arXiv subjects

Liangliang Xu

Publications and source records attributed to Liangliang Xu.

14 recordsLinked to original sources

arXiv2026

Energy-Efficient LLM Serving via Disaggregated Attention--FFN and Flexible Frequency Scaling

Large language model (LLM) serving spans diverse applications with stringent service-level objectives (SLOs), often requiring GPUs to run at maximum frequencies and increasing energy consumption. Existing energy-management approaches adapt GPU frequencies only at the request or inference-phase level, overlooking operator-level differences in frequency sensitivity between Attention and feed-forward networks (FFNs). We find that the energy-optimal frequencies of Attention and FFN (A/F) differ and vary with the inference phase, workload, and system configurations. However, runtime variability and independent A/F frequency control create a large search space and high communication overhead. To address these challenges, we present AFlex, a framework that jointly optimizes resource provisioning and GPU frequency scaling for disaggregated A/F serving. AFlex introduces a global scheduler and a local operator-level dynamic voltage and frequency scaling (DVFS) controller to determine A/F resource allocations and frequencies. It further introduces an interleaved A/F pipeline with dynamic microbatch depth and adaptive request batching to reduce pipeline bubbles. We implement AFlex in SGLang and evaluate it on NVIDIA A800 GPUs using Qwen3-32B and Mixtral-8$\times$7B under production Conversation and Coding traces. \AFlex reduces energy per token by up to 49\% over state-of-the-art disaggregated serving and 48\% over frequency-scaling systems while satisfying TTFT and TPOT SLOs.

cs.DC↗

arXiv2026

STQuant: Spatio-Temporal Adaptive Framework for Optimizer Quantization in Large Multimodal Model Training

Quantization is an effective way to reduce the memory cost of large-scale model training. However, most existing methods adopt fixed-precision policies, which ignore the fact that optimizer-state distributions vary significantly across layers and training steps. Such uniform designs often introduce noticeable accuracy degradation. To move beyond fixed quantization, we propose STQuant, a distributed training framework that reduces the memory footprint of optimizer states via dynamic precision allocation across layers, state variables, and training steps, while maintaining model quality. Naively applying dynamic quantization during training is challenging for two reasons. First, optimizer states are numerically sensitive, and quantization noise can destabilize quality. Second, jointly considering multiple states and layers induces a large combinatorial search space. STQuant addresses these challenges with two key techniques: 1) a provably near-optimal factor selection strategy that accurately identifies the most influential factors for precision adaptation. 2) a dynamic transition decision algorithm that reduces the search cost from exponential to linear complexity. Experiments on GPT-2 and ViT show that STQuant reduces optimizer-state memory by 84.4%, achieving an average bit-width of as low as 5.1 bits, compared with existing solutions. Moreover, STQuant incurs only O(N/K) computational overhead and requires O(1) extra space.

cs.LG↗

arXiv2026

Efficient and Principled Scientific Discovery through Bayesian Optimization: A Tutorial

Traditional scientific discovery relies on an iterative hypothesise-experiment-refine cycle that has driven progress for centuries, but its intuitive, ad-hoc implementation often wastes resources, yields inefficient designs, and misses critical insights. This tutorial presents Bayesian Optimisation (BO), a principled probability-driven framework that formalises and automates this core scientific cycle. BO uses surrogate models (e.g., Gaussian processes) to model empirical observations as evolving hypotheses, and acquisition functions to guide experiment selection, balancing exploitation of known knowledge and exploration of uncharted domains to eliminate guesswork and manual trial-and-error. We first frame scientific discovery as an optimisation problem, then unpack BO's core components, end-to-end workflows, and real-world efficacy via case studies in catalysis, materials science, organic synthesis, and molecule discovery. We also cover critical technical extensions for scientific applications, including batched experimentation, heteroscedasticity, contextual optimisation, and human-in-the-loop integration. Tailored for a broad audience, this tutorial bridges AI advances in BO with practical natural science applications, offering tiered content to empower cross-disciplinary researchers to design more efficient experiments and accelerate principled scientific discovery.

cs.LG↗

arXiv2025

DDiT: Dynamic Resource Allocation for Diffusion Transformer Model Serving

The Text-to-Video (T2V) model aims to generate dynamic and expressive videos from textual prompts. The generation pipeline typically involves multiple modules, such as language encoder, Diffusion Transformer (DiT), and Variational Autoencoders (VAE). Existing serving systems often rely on monolithic model deployment, while overlooking the distinct characteristics of each module, leading to inefficient GPU utilization. In addition, DiT exhibits varying performance gains across different resolutions and degrees of parallelism, and significant optimization potential remains unexplored. To address these problems, we present DDiT, a flexible system that integrates both inter-phase and intra-phase optimizations. DDiT focuses on two key metrics: optimal degree of parallelism, which prevents excessive parallelism for specific resolutions, and starvation time, which quantifies the sacrifice of each request. To this end, DDiT introduces a decoupled control mechanism to minimize the computational inefficiency caused by imbalances in the degree of parallelism between the DiT and VAE phases. It also designs a greedy resource allocation algorithm with a novel scheduling mechanism that operates at the single-step granularity, enabling dynamic and timely resource scaling. Our evaluation on the T5 encoder, OpenSora SDDiT, and OpenSora VAE models across diverse datasets reveals that DDiT significantly outperforms state-of-the-art baselines by up to 1.44x in p99 latency and 1.43x in average latency.

cs.DC↗

arXiv2025

New Wide Locally Recoverable Codes with Unified Locality

Wide Locally Recoverable Codes (LRCs) have recently been proposed as a solution for achieving high reliability, good performance, and ultra-low storage cost in distributed storage systems. However, existing wide LRCs struggle to balance optimal fault tolerance and high availability during frequent system events. By analyzing the existing LRCs, we reveal three limitations in the LRC construction which lay behind the non-optimal overall performance from multiple perspectives, including non-minimum local recovery cost, non cluster-topology-aware data distribution, and non XOR-based local coding. Thanks to the flexible design space offered by the locality property of wide LRCs, we present UniLRC, which unifies locality considerations in code construction. UniLRC achieves the optimal fault tolerance while overcoming the revealed limitations. We implement UniLRC prototype and conduct comprehensive theoretical and system evaluations, showing significant improvements in reliability and performance over existing wide LRCs deployed in Google and Azure clusters.

cs.DC↗

arXiv2025

Dynamical Mode Recognition of Turbulent Flames in a Swirl-stabilized Annular Combustor by a Time-series Learning Approach

Thermoacoustic instability in annular combustors, essential to aero engines and modern gas turbines, can severely impair operational stability and efficiency, accurately recognizing and understanding various combustion modes is the prerequisite for understanding and controlling combustion instabilities. However, the high-dimensional spatial-temporal dynamics of turbulent flames typically pose considerable challenges to mode recognition. Based on the bidirectional temporal and nonlinear dimensionality reduction models, this study introduces a two-layer bidirectional long short-term memory variational autoencoder, Bi-LSTM-VAE model, to effectively recognize dynamical modes in annular combustion systems. Specifically, leveraging 16 pressure signals from a swirl-stabilized annular combustor, the model maps complex dynamics into a low-dimensional latent space while preserving temporal dependency and nonlinear behavior features through the recurrent neural network structure. The results show that the novel Bi-LSTM-VAE method enables a clear representation of combustion states in two-dimensional state space. Analysis of latent variable distributions reveals distinct patterns corresponding to a wide range of equivalence ratios and premixed fuel and air mass flow rates, offering novel insights into mode classification and transitions, highlighting this model's potential for deciphering complex thermoacoustic phenomena.

cs.LG↗

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

A Critical Review of Inductive Logic Programming Techniques for Explainable AI

Despite recent advances in modern machine learning algorithms, the opaqueness of their underlying mechanisms continues to be an obstacle in adoption. To instill confidence and trust in artificial intelligence systems, Explainable Artificial Intelligence has emerged as a response to improving modern machine learning algorithms' explainability. Inductive Logic Programming (ILP), a subfield of symbolic artificial intelligence, plays a promising role in generating interpretable explanations because of its intuitive logic-driven framework. ILP effectively leverages abductive reasoning to generate explainable first-order clausal theories from examples and background knowledge. However, several challenges in developing methods inspired by ILP need to be addressed for their successful application in practice. For example, existing ILP systems often have a vast solution space, and the induced solutions are very sensitive to noises and disturbances. This survey paper summarizes the recent advances in ILP and a discussion of statistical relational learning and neural-symbolic algorithms, which offer synergistic views to ILP. Following a critical review of the recent advances, we delineate observed challenges and highlight potential avenues of further ILP-motivated research toward developing self-explanatory artificial intelligence systems.

cs.LG↗

arXiv2022

STOPS: Short-Term-based Volatility-controlled Policy Search and its Global Convergence

It remains challenging to deploy existing risk-averse approaches to real-world applications. The reasons are multi-fold, including the lack of global optimality guarantee and the necessity of learning from long-term consecutive trajectories. Long-term consecutive trajectories are prone to involving visiting hazardous states, which is a major concern in the risk-averse setting. This paper proposes Short-Term VOlatility-controlled Policy Search (STOPS), a novel algorithm that solves risk-averse problems by learning from short-term trajectories instead of long-term trajectories. Short-term trajectories are more flexible to generate, and can avoid the danger of hazardous state visitations. By using an actor-critic scheme with an overparameterized two-layer neural network, our algorithm finds a globally optimal policy at a sublinear rate with proximal policy optimization and natural policy gradient, with effectiveness comparable to the state-of-the-art convergence rate of risk-neutral policy-search methods. The algorithm is evaluated on challenging Mujoco robot simulation tasks under the mean-variance evaluation metric. Both theoretical analysis and experimental results demonstrate a state-of-the-art level of STOPS' performance among existing risk-averse policy search methods.

cs.LG↗

arXiv2021

Data analytics accelerates the experimental discovery of new thermoelectric materials with extremely high figure of merit

Thermoelectric (TE) materials are among very few sustainable yet feasible energy solutions of present time. This huge promise of energy harvesting is contingent on identifying/designing materials having higher efficiency than presently available ones. However, due to the vastness of the chemical space of materials, only its small fraction was scanned experimentally and/or computationally so far. Employing a compressed-sensing based symbolic regression in an active-learning framework, we have not only identified a trend in materials' compositions for superior TE performance, but have also predicted and experimentally synthesized several extremely high performing novel TE materials. Among these, we found Ag$_{0.55}$Cu$_{0.45}$GaTe$_2$ to possess an experimental figure of merit as high as ~2.8 at 827 K, which is a breakthrough in the field. The presented methodology demonstrates the importance and tremendous potential of physically informed descriptors in material science, in particular for relatively small data sets typically available from experiments at well-controlled conditions.

cond-mat.mtrl-sci↗

arXiv2020

Deterministic Data Distribution for Efficient Recovery in Erasure-Coded Storage Systems

Due to individual unreliable commodity components, failures are common in large-scale distributed storage systems. Erasure codes are widely deployed in practical storage systems to provide fault tolerance with low storage overhead. However, random data distribution (RDD), commonly used in erasure-coded storage systems, induces heavy cross-rack traffic, load imbalance, and random access, which adversely affects failure recovery. In this paper, with orthogonal arrays, we define a Deterministic Data Distribution ($D^3$) to uniformly distribute data/parity blocks among nodes, and propose an efficient failure recovery approach based on $D^3$, which minimizes the cross-rack repair traffic against a single node failure. Thanks to the uniformity of $D^3$, the proposed recovery approach balances the repair traffic not only among nodes within a rack but also among racks. We implement $D^3$ over Reed-Solomon codes and Locally Repairable Codes in Hadoop Distributed File System (HDFS) with a cluster of 28 machines. Compared with RDD, our experiments show that $D^3$ significantly speeds up the failure recovery up to 2.49 times for RS codes and 1.38 times for LRCs. Moreover, $D^3$ supports front-end applications better than RDD in both of normal and recovery states.

cs.DC↗

arXiv2018

GeoMFree3D: An Under-Development Meshfree Software Package for Geomechanics

This paper briefly reports the GeoMFree3D, a meshfree / meshless software package designed for analyzing the problems of large deformations and crack propagations of rock and soil masses in geotechnics. The GeoMFree3D is developed based on the meshfree RPIM, and accelerated by exploiting the parallel computing on multi-core CPU and many-core GPU. The GeoMFree3D is currently being under intensive developments. To demonstrate the correctness and effectiveness of the GeoMFree3D, several simple verification examples are presented in this paper. Moreover, future work on the development of the GeoMFree3D is introduced.

physics.comp-ph↗

arXiv2016

Improving GPU-accelerated Adaptive IDW Interpolation Algorithm Using Fast kNN Search

This paper presents an efficient parallel Adaptive Inverse Distance Weighting (AIDW) interpolation algorithm on modern Graphics Processing Unit (GPU). The presented algorithm is an improvement of our previous GPU-accelerated AIDW algorithm by adopting fast k-Nearest Neighbors (kNN) search. In AIDW, it needs to find several nearest neighboring data points for each interpolated point to adaptively determine the power parameter; and then the desired prediction value of the interpolated point is obtained by weighted interpolating using the power parameter. In this work, we develop a fast kNN search approach based on the space-partitioning data structure, even grid, to improve the previous GPU-accelerated AIDW algorithm. The improved algorithm is composed of the stages of kNN search and weighted interpolating. To evaluate the performance of the improved algorithm, we perform five groups of experimental tests. Experimental results show that: (1) the improved algorithm can achieve a speedup of up to 1017 over the corresponding serial algorithm; (2) the improved algorithm is at least two times faster than our previous GPU-accelerated AIDW algorithm; and (3) the utilization of fast kNN search can significantly improve the computational efficiency of the entire GPU-accelerated AIDW algorithm.

cs.DC↗

arXiv2015

Accelerating Adaptive IDW Interpolation Algorithm on a Single GPU

This paper focuses on the design and implementing of GPU-accelerated Adaptive Inverse Distance Weighting (AIDW) interpolation algorithm. The AIDW is an improved version of the standard IDW, which can adaptively determine the power parameter according to the spatial points distribution pattern and achieve more accurate predictions than those by IDW. In this paper, we first present two versions of the GPU accelerated AIDW, the naive version without profiting from shared memory and the tiled version taking advantage of shared memory. We also implement the naive version and the tiled version using the data layouts, Structure of Arrays (AoS) and Array of aligned Structures (AoaS), on single and double precision. We then evaluate the performance of the GPU-accelerated AIDW by comparing it with its original CPU version. Experimental results show that: on single precision the naive version and the tiled version can achieve the speedups of approximately 270 and 400, respectively. In addition, on single precision the implementations using the layout SoA are always slightly faster than those using layout AoaS. However, on double precision, the speedup is only about 8; and we have also observed that: (1) there are no performance gains obtained from the tiled version against the naive version; and (2) the use of SoA and AoaS does not lead to significant differences in computational efficiency.

cs.DC↗