arXiv subjects

Jiang Xu

Publications and source records attributed to Jiang Xu.

At least 37 records · Page 2Linked to original sources

arXiv2025

DeepServe: Serverless Large Language Model Serving at Scale

In this paper, we propose DEEPSERVE, a scalable and serverless AI platform designed to efficiently serve large language models (LLMs) at scale in cloud environments. DEEPSERVE addresses key challenges such as resource allocation, serving efficiency, and cold start latencies through four main design components. First, DEEPSERVE uses a simple serverless abstraction called the request-job-task model, which helps manage diverse AI workloads across posttraining and model-serving tasks. Second, DEEPSERVE integrates an in-house serving engine named FLOWSERVE using a microkernel-inspired design, NPU-centric execution, and SPMD-based parallelism to optimize LLM serving. Third, DEEPSERVE includes novel scheduling policies tailored for a configuration with both PD-disaggregated and PD-colocated instances. Fourth, DEEPSERVE includes optimizations such as pre-warmed pods, DRAM pre-loading, and NPU-fork, which allow DEEPSERVE to scale up to 64 instances in seconds. DEEPSERVE has been in production for over a year, operating on a large Ascend NPU cluster and providing industrystandard APIs for fine-tuning, agent serving, and model serving to our customers.

cs.DC↗

arXiv2025

SpNeRF: Memory Efficient Sparse Volumetric Neural Rendering Accelerator for Edge Devices

Neural rendering has gained prominence for its high-quality output, which is crucial for AR/VR applications. However, its large voxel grid data size and irregular access patterns challenge real-time processing on edge devices. While previous works have focused on improving data locality, they have not adequately addressed the issue of large voxel grid sizes, which necessitate frequent off-chip memory access and substantial on-chip memory. This paper introduces SpNeRF, a software-hardware co-design solution tailored for sparse volumetric neural rendering. We first identify memory-bound rendering inefficiencies and analyze the inherent sparsity in the voxel grid data of neural rendering. To enhance efficiency, we propose novel preprocessing and online decoding steps, reducing the memory size for voxel grid. The preprocessing step employs hash mapping to support irregular data access while maintaining a minimal memory size. The online decoding step enables efficient on-chip sparse voxel grid processing, incorporating bitmap masking to mitigate PSNR loss caused by hash collisions. To further optimize performance, we design a dedicated hardware architecture supporting our sparse voxel grid processing technique. Experimental results demonstrate that SpNeRF achieves an average 21.07$\times$ reduction in memory size while maintaining comparable PSNR levels. When benchmarked against Jetson XNX, Jetson ONX, RT-NeRF.Edge and NeuRex.Edge, our design achieves speedups of 95.1$\times$, 63.5$\times$, 1.5$\times$ and 10.3$\times$, and improves energy efficiency by 625.6$\times$, 529.1$\times$, 4$\times$, and 4.4$\times$, respectively.

cs.AR↗

arXiv2025

Knowledge Phenomenology Research of Future Industrial Iconic Product Innovation

Iconic products, as innovative carriers supporting the development of future industries, are key breakthrough points for driving the transformation of new quality productive forces. This article is grounded in the philosophy of technology and examines the evolution of human civilization to accurately identify the patterns of product innovation. By integrating theories from systems science, it analyzes the intrinsic logical differences between traditional products and iconic products. The study finds that iconic products are based on a comprehensive knowledge system that integrates explicit and tacit knowledge, enabling them to adapt to complex dynamic environments. Therefore, based on the method of phenomenological essence reduction and the process of specialized knowledge acquisition, this study establishes the first principle of knowledge phenomenology: "knowledge generation-moving from the tacit to the explicit-moving from the explicit to the tacit-fusion of the explicit and tacit." Grounded in knowledge phenomenology, it reconstructs the product design evolution process and establishes a forward innovative design framework for iconic products, consisting of "design problem space-explicit knowledge space-tacit knowledge space-innovative solution space." Furthermore, based on FBS design theory, it develops a disruptive technology innovation forecasting framework of "technology problem space-knowledge base prediction-application scenario prediction-coupled technology prediction," which collectively advances the innovation systems engineering of iconic products. In light of the analysis of the global future industrial competitive landscape, it proposes a strategy for enhancing embodied intelligence in iconic products.

econ.TH↗

arXiv2024

MemServe: Context Caching for Disaggregated LLM Serving with Elastic Memory Pool

Large language model (LLM) serving has transformed from stateless to stateful systems, utilizing techniques like context caching and disaggregated inference. These optimizations extend the lifespan and domain of the KV cache, necessitating a new architectural approach. We present MemServe, a unified system that integrates both inter-request and intra-request optimizations. MemServe introduces MemPool, an elastic memory pool managing distributed memory and KV caches across serving instances. Using MemPool APIs, MemServe combines context caching with disaggregated inference for the first time, supported by a global scheduler that enhances cache reuse through a global prompt tree-based locality-aware policy. Tests show that MemServe significantly improves job completion time and time-to-first-time.

cs.DC↗

arXiv2024

Intuitive interaction flow: A Dual-Loop Human-Machine Collaboration Task Allocation Model and an experimental study

This study investigates the issue of task allocation in Human-Machine Collaboration (HMC) within the context of Industry 4.0. By integrating philosophical insights and cognitive science, it clearly defines two typical modes of human behavior in human-machine interaction(HMI): skill-based intuitive behavior and knowledge-based intellectual behavior. Building on this, the concept of 'intuitive interaction flow' is innovatively introduced by combining human intuition with machine humanoid intelligence, leading to the construction of a dual-loop HMC task allocation model. Through comparative experiments measuring electroencephalogram (EEG) and electromyogram (EMG) activities, distinct physiological patterns associated with these behavior modes are identified, providing a preliminary foundation for future adaptive HMC frameworks. This work offers a pathway for developing intelligent HMC systems that effectively integrate human intuition and machine intelligence in Industry 4.0.

cs.HC↗

arXiv2024

Diamond of Thought: A Design Thinking-Based Framework for LLMs in Wearable Design

Wearable design is an interdisciplinary field that balances technological innovation, human factors, and human-computer interactions. Despite contributions from various disciplines, many projects lack stable interdisciplinary teams, which often leads to design failures. Large language models (LLMs) integrate diverse information and generate innovative solutions, making them a valuable tool for enhancing design processes. Thus, we have explored the use of LLMs in wearable design by combining design-thinking principles with LLM capabilities. We have developed the "Diamond of Thought" framework and analysed 1,603 prototypes and 1,129 products from a body-centric perspective to create a comprehensive database. We employed retrieval-augmented generation to input database details into the LLMs, ensuring applicability to wearable design challenges and integration of embodied cognition into the process. Our LLM-based methodology for wearables has been experimentally validated, demonstrating the potential of LLMs for the advancement of design practices. This study offers new tools and methods for future wearable designs.

cs.HC↗

arXiv2024

A new characterization of the dissipation structure and the relaxation limit for the compressible Euler-Maxwell system

We investigate the three-dimensional compressible Euler-Maxwell system, a model for simulating the transport of electrons interacting with propagating electromagnetic waves in semiconductor devices. First, we show the global well-posedness of classical solutions being a sharp small perturbation of constant equilibrium in a critical regularity setting, uniformly with respect to the relaxation parameter $\varepsilon>0$. Then, for all times $t>0$, we derive quantitative error estimates at the rate $O(\varepsilon)$ between the rescaled Euler-Maxwell system and the limit drift-diffusion model. To the best of our knowledge, this work provides the first global-in-time strong convergence for the relaxation procedure in the case of ill-prepared data. In order to prove our results, we develop a new characterization of the dissipation structure for the linearized Euler-Maxwell system with respect to the relaxation parameter $\varepsilon$. This is done by partitioning the frequency space into three distinct regimes: low, medium and high frequencies, each associated with a different behaviour of the solution. Then, in each regime, the use of efficient unknowns and Lyapunov functionals based on the hypocoercivity theory leads to uniform a priori estimates.

math.AP↗

arXiv2024

The pressureless damped Euler-Riesz system in the critical regularity framework

We are concerned with a system governing the evolution of the pressureless compressible Euler equations with Riesz interaction and damping in $\mathbb{R}^{d}$ ($d\geq1$), where the interaction force is given by $\nabla(-Δ)^{\smash{\frac{α-d}{2}}}(ρ-\barρ)$ with $d-2<α<d$. Referring to the standard dissipative structure of first-order hyperbolic systems, the purpose of this paper is to investigate the weaker dissipation effect arising from the interaction force and to establish the global existence and large-time behavior of solutions to the Cauchy problem in the critical $L^p$ framework. More precisely, it is observed by the spectral analysis that the density behaves like fractional heat diffusion at low frequencies. Furthermore, if the low-frequency part of the initial perturbation is bounded in some Besov space $\dot{B}^{σ_1}_{p,\infty}$ with $-d/p-1\leq σ_1<d/p-1$, it is shown that the $L^p$-norm of the $σ$-order derivative for the density converges to its equilibrium at the rate $(1+t)^{-\smash{\frac{σ-σ_1}{α-d+2}}}$, which coincides with that of the fractional heat kernel.

math.AP↗

arXiv2024

The Cattaneo-Christov approximation of Fourier heat-conductive compressible fluids

We investigate the Navier-Stokes-Cattaneo-Christov (NSC) system in $\mathbb{R}^d$ ($d\geq3$), a model of heat-conductive compressible flows serving as a finite speed of propagation approximation of the Navier-Stokes-Fourier (NSF) system. Due to the presence of Oldroyd's upper-convected derivatives, the system (NSC) exhibits a \textit{lack of hyperbolicity} which makes it challenging to establish its well-posedness, especially in multi-dimensional contexts. In this paper, within a critical regularity functional framework, we prove the global-in-time well-posedness of (NSC) for initial data that are small perturbations of constant equilibria, uniformly with respect to the approximation parameter $\varepsilon>0$. Then, building upon this result, we obtain the sharp large-time asymptotic behaviour of (NSC) and, for all time $t>0$, we derive quantitative error estimates between the solutions of (NSC) and (NSF). To the best of our knowledge, our work provides the first strong convergence result for this relaxation procedure in the three-dimensional setting and for ill-prepared data. The (NSC) system is partially dissipative and incorporates both partial diffusion and partial damping mechanisms. To address these aspects and ensure the large-time stability of the solutions, we construct localized-in-frequency perturbed energy functionals based on the hypocoercivity theory. More precisely, our analysis relies on partitioning the frequency space into \textit{three} distinct regimes: low, medium and high frequencies. Within each frequency regime, we introduce effective unknowns and Lyapunov functionals, revealing the spectrally expected dissipative structures.

math.AP↗

arXiv2024

Learning Exhaustive Correlation for Spectral Super-Resolution: Where Spatial-Spectral Attention Meets Linear Dependence

Spectral super-resolution that aims to recover hyperspectral image (HSI) from easily obtainable RGB image has drawn increasing interest in the field of computational photography. The crucial aspect of spectral super-resolution lies in exploiting the correlation within HSIs. However, two types of bottlenecks in existing Transformers limit performance improvement and practical applications. First, existing Transformers often separately emphasize either spatial-wise or spectral-wise correlation, disrupting the 3D features of HSI and hindering the exploitation of unified spatial-spectral correlation. Second, existing self-attention mechanism always establishes full-rank correlation matrix by learning the correlation between pairs of tokens, leading to its inability to describe linear dependence widely existing in HSI among multiple tokens. To address these issues, we propose a novel Exhaustive Correlation Transformer (ECT) for spectral super-resolution. First, we propose a Spectral-wise Discontinuous 3D (SD3D) splitting strategy, which models unified spatial-spectral correlation by integrating spatial-wise continuous splitting strategy and spectral-wise discontinuous splitting strategy. Second, we propose a Dynamic Low-Rank Mapping (DLRM) model, which captures linear dependence among multiple tokens through a dynamically calculated low-rank dependence map. By integrating unified spatial-spectral attention and linear dependence, our ECT can model exhaustive correlation within HSI. The experimental results on both simulated and real data indicate that our method achieves state-of-the-art performance. Codes and pretrained models will be available later.

eess.IV↗

arXiv2024

Bodioid: philosophical reflections on the hybrid of bodies and artefacts towards post-human

The advent of the post-human era has blurred the boundary between the body and artefacts. Further, external materials and information are more deeply integrated into the body, making emerging technology a key driving force for shaping post-human existence and promoting bodily evolution. Based on this, this study analyses the transformation process of three technological forms, namely tools, machines, and cyborgs, and reveals the construction of bodies and artefacts. From the phenomenological perspective, the essences of body and artefact existences are reflected upon, and the 'existence is construction' viewpoint is proposed. Furthermore, a technological design concept, 'bodioid', is proposed to meticulously depict the characteristics of integrating similarities and differences towards unity between the body and artefacts, based on the theoretical foundation of technology mediation and the materialization of morality. Finally, through analogizing the organizational form of language, the two key forms and specific mechanisms of bodioid construction, namely extension and mirroring, are indicated. With this in mind, the post-human existence landscape is discussed with the objective of providing theoretical insights into the study of the underlying philosophical principles of technological design.

cs.HC↗

arXiv2024

Inference without Interference: Disaggregate LLM Inference for Mixed Downstream Workloads

Transformer-based large language model (LLM) inference serving is now the backbone of many cloud services. LLM inference consists of a prefill phase and a decode phase. However, existing LLM deployment practices often overlook the distinct characteristics of these phases, leading to significant interference. To mitigate interference, our insight is to carefully schedule and group inference requests based on their characteristics. We realize this idea in TetriInfer through three pillars. First, it partitions prompts into fixed-size chunks so that the accelerator always runs close to its computationsaturated limit. Second, it disaggregates prefill and decode instances so each can run independently. Finally, it uses a smart two-level scheduling algorithm augmented with predicted resource usage to avoid decode scheduling hotspots. Results show that TetriInfer improves time-to-first-token (TTFT), job completion time (JCT), and inference efficiency in turns of performance per dollar by a large margin, e.g., it uses 38% less resources all the while lowering average TTFT and average JCT by 97% and 47%, respectively.

cs.DC↗

arXiv2023

Global existence and analyticity of $L^p$ solutions to the compressible fluid model of Korteweg type

We are concerned with a system of equations in $\mathbb{R}^{d}(d\geq2)$ governing the evolution of isothermal, viscous and compressible fluids of Korteweg type, that can be used as a phase transition model. In the case of zero sound speed $P'(ρ^{\ast})=0$, it is found that the linearized system admits the \textit{purely} parabolic structure, which enables us to establish the global-in-time existence and Gevrey analyticity of strong solutions in hybrid Besov spaces of $L^p$-type. Precisely, if the full viscosity coefficient and capillary coefficient satisfy $\barν^2\geq4\barκ$, then the acoustic waves are not available in compressible fluids. Consequently, the prior $L^2$ bounds on the low frequencies of density and velocity could be improved to the general $L^p$ version with $1\leq p\leq d$ if $d\geq2$. The proof mainly relies on new nonlinear Besov (-Gevrey) estimates for product and composition of functions.

math.AP↗

arXiv2022

Decay of higher order derivatives for $L^p$ solutions to the compressible fluid model of Korteweg type

We present a new derivation for the optimal decay of \textit{arbitrary} higher order derivatives for $L^p$ solutions to the compressible fluid model of Korteweg type. This approach, based on Gevrey estimates, is to establish uniform bounds on the growth of the radius of analyticity of the solution in negative Besov norms. For that end, the maximal regularity property involving Gevrey multiplier of heat kernel and non standard product Besov estimates are well developed. Our approach is partly inspired by Oliver-Titi's work and is applicable to a wide range of dissipative systems.

math.AP↗

arXiv2020

Global existence in critical spaces for non Newtonian compressible viscoelastic flows

We are interested in the multi-dimentional compressible viscoelastic flows of Oldroyd type, which is one of non-Newtonian fluids exhibiting the elastic behavior. In order to capture the damping effect of the additional deformation tensor, to the best of our knowledge, the "div-curl" structural condition plays a key role in previous efforts. Our aim of this paper is to remove the structural condition and prove a global existence of strong solutions to compressible viscoelastic flows in critical spaces. The new ingredient lies in the introduction of effective flux $(θ,\mathcal{G})$, which enables us to capture the dissipation arising from \textit{combination} of density and deformation tensor. In absence of compatible conditions, the partial dissipation is found in non-Newtonian compressible fluids, which is weaker than that of usual Navier-Stokes equations.

math.AP↗

arXiv2019

The large-time behavior of solutions in the critical $L^p$ framework for compressible viscous and heat-conductive gas flows

The $L^{p}$ theory for non-isentropic Navier-Stokes equations governing compressible viscous and heat-conductive gases is not yet proved completely so far, because the critical regularity cannot control all non linear coupling terms. In this paper, we pose an additional regularity assumption of low frequencies in $\mathbb{R}^d(d\geq 3)$, and then the sharp time-weighted inequality can be established, which leads to the time-decay estimates of global strong solutions in the $L^{p}$ critical Besov spaces. Precisely, we show that if the initial data belong to some Besov space $\dot{B}^{-s_{1}}_{2,\infty}$ with $s_{1}\in (1-\frac{d}{2}, s_0](s_0\triangleq \frac{2d}{p}-\frac{d}{2})$, then the $L^{p}$ norm of the critical global solutions admits the time decay $t^{-\frac{s_{1}}{2}-\frac{d}{2}(\frac{1}{2}-\frac{1}{p})}$ (in particular, $t^{-\frac{d}{2p}}$ if $s_1=s_0$), which coincides with that of heat kernel in the $L^p$ framework. In comparison with \cite{DX2}, the low-frequency regularity $s_1$ can be improved to be \textit{the whole range}.

math.AP↗

arXiv2019

Optimal decay for the compressible Navier-Stokes equations without additional smallness assumptions

This work is concerned with the large time behavior of solutions to the barotropic compressible Navier-Stokes equations in $\mathbb{R}^{d}(d\geq2)$. Precisely, it is shown that if the initial density and velocity additionally belong to some Besov space $\dot{B}^{-σ_1}_{2,\infty}$ with $σ_1\in (1-d/2, 2d/p-d/2]$, then the $L^p$ norm (the slightly stronger $\dot{B}^{0}_{p,1}$ norm in fact) of global solutions admits the optimal decay $t^{-\frac{d}{2}(\frac 12-\frac 1p)-\frac{σ_1}{2}}$ for $t\rightarrow+\infty$. In contrast to refined time-weighted approaches ([11,43]), a pure energy argument (independent of the spectral analysis) has been developed in more general $L^p$ critical framework, which allows to remove the smallness of low frequencies of initial data. Indeed, bounding the evolution of $\dot{B}^{-σ_1}_{2,\infty}$-norm restricted in low frequencies is the key ingredient, whose proof mainly depends on non standard $L^p$ product estimates with respect to different Sobolev embeddings. The result can hold true in case of large highly oscillating initial velocities.

math.AP↗

arXiv2019

The Cauchy problem for the inhomogeneous non-cutoff Kac equation in critical Besov space

In this work, we investigate the Cauchy problem for the spatially inhomogeneous non-cutoff Kac equation. If the initial datum belongs to the spatially critical Besov space, we can prove the well-posedness of weak solution under a perturbation framework. Furthermore, it is shown that the solution enjoys Gelfand-Shilov regularizing properties with respect to the velocity variable and Gevrey regularizing properties with respect to the position variable. In comparison with the recent result in [18], the Gelfand-Shilov regularity index is improved to be optimal. To the best of our knowledge, our work is the first one that exhibits smoothing effect for the kinetic equation in Besov spaces.

math.AP↗