important information when inverse-)projecting input samples between the visible low-dimensional and the invisible high-dimensional space, for causal analyses. Our extensive experiments show that, comparing to baseline approaches, we achieve the best visualization performance regarding the spatial/temporal properties and visualization efficiency. Moreover, our case study shows that our visual solution can well reflect the characteristics of various training scenarios, showing good potential of DVI as a debugging tool for analyzing deep learning training processes.

Xuezhi Song, Yun Lin, Siang Hwee Ng et al.

Bug datasets consisting of real-world bugs are important artifacts for researchers and programmers, which lay empirical and experimental foundation for various SE/PL research such as fault localization, software testing, and program repair. All known state-of-the-art datasets are constructed manually, which inevitably limits their scalability, representativeness, and the support for the emerging data-driven research. In this work, we propose an approach to automate the process of harvesting replicable regression bugs from the code evolutionary history. We focus on regression bug dataset, as they (1) manifest how a bug is introduced and fixed (as normal bugs), (2) support regression bug analysis, and (3) incorporate a much stronger specification (i.e., the original passing version) for general bug analysis. Technically, we address an information retrieval problem on code evolution history. Given a code repository, we search for regressions where a test can pass a regression-fixing commit, fail a regressioninducing commit, and pass a working commit. In this work, we address the challenges of (1) identifying potential regression-fixing commits from the code evolution history, (2) migrating the test and its code dependencies over the history, and (3) minimizing the compilation overhead during the regression search. We build our tool, RegMiner, which harvested 537 regressions over 66 projects for 3 weeks, created the largest replicable regression dataset within shortest period, to the best of our knowledge. Moreover, our empirical study on our regression dataset shows a gap between the popular regression fault localization techniques (e.g, delta-debugging) and the real fix, revealing new data-driven research opportunities.

Wenjie Chu, Wei Zhang, Haiyan Zhao et al.

Self-assembly plays an essential role in many natural processes, involving the formation and evolution of living or non-living structures, and shows potential applications in many emerging domains. In existing research and practice, there still lacks an ideal self-assembly mechanism that manifests efficiency, scalability, and stability at the same time. Inspired by phototaxis observed in nature, we propose a computational approach for massive self-assembly of connected shapes in grid environments. The key component of this approach is an artificial light field superimposed on a grid environment, which is determined by the positions of all agents and at the same time drives all agents to change their positions, forming a dynamic mutual feedback process. This work advances the understanding and potential applications of self-assembly.

Yan Li, Bo An, Junming Ma et al.

Hyper-parameter tuning (HPT) is crucial for many machine learning (ML) algorithms. But due to the large searching space, HPT is usually time-consuming and resource-intensive. Nowadays, many researchers use public cloud resources to train machine learning models, convenient yet expensive. How to speed up the HPT process while at the same time reduce cost is very important for cloud ML users. In this paper, we propose SpotTune, an approach that exploits transient revocable resources in the public cloud with some tailored strategies to do HPT in a parallel and cost-efficient manner. Orchestrating the HPT process upon transient servers, SpotTune uses two main techniques, fine-grained cost-aware resource provisioning, and ML training trend predicting, to reduce the monetary cost and runtime of HPT processes. Our evaluations show that SpotTune can reduce the cost by up to 90% and achieve a 16.61x performance-cost rate improvement.

Xuanzhe Liu, Huoran Li, Xuan Lu et al.

The prevalence of smart mobile devices has promoted the popularity of mobile applications (a.k.a. apps). Supporting mobility has become a promising trend in software engineering research. This article presents an empirical study of behavioral service profiles collected from millions of users whose devices are deployed with Wandoujia, a leading Android app store service in China. The dataset of Wandoujia service profiles consists of two kinds of user behavioral data from using 0.28 million free Android apps, including (1) app management activities (i.e., downloading, updating, and uninstalling apps) from over 17 million unique users and (2) app network usage from over 6 million unique users. We explore multiple aspects of such behavioral data and present patterns of app usage. Based on the findings as well as derived knowledge, we also suggest some new open opportunities and challenges that can be explored by the research community, including app development, deployment, delivery, revenue, etc.

Xuanzhe Liu, Yun Ma, Shuailiang Dong et al.

Redundant transfer of resources is a critical issue for compromising the performance of mobile Web applications (a.k.a., apps) in terms of data traffic, load time, and even energy consumption. Evidence shows that the current cache mechanisms are far from satisfactory. With lessons learned from how native apps manage their resources, in this paper, we propose the ReWAP approach to fundamentally reducing redundant transfers by restructuring the resource loading of mobile Web apps. ReWAP is based on an efficient mechanism of resource packaging where stable resources are encapsulated and maintained into a package, and such a package shall be loaded always from the local storage and updated by explicitly refreshing. By retrieving and analyzing the update of resources, ReWAP maintains resource packages that can accurately identify which resources can be loaded from the local storage for a considerably long period. ReWAP also provides a wrapper for mobile Web apps to enable loading and updating resource packages in the local storage as well as loading resources from resource packages. ReWAP can be easily and seamlessly deployed into existing mobile Web architectures with minimal modifications, and is transparent to end-users. We evaluate ReWAP based on continuous 15-day access traces of 50 mobile Web apps that suffer heavily from the problem of redundant transfers. Compared to the original mobile Web apps with cache enabled, ReWAP can significantly reduce the data traffic, with the median saving up to 51%. In addition, ReWAP can incur only very minor runtime overhead of the client-side browsers.

Xuanzhe Liu, Mengwei Xu, Gang Huang et al.

In enterprise organizations, the Bring-Your-Own-Device (BYOD) requirement has become prevalent as employees use their own mobile devices to process the workflow-oriented tasks. Consequently, it calls for approaches that can quickly develop and integrate mobile user interactions into existing business processes, and adapt to various contexts. However, designing, developing and deploying adaptive and mobile-oriented user interfaces for existing process engines are non-trivial, and require significant systematic efforts. To address this issue, we present a novel middleware-based approach, called MUIT, to developing and deploying the Mobility, User Interactions and Tasks into WS-BPEL engines. MUIT can be seamlessly into WS-BPEL without intrusions of existing process instances. MUIT provides a Domain-Specific Language (DSL) that provides some intuitive APIs to support the declarative development of adaptive, mobile-oriented, and Web-based user interfaces in WS-BPEL. The DSL can significantly improve the development of user interactions by preventing arbitrarily mixed codes, and its runtime supports satisfactory user experiences. We implement a proof- of-concept prototype by integrating MUIT into the commodity WS-BPEL-based Apusic Platform, and evaluate the performance and usability of MUIT platform.

Lin Yan, Yao Guo, Xiangqun Chen et al.

Power side channel is a very important category of side channels, which can be exploited to steal confidential information from a computing system by analyzing its power consumption. In this paper, we demonstrate the existence of various power side channels on popular mobile devices such as smartphones. Based on unprivileged power consumption traces, we present a list of real-world attacks that can be initiated to identify running apps, infer sensitive UIs, guess password lengths, and estimate geo-locations. These attack examples demonstrate that power consumption traces can be used as a practical side channel to gain various confidential information of mobile apps running on smartphones. Based on these power side channels, we discuss possible exploitations and present a general approach to exploit a power side channel on an Android smartphone, which demonstrates that power side channels pose imminent threats to the security and privacy of mobile users. We also discuss possible countermeasures to mitigate the threats of power side channels.