Yonggang Wen

Hongling Zheng, Li Shen, Anke Tang et al.

Foundation models (FM) have demonstrated remarkable performance across a wide range of tasks (especially in the fields of natural language processing and computer vision), primarily attributed to their ability to comprehend instructions and access extensive, high-quality data. This not only showcases their current effectiveness but also sets a promising trajectory towards the development of artificial general intelligence. Unfortunately, due to multiple constraints, the raw data of the model used for large model training are often inaccessible, so the use of end-to-end models for downstream tasks has become a new research trend, which we call Learn From Model (LFM) in this article. LFM focuses on the research, modification, and design of FM based on the model interface, so as to better understand the model structure and weights (in a black box environment), and to generalize the model to downstream tasks. The study of LFM techniques can be broadly categorized into five major areas: model tuning, model distillation, model reuse, meta learning and model editing. Each category encompasses a repertoire of methods and strategies that aim to enhance the capabilities and performance of FM. This paper gives a comprehensive review of the current methods based on FM from the perspective of LFM, in order to help readers better understand the current research status and ideas. To conclude, we summarize the survey by highlighting several critical areas for future exploration and addressing open issues that require further attention from the research community. The relevant papers we investigated in this article can be accessed at https://github.com/ruthless-man/Awesome-Learn-from-Model

Zhenpeng Yao, Yanwei Lum, Andrew Johnston et al.

Transitioning from fossil fuels to renewable energy sources is a critical global challenge; it demands advances at the levels of materials, devices, and systems for the efficient harvesting, storage, conversion, and management of renewable energy. Researchers globally have begun incorporating machine learning (ML) techniques with the aim of accelerating these advances. ML technologies leverage statistical trends in data to build models for prediction of material properties, generation of candidate structures, optimization of processes, among other uses; as a result, they can be incorporated into discovery and development pipelines to accelerate progress. Here we review recent advances in ML-driven energy research, outline current and future challenges, and describe what is required moving forward to best lever ML techniques. To start, we give an overview of key ML concepts. We then introduce a set of key performance indicators to help compare the benefits of different ML-accelerated workflows for energy research. We discuss and evaluate the latest advances in applying ML to the development of energy harvesting (photovoltaics), storage (batteries), conversion (electrocatalysis), and management (smart grids). Finally, we offer an outlook of potential research areas in the energy field that stand to further benefit from the application of ML.

Wei Gao, Qinghao Hu, Zhisheng Ye et al.

Deep learning (DL) shows its prosperity in a wide variety of fields. The development of a DL model is a time-consuming and resource-intensive procedure. Hence, dedicated GPU accelerators have been collectively constructed into a GPU datacenter. An efficient scheduler design for such GPU datacenter is crucially important to reduce the operational cost and improve resource utilization. However, traditional approaches designed for big data or high performance computing workloads can not support DL workloads to fully utilize the GPU resources. Recently, substantial schedulers are proposed to tailor for DL workloads in GPU datacenters. This paper surveys existing research efforts for both training and inference workloads. We primarily present how existing schedulers facilitate the respective workloads from the scheduling objectives and resource consumption features. Finally, we prospect several promising future research directions. More detailed summary with the surveyed paper and code links can be found at our project website: https://github.com/S-Lab-System-Group/Awesome-DL-Scheduling-Papers

Qihuang Zhong, Liang Ding, Yibing Zhan et al.

This technical report briefly describes our JDExplore d-team's Vega v2 submission on the SuperGLUE leaderboard. SuperGLUE is more challenging than the widely used general language understanding evaluation (GLUE) benchmark, containing eight difficult language understanding tasks, including question answering, natural language inference, word sense disambiguation, coreference resolution, and reasoning. [Method] Instead of arbitrarily increasing the size of a pretrained language model (PLM), our aim is to 1) fully extract knowledge from the input pretraining data given a certain parameter budget, e.g., 6B, and 2) effectively transfer this knowledge to downstream tasks. To achieve goal 1), we propose self-evolution learning for PLMs to wisely predict the informative tokens that should be masked, and supervise the masked language modeling (MLM) process with rectified smooth labels. For goal 2), we leverage the prompt transfer technique to improve the low-resource tasks by transferring the knowledge from the foundation model and related downstream tasks to the target task. [Results] According to our submission record (Oct. 2022), with our optimized pretraining and fine-tuning strategies, our 6B Vega method achieved new state-of-the-art performance on 4/8 tasks, sitting atop the SuperGLUE leaderboard on Oct. 8, 2022, with an average score of 91.3.

Shenggui Li, Chao Wang, Yikai Zhu et al.

Large language models incur high inference latency due to sequential autoregressive decoding. Speculative decoding alleviates this bottleneck by using a lightweight draft model to propose multiple tokens for batched verification. However, its adoption has been limited by the lack of high-quality draft models and scalable training infrastructure. We introduce SpecForge, an open-source, production-oriented framework for training speculative decoding models with full support for EAGLE-3. SpecForge incorporates target-draft decoupling, hybrid parallelism, optimized training kernels, and integration with production-grade inference engines, enabling up to 9.9x faster EAGLE-3 training for Qwen3-235B-A22B. In addition, we release SpecBundle, a suite of production-grade EAGLE-3 draft models trained with SpecForge for mainstream open-source LLMs. Through a systematic study of speculative decoding training recipes, SpecBundle addresses the scarcity of high-quality drafts in the community, and our draft models achieve up to 4.48x end-to-end inference speedup on SGLang, establishing SpecForge as a practical foundation for real-world speculative decoding deployment.

Rui Xu, Yong Luo, Han Hu et al.

Weakly supervised object localization (WSOL) is one of the most popular and challenging tasks in computer vision. This task is to localize the objects in the images given only the image-level supervision. Recently, dividing WSOL into two parts (class-agnostic object localization and object classification) has become the state-of-the-art pipeline for this task. However, existing solutions under this pipeline usually suffer from the following drawbacks: 1) they are not flexible since they can only localize one object for each image due to the adopted single-class regression (SCR) for localization; 2) the generated pseudo bounding boxes may be noisy, but the negative impact of such noise is not well addressed. To remedy these drawbacks, we first propose to replace SCR with a binary-class detector (BCD) for localizing multiple objects, where the detector is trained by discriminating the foreground and background. Then we design a weighted entropy (WE) loss using the unlabeled data to reduce the negative impact of noisy bounding boxes. Extensive experiments on the popular CUB-200-2011 and ImageNet-1K datasets demonstrate the effectiveness of our method.

Weiming Zhuang, Xin Gan, Yonggang Wen et al.

Given labeled data in a source domain, unsupervised domain adaptation has been widely adopted to generalize models for unlabeled data in a target domain, whose data distributions are different. However, existing works are inapplicable to face recognition under privacy constraints because they require sharing of sensitive face images between domains. To address this problem, we propose federated unsupervised domain adaptation for face recognition, FedFR. FedFR jointly optimizes clustering-based domain adaptation and federated learning to elevate performance on the target domain. Specifically, for unlabeled data in the target domain, we enhance a clustering algorithm with distance constrain to improve the quality of predicted pseudo labels. Besides, we propose a new domain constraint loss (DCL) to regularize source domain training in federated learning. Extensive experiments on a newly constructed benchmark demonstrate that FedFR outperforms the baseline and classic methods on the target domain by 3% to 14% on different evaluation metrics.

Weiming Zhuang, Yonggang Wen, Shuai Zhang

Self-supervised learning (SSL) is capable of learning remarkable representations from centrally available data. Recent works further implement federated learning with SSL to learn from rapidly growing decentralized unlabeled images (e.g., from cameras and phones), often resulted from privacy constraints. Extensive attention has been paid to SSL approaches based on Siamese networks. However, such an effort has not yet revealed deep insights into various fundamental building blocks for the federated self-supervised learning (FedSSL) architecture. We aim to fill in this gap via in-depth empirical study and propose a new method to tackle the non-independently and identically distributed (non-IID) data problem of decentralized data. Firstly, we introduce a generalized FedSSL framework that embraces existing SSL methods based on Siamese networks and presents flexibility catering to future methods. In this framework, a server coordinates multiple clients to conduct SSL training and periodically updates local models of clients with the aggregated global model. Using the framework, our study uncovers unique insights of FedSSL: 1) stop-gradient operation, previously reported to be essential, is not always necessary in FedSSL; 2) retaining local knowledge of clients in FedSSL is particularly beneficial for non-IID data. Inspired by the insights, we then propose a new approach for model update, Federated Divergence-aware Exponential Moving Average update (FedEMA). FedEMA updates local models of clients adaptively using EMA of the global model, where the decay rate is dynamically measured by model divergence. Extensive experiments demonstrate that FedEMA outperforms existing methods by 3-4% on linear evaluation. We hope that this work will provide useful insights for future research.

Weiming Zhuang, Yonggang Wen, Lingjuan Lyu et al.

Federated learning (FL) is an emerging distributed machine learning method that empowers in-situ model training on decentralized edge devices. However, multiple simultaneous FL tasks could overload resource-constrained devices. In this work, we propose the first FL system to effectively coordinate and train multiple simultaneous FL tasks. We first formalize the problem of training simultaneous FL tasks. Then, we present our new approach, MAS (Merge and Split), to optimize the performance of training multiple simultaneous FL tasks. MAS starts by merging FL tasks into an all-in-one FL task with a multi-task architecture. After training for a few rounds, MAS splits the all-in-one FL task into two or more FL tasks by using the affinities among tasks measured during the all-in-one training. It then continues training each split of FL tasks based on model parameters from the all-in-one training. Extensive experiments demonstrate that MAS outperforms other methods while reducing training time by 2x and reducing energy consumption by 40%. We hope this work will inspire the community to further study and optimize training simultaneous FL tasks.

Dui Wang, Li Shen, Yong Luo et al.

Federated learning aims to collaboratively train models without accessing their client's local private data. The data may be Non-IID for different clients and thus resulting in poor performance. Recently, personalized federated learning (PFL) has achieved great success in handling Non-IID data by enforcing regularization in local optimization or improving the model aggregation scheme on the server. However, most of the PFL approaches do not take into account the unfair competition issue caused by the imbalanced data distribution and lack of positive samples for some classes in each client. To address this issue, we propose a novel and generic PFL framework termed Federated Averaging via Binary Classification, dubbed FedABC. In particular, we adopt the ``one-vs-all'' training strategy in each client to alleviate the unfair competition between classes by constructing a personalized binary classification problem for each class. This may aggravate the class imbalance challenge and thus a novel personalized binary classification loss that incorporates both the under-sampling and hard sample mining strategies is designed. Extensive experiments are conducted on two popular datasets under different settings, and the results demonstrate that our FedABC can significantly outperform the existing counterparts.

Mengya Han, Heliang Zheng, Chaoyue Wang et al.

In this work, we address the task of few-shot part segmentation, which aims to segment the different parts of an unseen object using very few labeled examples. It is found that leveraging the textual space of a powerful pre-trained image-language model (such as CLIP) can be beneficial in learning visual features. Therefore, we develop a novel method termed PartSeg for few-shot part segmentation based on multimodal learning. Specifically, we design a part-aware prompt learning method to generate part-specific prompts that enable the CLIP model to better understand the concept of ``part'' and fully utilize its textual space. Furthermore, since the concept of the same part under different object categories is general, we establish relationships between these parts during the prompt learning process. We conduct extensive experiments on the PartImageNet and Pascal$\_$Part datasets, and the experimental results demonstrated that our proposed method achieves state-of-the-art performance.

Mengya Han, Yibing Zhan, Yong Luo et al.

Few-shot visual recognition refers to recognize novel visual concepts from a few labeled instances. Many few-shot visual recognition methods adopt the metric-based meta-learning paradigm by comparing the query representation with class representations to predict the category of query instance. However, current metric-based methods generally treat all instances equally and consequently often obtain biased class representation, considering not all instances are equally significant when summarizing the instance-level representations for the class-level representation. For example, some instances may contain unrepresentative information, such as too much background and information of unrelated concepts, which skew the results. To address the above issues, we propose a novel metric-based meta-learning framework termed instance-adaptive class representation learning network (ICRL-Net) for few-shot visual recognition. Specifically, we develop an adaptive instance revaluing network with the capability to address the biased representation issue when generating the class representation, by learning and assigning adaptive weights for different instances according to their relative significance in the support set of corresponding class. Additionally, we design an improved bilinear instance representation and incorporate two novel structural losses, i.e., intra-class instance clustering loss and inter-class representation distinguishing loss, to further regulate the instance revaluation process and refine the class representation. We conduct extensive experiments on four commonly adopted few-shot benchmarks: miniImageNet, tieredImageNet, CIFAR-FS, and FC100 datasets. The experimental results compared with the state-of-the-art approaches demonstrate the superiority of our ICRL-Net.

Lizhuo Luo, Shenggui Li, Yonggang Wen et al.

Diffusion large language models (dLLMs) have emerged as a promising alternative for text generation, distinguished by their native support for parallel decoding. In practice, block inference is crucial for avoiding order misalignment in global bidirectional decoding and improving output quality. However, the widely-used fixed, predefined block (naive) schedule is agnostic to semantic difficulty, making it a suboptimal strategy for both quality and efficiency: it can force premature commitments to uncertain positions while delaying easy positions near block boundaries. In this work, we analyze the limitations of naive block scheduling and disclose the importance of dynamically adapting the schedule to semantic difficulty for reliable and efficient inference. Motivated by this, we propose Dynamic Sliding Block (DSB), a training-free block scheduling method that uses a sliding block with a dynamic size to overcome the rigidity of the naive block. To further improve efficiency, we introduce DSB Cache, a training-free KV-cache mechanism tailored to DSB. Extensive experiments across multiple models and benchmarks demonstrate that DSB, together with DSB Cache, consistently improves both generation quality and inference efficiency for dLLMs. Code is released at https://github.com/lizhuo-luo/DSB.

Weiming Zhuang, Xin Gan, Yonggang Wen et al.

The increasingly stringent data privacy regulations limit the development of person re-identification (ReID) because person ReID training requires centralizing an enormous amount of data that contains sensitive personal information. To address this problem, we introduce federated person re-identification (FedReID) -- implementing federated learning, an emerging distributed training method, to person ReID. FedReID preserves data privacy by aggregating model updates, instead of raw data, from clients to a central server. Furthermore, we optimize the performance of FedReID under statistical heterogeneity via benchmark analysis. We first construct a benchmark with an enhanced algorithm, two architectures, and nine person ReID datasets with large variances to simulate the real-world statistical heterogeneity. The benchmark results present insights and bottlenecks of FedReID under statistical heterogeneity, including challenges in convergence and poor performance on datasets with large volumes. Based on these insights, we propose three optimization approaches: (1) We adopt knowledge distillation to facilitate the convergence of FedReID by better transferring knowledge from clients to the server; (2) We introduce client clustering to improve the performance of large datasets by aggregating clients with similar data distributions; (3) We propose cosine distance weight to elevate performance by dynamically updating the weights for aggregation depending on how well models are trained in clients. Extensive experiments demonstrate that these approaches achieve satisfying convergence with much better performance on all datasets. We believe that FedReID will shed light on implementing and optimizing federated learning on more computer vision applications.

Meng Zhang, Qinghao Hu, Peng Sun et al.

Training Graph Neural Networks (GNNs) on large graphs is challenging due to the conflict between the high memory demand and limited GPU memory. Recently, distributed full-graph GNN training has been widely adopted to tackle this problem. However, the substantial inter-GPU communication overhead can cause severe throughput degradation. Existing communication compression techniques mainly focus on traditional DNN training, whose bottleneck lies in synchronizing gradients and parameters. We find they do not work well in distributed GNN training as the barrier is the layer-wise communication of features during the forward pass & feature gradients during the backward pass. To this end, we propose an efficient distributed GNN training framework Sylvie, which employs one-bit quantization technique in GNNs and further pipelines the curtailed communication with computation to enormously shrink the overhead while maintaining the model quality. In detail, Sylvie provides a lightweight Low-bit Module to quantize the sent data and dequantize the received data back to full precision values in each layer. Additionally, we propose a Bounded Staleness Adaptor to control the introduced staleness to achieve further performance enhancement. We conduct theoretical convergence analysis and extensive experiments on various models & datasets to demonstrate Sylvie can considerably boost the training throughput by up to 28.1x.

Meng Zhang, Jie Sun, Qinghao Hu et al.

Graph Transformer is a new architecture that surpasses GNNs in graph learning. While there emerge inspiring algorithm advancements, their practical adoption is still limited, particularly on real-world graphs involving up to millions of nodes. We observe existing graph transformers fail on large-scale graphs mainly due to heavy computation, limited scalability and inferior model quality. Motivated by these observations, we propose TorchGT, the first efficient, scalable, and accurate graph transformer training system. TorchGT optimizes training at different levels. At algorithm level, by harnessing the graph sparsity, TorchGT introduces a Dual-interleaved Attention which is computation-efficient and accuracy-maintained. At runtime level, TorchGT scales training across workers with a communication-light Cluster-aware Graph Parallelism. At kernel level, an Elastic Computation Reformation further optimizes the computation by reducing memory access latency in a dynamic way. Extensive experiments demonstrate that TorchGT boosts training by up to 62.7x and supports graph sequence lengths of up to 1M.

Weiming Zhuang, Yonggang Wen, Shuai Zhang

Federated learning (FL) is an emerging distributed machine learning method that empowers in-situ model training on decentralized edge devices. However, multiple simultaneous training activities could overload resource-constrained devices. In this work, we propose a smart multi-tenant FL system, MuFL, to effectively coordinate and execute simultaneous training activities. We first formalize the problem of multi-tenant FL, define multi-tenant FL scenarios, and introduce a vanilla multi-tenant FL system that trains activities sequentially to form baselines. Then, we propose two approaches to optimize multi-tenant FL: 1) activity consolidation merges training activities into one activity with a multi-task architecture; 2) after training it for rounds, activity splitting divides it into groups by employing affinities among activities such that activities within a group have better synergy. Extensive experiments demonstrate that MuFL outperforms other methods while consuming 40% less energy. We hope this work will inspire the community to further study and optimize multi-tenant FL.

Ziyi Zhang, Sen Zhang, Yibing Zhan et al.

Bridging the gap between diffusion models and human preferences is crucial for their integration into practical generative workflows. While optimizing downstream reward models has emerged as a promising alignment strategy, concerns arise regarding the risk of excessive optimization with learned reward models, which potentially compromises ground-truth performance. In this work, we confront the reward overoptimization problem in diffusion model alignment through the lenses of both inductive and primacy biases. We first identify a mismatch between current methods and the temporal inductive bias inherent in the multi-step denoising process of diffusion models, as a potential source of reward overoptimization. Then, we surprisingly discover that dormant neurons in our critic model act as a regularization against reward overoptimization while active neurons reflect primacy bias. Motivated by these observations, we propose Temporal Diffusion Policy Optimization with critic active neuron Reset (TDPO-R), a policy gradient algorithm that exploits the temporal inductive bias of diffusion models and mitigates the primacy bias stemming from active neurons. Empirical results demonstrate the superior efficacy of our methods in mitigating reward overoptimization. Code is avaliable at https://github.com/ZiyiZhang27/tdpo.

Xinpeng Yang, Meng Hao, Chenkai Weng et al.

Private set intersection (PSI) enables a sender holding a set $Q$ of size $m$ and a receiver holding a set $W$ of size $n$ to securely compute the intersection $Q \cap W$. Fuzzy PSI (FPSI) is a PSI variant where the receiver learns the items $q \in Q$ for which there exists some $w \in W$ satisfying $\mathsf{dist}(q, w) \le δ$ under a given distance metric. Although several FPSI works are proposed for $L_{p}$ distance metrics with $p \in [1, \infty]$, they either heavily rely on expensive homomorphic encryptions, or incur undesirable complexity, e.g., exponential to the element dimension, both of which lead to poor practical efficiency. In this work, we propose efficient FPSI protocols for $L_{p \in [1, \infty]}$ distance metrics, primarily leveraging significantly cheaper symmetric-key operations. Our protocols achieve linear communication and computation complexity in the set sizes $m,n$, the dimension $d$, and the distance threshold $δ$. Our core building block is an oblivious programmable PRF with secret-shared outputs, which may be of independent interest. Furthermore, we incorporate a prefix technique that reduces the dependence on the distance threshold $δ$ to logarithmic, which is particularly suitable for large $δ$. We implement our FPSI protocols and compare them with state-of-the-art constructions. Experimental results demonstrate that our protocols consistently and significantly outperform existing works across all settings. Specifically, our protocols achieve a speedup of $12{\sim}145\times$ in running time and a reduction of $3{\sim}8\times$ in communication cost compared to Gao et al.~(ASIACRYPT'24) and a speedup of $9{\sim}80\times$ in running time and a reduction of $5{\sim}19\times$ in communication cost compared to Dang et al.~(CCS'25).

Shuai Wang, Yibing Zhan, Yong Luo et al.

Incremental learning is nontrivial due to severe catastrophic forgetting. Although storing a small amount of data on old tasks during incremental learning is a feasible solution, current strategies still do not 1) adequately address the class bias problem, and 2) alleviate the mutual interference between new and old tasks, and 3) consider the problem of class bias within tasks. This motivates us to propose a joint input and output coordination (JIOC) mechanism to address these issues. This mechanism assigns different weights to different categories of data according to the gradient of the output score, and uses knowledge distillation (KD) to reduce the mutual interference between the outputs of old and new tasks. The proposed mechanism is general and flexible, and can be incorporated into different incremental learning approaches that use memory storage. Extensive experiments show that our mechanism can significantly improve their performance.

Liangtao Lin, Zhaomeng Zhu, Tianwei Zhang et al.

Standard Operating Procedures (SOPs) are essential for ensuring operational safety and consistency in industrial environments. However, retrieving and following these procedures presents unique challenges, such as rigid proprietary structures, condition-dependent relevance, and actionable execution requirement, which standard semantic-driven Retrieval-Augmented Generation (RAG) paradigms fail to address. Inspired by the Mixture-of-Experts (MoE) paradigm, we propose SOPRAG, a novel framework specifically designed to address the above pain points in SOP retrieval. SOPRAG replaces flat chunking with specialized Entity, Causal, and Flow graph experts to resolve industrial structural and logical complexities. To optimize and coordinate these experts, we propose a Procedure Card layer that prunes the search space to eliminate computational noise, and an LLM-Guided gating mechanism that dynamically weights these experts to align retrieval with operator intent. To address the scarcity of domain-specific data, we also introduce an automated, multi-agent workflow for benchmark construction. Extensive experiments across four industrial domains demonstrate that SOPRAG significantly outperforms strong lexical, dense, and graph-based RAG baselines in both retrieval accuracy and response utility, achieving perfect execution scores in real-world critical tasks.

Qiaoling Chen, Zijun Liu, Peng Sun et al.

Adapting large language models (LLMs) via reinforcement learning (RL) is often bottlenecked by the generation stage, which can consume over 75\% of the training time. Speculative decoding (SD) accelerates autoregressive generation in serving systems, but its behavior under RL training remains largely unexplored. We identify three critical gaps that hinder the naive integration of SD into RL systems: diminishing speedups at large batch sizes, drafter staleness under continual actor updates, and drafter-induced policy degradation. To address these gaps, we present ReSpec, a system that adapts SD to RL through three complementary mechanisms: dynamically tuning SD configurations, evolving the drafter via knowledge distillation, and weighting updates by rollout rewards. On Qwen models (3B--14B), ReSpec achieves up to 4.5x speedup while preserving reward convergence and training stability, providing a practical solution for efficient RL-based LLM adaptation.

Zhiwei Hao, Jianyuan Guo, Li Shen et al.

Large language models (LLMs) have achieved impressive performance across various domains. However, the substantial hardware resources required for their training present a significant barrier to efficiency and scalability. To mitigate this challenge, low-precision training techniques have been widely adopted, leading to notable advancements in training efficiency. Despite these gains, low-precision training involves several components$\unicode{x2013}$such as weights, activations, and gradients$\unicode{x2013}$each of which can be represented in different numerical formats. The resulting diversity has created a fragmented landscape in low-precision training research, making it difficult for researchers to gain a unified overview of the field. This survey provides a comprehensive review of existing low-precision training methods. To systematically organize these approaches, we categorize them into three primary groups based on their underlying numerical formats, which is a key factor influencing hardware compatibility, computational efficiency, and ease of reference for readers. The categories are: (1) fixed-point and integer-based methods, (2) floating-point-based methods, and (3) customized format-based methods. Additionally, we discuss quantization-aware training approaches, which share key similarities with low-precision training during forward propagation. Finally, we highlight several promising research directions to advance this field. A collection of papers discussed in this survey is provided in https://github.com/Hao840/Awesome-Low-Precision-Training.

Runyi Hu, Jie Zhang, Shiqian Zhao et al.

Token-level steering has emerged as a pivotal approach for inference-time alignment, enabling fine grained control over large language models by modulating their output distributions without parameter updates. While effective, existing methods rely on dense intervention at every decoding step. This persistent manipulation not only incurs substantial computational overhead but also risks compromising generation quality by excessively drifting from the model's intrinsic distribution. In this work, we show that dense intervention is unnecessary and propose Sparse Inference time Alignment (SIA), which performs sparse junction steering by intervening only at critical decision points along the generation trajectory. Our key insight is that high entropy junctions mark pivotal decision points in the generation trajectory and are particularly susceptible to misalignment, indicating the need to introduce alignment related reward signals at these points. Extensive experiments across different model families and alignment objectives show that steering only 20% to 80% of tokens achieves superior alignment-efficiency trade offs. For strong base models such as Qwen3, intervening on as few as 20% of tokens matches or even surpasses heavily post-trained instruct models. This sparsity enables stronger guidance while better preserving the model's native distribution, integrates seamlessly with search based methods such as Best-of-N, and reduces computational cost by up to 6x.

Zhiwei Hao, Jianyuan Guo, Li Shen et al.

Recent advancements in multimodal fusion have witnessed the remarkable success of vision-language (VL) models, which excel in various multimodal applications such as image captioning and visual question answering. However, building VL models requires substantial hardware resources, where efficiency is restricted by two key factors: the extended input sequence of the language model with vision features demands more computational operations, and a large number of additional learnable parameters increase memory complexity. These challenges significantly restrict the broader applicability of such models. To bridge this gap, we propose ADEM-VL, an efficient vision-language method that tunes VL models based on pretrained large language models (LLMs) by adopting a parameter-free cross-attention mechanism for similarity measurements in multimodal fusion. This approach only requires embedding vision features into the language space, significantly reducing the number of trainable parameters and accelerating both training and inference speeds. To enhance representation learning in fusion module, we introduce an efficient multiscale feature generation scheme that requires only a single forward pass through the vision encoder. Moreover, we propose an adaptive fusion scheme that dynamically discards less relevant visual information for each text token based on its attention score. This ensures that the fusion process prioritizes the most pertinent visual features. With experiments on various tasks including visual question answering, image captioning, and instruction-following, we demonstrate that our framework outperforms existing approaches. Specifically, our method surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset, with reduced training and inference latency, demonstrating the superiority of our framework. The code is available at https://github.com/Hao840/ADEM-VL.

Wei Gao, Xinyu Zhou, Peng Sun et al.

Key-Value cache (\texttt{KV} \texttt{cache}) compression has emerged as a promising technique to optimize Large Language Model (LLM) serving. It primarily decreases the memory consumption of \texttt{KV} \texttt{cache} to reduce the computation cost. Despite the development of many compression algorithms, their applications in production environments are still not prevalent. In this paper, we revisit mainstream \texttt{KV} \texttt{cache} compression solutions from a practical perspective. Our contributions are three-fold. First, we comprehensively review existing algorithmic designs and benchmark studies for \texttt{KV} \texttt{cache} compression and identify missing pieces in their performance measurement, which could hinder their adoption in practice. Second, we empirically evaluate representative \texttt{KV} \texttt{cache} compression methods to uncover two key issues that affect the computational efficiency: (1) while compressing \texttt{KV} \texttt{cache} can reduce memory consumption, current implementations (e.g., FlashAttention, PagedAttention) do not optimize for production-level LLM serving, resulting in suboptimal throughput performance; (2) compressing \texttt{KV} \texttt{cache} may lead to longer outputs, resulting in increased end-to-end latency. We further investigate the accuracy performance of individual samples rather than the overall performance, revealing the intrinsic limitations in \texttt{KV} \texttt{cache} compression when handling specific LLM tasks. Third, we provide tools to shed light on future \texttt{KV} \texttt{cache} compression studies and facilitate their practical deployment in production. They are open-sourced in \href{https://github.com/LLMkvsys/rethink-kv-compression}{https://github.com/LLMkvsys/rethink-kv-compression}.

Minghao Li, Ruihang Wang, Rui Tan et al.

Modern data centers (DCs) hosting artificial intelligence (AI)-dedicated devices operate at high power densities with rapidly varying workloads, making minute-level adaptation essential for safe and energy-efficient operation. However, manually designing piecewise deep reinforcement learning (DRL) agents cannot keep pace with frequent dynamics shifts and service-level agreement (SLA) changes of an evolving DC. This specification-to-policy lag causes a lack of timely, effective control policies, which may lead to service outages. To bridge the gap, we present DCoPilot, a hybrid framework for generative control policies in dynamic DC operation. DCoPilot synergizes two distinct generative paradigms, i.e., a large language model (LLM) that performs symbolic generation of structured reward forms, and a hypernetwork that conducts parametric generation of policy weights. DCoPilot operates through three coordinated phases: (i) simulation scale-up, which stress-tests reward candidates across diverse simulation-ready (SimReady) scenes; (ii) meta policy distillation, where a hypernetwork is trained to output policy weights conditioned on SLA and scene embeddings; and (iii) online adaptation, enabling zero-shot policy generation in response to updated specifications. Evaluated across five control task families spanning diverse DC components, DCoPilot achieves near-zero constraint violations and outperforms all baselines across specification variations. Ablation studies validate the effectiveness of LLM-based unified reward generation in enabling stable hypernetwork convergence.

Liangtao Lin, Zhaomeng Zhu, Tianwei Zhang et al.

Mission-critical industrial infrastructure, such as data centers, increasingly depends on complex management software. Its operations, however, pose significant challenges due to the escalating system complexity, multi-vendor integration, and a shortage of expert operators. While Robotic Process Automation (RPA) offers partial automation through handcrafted scripts, it suffers from limited flexibility and high maintenance costs. Recent advances in Large Language Model (LLM)-based graphical user interface (GUI) agents have enabled more flexible automation, yet these general-purpose agents face five critical challenges when applied to industrial management, including unfamiliar element understanding, precision and efficiency, state localization, deployment constraints, and safety requirements. To address these issues, we propose InfraMind, a novel exploration-based GUI agentic framework specifically tailored for industrial management systems. InfraMind integrates five innovative modules to systematically resolve different challenges in industrial management: (1) systematic search-based exploration with virtual machine snapshots for autonomous understanding of complex GUIs; (2) memory-driven planning to ensure high-precision and efficient task execution; (3) advanced state identification for robust localization in hierarchical interfaces; (4) structured knowledge distillation for efficient deployment with lightweight models; and (5) comprehensive, multi-layered safety mechanisms to safeguard sensitive operations. Extensive experiments on both open-source and commercial DCIM platforms demonstrate that our approach consistently outperforms existing frameworks in terms of task success rate and operational efficiency, providing a rigorous and scalable solution for industrial management automation.

Yixin Cao, Xiang Ji, Xin Lv et al.

We present InferWiki, a Knowledge Graph Completion (KGC) dataset that improves upon existing benchmarks in inferential ability, assumptions, and patterns. First, each testing sample is predictable with supportive data in the training set. To ensure it, we propose to utilize rule-guided train/test generation, instead of conventional random split. Second, InferWiki initiates the evaluation following the open-world assumption and improves the inferential difficulty of the closed-world assumption, by providing manually annotated negative and unknown triples. Third, we include various inference patterns (e.g., reasoning path length and types) for comprehensive evaluation. In experiments, we curate two settings of InferWiki varying in sizes and structures, and apply the construction process on CoDEx as comparative datasets. The results and empirical analyses demonstrate the necessity and high-quality of InferWiki. Nevertheless, the performance gap among various inferential assumptions and patterns presents the difficulty and inspires future research direction. Our datasets can be found in https://github.com/TaoMiner/inferwiki

Wen Wang, Yang Cao, Jing Zhang et al.

Detection transformers have recently shown promising object detection results and attracted increasing attention. However, how to develop effective domain adaptation techniques to improve its cross-domain performance remains unexplored and unclear. In this paper, we delve into this topic and empirically find that direct feature distribution alignment on the CNN backbone only brings limited improvements, as it does not guarantee domain-invariant sequence features in the transformer for prediction. To address this issue, we propose a novel Sequence Feature Alignment (SFA) method that is specially designed for the adaptation of detection transformers. Technically, SFA consists of a domain query-based feature alignment (DQFA) module and a token-wise feature alignment (TDA) module. In DQFA, a novel domain query is used to aggregate and align global context from the token sequence of both domains. DQFA reduces the domain discrepancy in global feature representations and object relations when deploying in the transformer encoder and decoder, respectively. Meanwhile, TDA aligns token features in the sequence from both domains, which reduces the domain gaps in local and instance-level feature representations in the transformer encoder and decoder, respectively. Besides, a novel bipartite matching consistency loss is proposed to enhance the feature discriminability for robust object detection. Experiments on three challenging benchmarks show that SFA outperforms state-of-the-art domain adaptive object detection methods. Code has been made available at: https://github.com/encounter1997/SFA.

Huaizheng Zhang, Yuanming Li, Qiming Ai et al.

Combining \underline{v}ideo streaming and online \underline{r}etailing (V2R) has been a growing trend recently. In this paper, we provide practitioners and researchers in multimedia with a cloud-based platform named Hysia for easy development and deployment of V2R applications. The system consists of: 1) a back-end infrastructure providing optimized V2R related services including data engine, model repository, model serving and content matching; and 2) an application layer which enables rapid V2R application prototyping. Hysia addresses industry and academic needs in large-scale multimedia by: 1) seamlessly integrating state-of-the-art libraries including NVIDIA video SDK, Facebook faiss, and gRPC; 2) efficiently utilizing GPU computation; and 3) allowing developers to bind new models easily to meet the rapidly changing deep learning (DL) techniques. On top of that, we implement an orchestrator for further optimizing DL model serving performance. Hysia has been released as an open source project on GitHub, and attracted considerable attention. We have published Hysia to DockerHub as an official image for seamless integration and deployment in current cloud environments.

Huaizheng Zhang, Yuanming Li, Yizheng Huang et al.

MLModelCI provides multimedia researchers and developers with a one-stop platform for efficient machine learning (ML) services. The system leverages DevOps techniques to optimize, test, and manage models. It also containerizes and deploys these optimized and validated models as cloud services (MLaaS). In its essence, MLModelCI serves as a housekeeper to help users publish models. The models are first automatically converted to optimized formats for production purpose and then profiled under different settings (e.g., batch size and hardware). The profiling information can be used as guidelines for balancing the trade-off between performance and cost of MLaaS. Finally, the system dockerizes the models for ease of deployment to cloud environments. A key feature of MLModelCI is the implementation of a controller, which allows elastic evaluation which only utilizes idle workers while maintaining online service quality. Our system bridges the gap between current ML training and serving systems and thus free developers from manual and tedious work often associated with service deployment. We release the platform as an open-source project on GitHub under Apache 2.0 license, with the aim that it will facilitate and streamline more large-scale ML applications and research projects.

Linsen Dong, Guanyu Gao, Xinyi Zhang et al.

Model-Based Reinforcement Learning (MBRL) is one category of Reinforcement Learning (RL) algorithms which can improve sampling efficiency by modeling and approximating system dynamics. It has been widely adopted in the research of robotics, autonomous driving, etc. Despite its popularity, there still lacks some sophisticated and reusable open-source frameworks to facilitate MBRL research and experiments. To fill this gap, we develop a flexible and modularized framework, Baconian, which allows researchers to easily implement a MBRL testbed by customizing or building upon our provided modules and algorithms. Our framework can free users from re-implementing popular MBRL algorithms from scratch thus greatly save users' efforts on MBRL experiments.

Huaizheng Zhang, Han Hu, Guanyu Gao et al.

Motivated by the prowess of deep learning (DL) based techniques in prediction, generalization, and representation learning, we develop a novel framework called DeepQoE to predict video quality of experience (QoE). The end-to-end framework first uses a combination of DL techniques (e.g., word embeddings) to extract generalized features. Next, these features are combined and fed into a neural network for representation learning. Such representations serve as inputs for classification or regression tasks. Evaluating the performance of DeepQoE with two datasets, we show that for the small dataset, the accuracy of all shallow learning algorithm is improved by using the representation derived from DeepQoE. For the large dataset, our DeepQoE framework achieves significant performance improvement in comparison to the best baseline method (90.94% vs. 82.84%). Moreover, DeepQoE, also released as an open source tool, provides video QoE research much-needed flexibility in fitting different datasets, extracting generalized features, and learning representations.

Guanyu Gao, Yonggang Wen, Cedric Westphal

Adaptive bitrate streaming (ABR) has been widely adopted to support video streaming services over heterogeneous devices and varying network conditions. With ABR, each video content is transcoded into multiple representations in different bitrates and resolutions. However, video transcoding is computing intensive, which requires the transcoding service providers to deploy a large number of servers for transcoding the video contents published by the content producers. As such, a natural question for the transcoding service provider is how to provision the computing resource for transcoding the video contents while maximizing service profit. To address this problem, we design a cloud video transcoding system by taking the advantage of cloud computing technology to elastically allocate computing resource. We propose a method for jointly considering the task scheduling and resource provisioning problem in two timescales, and formulate the service profit maximization as a two-timescale stochastic optimization problem. We derive some approximate policies for the task scheduling and resource provisioning. Based on our proposed methods, we implement our open source cloud video transcoding system Morph and evaluate its performance in a real environment. The experiment results demonstrate that our proposed method can reduce the resource consumption and achieve a higher profit compared with the baseline schemes.

Qinghao Hu, Zhisheng Ye, Zerui Wang et al.

Large Language Models (LLMs) have presented impressive performance across several transformative tasks. However, it is non-trivial to efficiently utilize large-scale cluster resources to develop LLMs, often riddled with numerous challenges such as frequent hardware failures, intricate parallelization strategies, and imbalanced resource utilization. In this paper, we present an in-depth characterization study of a six-month LLM development workload trace collected from our GPU datacenter Acme. Specifically, we investigate discrepancies between LLMs and prior task-specific Deep Learning (DL) workloads, explore resource utilization patterns, and identify the impact of various job failures. Our analysis summarizes hurdles we encountered and uncovers potential opportunities to optimize systems tailored for LLMs. Furthermore, we introduce our system efforts: (1) fault-tolerant pretraining, which enhances fault tolerance through LLM-involved failure diagnosis and automatic recovery. (2) decoupled scheduling for evaluation, which achieves timely performance feedback via trial decomposition and scheduling optimization.

Xuming An, Dui Wang, Li Shen et al.

Federated learning aims to collaboratively learn a model by using the data from multiple users under privacy constraints. In this paper, we study the multi-label classification problem under the federated learning setting, where trivial solution and extremely poor performance may be obtained, especially when only positive data w.r.t. a single class label are provided for each client. This issue can be addressed by adding a specially designed regularizer on the server-side. Although effective sometimes, the label correlations are simply ignored and thus sub-optimal performance may be obtained. Besides, it is expensive and unsafe to exchange user's private embeddings between server and clients frequently, especially when training model in the contrastive way. To remedy these drawbacks, we propose a novel and generic method termed Federated Averaging by exploring Label Correlations (FedALC). Specifically, FedALC estimates the label correlations in the class embedding learning for different label pairs and utilizes it to improve the model training. To further improve the safety and also reduce the communication overhead, we propose a variant to learn fixed class embedding for each client, so that the server and clients only need to exchange class embeddings once. Extensive experiments on multiple popular datasets demonstrate that our FedALC can significantly outperform existing counterparts.

Qingang Zhang, Yuejun Yan, Guangyu Wu et al.

The growing scale and complexity of modern data centers present major challenges in balancing energy efficiency with outage risk. Although Deep Reinforcement Learning (DRL) shows strong potential for intelligent control, its deployment in mission-critical systems is limited by data scarcity and the lack of real-time pre-evaluation mechanisms. This paper introduces the Dual-Loop Control Framework (DLCF), a digital twin-based architecture designed to overcome these challenges. The framework comprises three core entities: the physical system, a digital twin, and a policy reservoir of diverse DRL agents. These components interact through a dual-loop mechanism involving real-time data acquisition, data assimilation, DRL policy training, pre-evaluation, and expert verification. Theoretical analysis shows how DLCF can improve sample efficiency, generalization, safety, and optimality. Leveraging DLCF, we implemented the DCVerse platform and validated it through case studies on a real-world data center cooling system. The evaluation shows that our approach achieves up to 4.09% energy savings over conventional control strategies without violating SLA requirements. Additionally, the framework improves policy interpretability and supports more trustworthy DRL deployment. This work provides a foundation for reliable AI-based control in data centers and points toward future extensions for holistic, system-wide optimization.

Hanwen Zhang, Ruichen Zhang, Wei Zhang et al.

Generative artificial intelligence, particularly through large language models (LLMs), is poised to transform energy optimization and demand side management (DSM) within microgrids. This paper explores the integration of LLMs into energy management, emphasizing their roles in automating the optimization of DSM strategies with Internet of electric vehicles. We investigate challenges and solutions associated with DSM and explore the new opportunities presented by leveraging LLMs. Then, we propose an innovative solution that enhances LLMs with retrieval-augmented generation for automatic problem formulation, code generation, and customizing optimization. We present a case study to demonstrate the effectiveness of our proposed solution in charging scheduling and optimization for electric vehicles, highlighting our solution's significant advancements in energy efficiency and user adaptability. This work underscores the potential of LLMs for energy optimization and fosters a new era of intelligent DSM solutions.

Minghao LI, Ruihang Wang, Rui Tan et al.

Data center (DC) infrastructure serves as the backbone to support the escalating demand for computing capacity. Traditional design methodologies that blend human expertise with specialized simulation tools scale poorly with the increasing system complexity. Recent studies adopt generative artificial intelligence to design plausible human-centric indoor layouts. However, they do not consider the underlying physics, making them unsuitable for the DC design that sets quantifiable operational objectives and strict physical constraints. To bridge the gap, we propose Phythesis, a novel framework that synergizes large language models (LLMs) and physics-guided evolutionary optimization to automate simulation-ready (SimReady) scene synthesis for energy-efficient DC design. Phythesis employs an iterative bi-level optimization architecture, where (i) the LLM-driven optimization level generates physically plausible three-dimensional layouts and self-criticizes them to refine the scene topology, and (ii) the physics-informed optimization level identifies the optimal asset parameters and selects the best asset combination. Experiments on three generation scales show that Phythesis achieves 57.3% generation success rate increase and 11.5% power usage effectiveness (PUE) improvement, compared with the vanilla LLM-based solution.

Vijay Babu Pamshetti, Wei Zhang, Sumei Sun et al.

Battery health prognostics are critical for ensuring safety, efficiency, and sustainability in modern energy systems. However, it has been challenging to achieve accurate and robust prognostics due to complex battery degradation behaviors with nonlinearity, noise, capacity regeneration, etc. Existing data-driven models capture temporal degradation features but often lack knowledge guidance, which leads to unreliable long-term health prognostics. To overcome these limitations, we propose Karma, a knowledge-aware model with frequency-adaptive learning for battery capacity estimation and remaining useful life prediction. The model first performs signal decomposition to derive battery signals in different frequency bands. A dual-stream deep learning architecture is developed, where one stream captures long-term low-frequency degradation trends and the other models high-frequency short-term dynamics. Karma regulates the prognostics with knowledge, where battery degradation is modeled as a double exponential function based on empirical studies. Our dual-stream model is used to optimize the parameters of the knowledge with particle filters to ensure physically consistent and reliable prognostics and uncertainty quantification. Experimental study demonstrates Karma's superior performance, achieving average error reductions of 50.6% and 32.6% over state-of-the-art algorithms for battery health prediction on two mainstream datasets, respectively. These results highlight Karma's robustness, generalizability, and potential for safer and more reliable battery management across diverse applications.

Zerui Wang, Qinghao Hu, Ana Klimovic et al.

Deep learning (DL) schedulers are pivotal in optimizing resource allocation in GPU clusters, but operate with a critical limitation: they are largely blind to the semantic context of the jobs they manage. This forces them to rely on limited metadata, leading to high profiling overhead, unreliable duration estimation, inadequate failure handling, and poor observability. To this end, we propose SchedMate, a framework that bridges this semantic gap by systematically extracting deep insights from overlooked, unstructured data sources: source code, runtime logs, and historical jobs. SchedMate enhances existing schedulers non-intrusively through three LLM-based components. Our implementation integrates seamlessly with existing deep learning schedulers. Evaluations on a 128-GPU physical cluster and extensive simulations on production traces show SchedMate reduces average job completion times by up to 1.91x, substantially enhancing the scheduling performance, demonstrating the critical role of semantic-awareness in modern DL scheduling.

Guanyu Xu, Zhiwei Hao, Li Shen et al.

The impressive performance of transformer models has sparked the deployment of intelligent applications on resource-constrained edge devices. However, ensuring high-quality service for real-time edge systems is a significant challenge due to the considerable computational demands and resource requirements of these models. Existing strategies typically either offload transformer computations to other devices or directly deploy compressed models on individual edge devices. These strategies, however, result in either considerable communication overhead or suboptimal trade-offs between accuracy and efficiency. To tackle these challenges, we propose a collaborative inference system for general transformer models, termed CoFormer. The central idea behind CoFormer is to exploit the divisibility and integrability of transformer. An off-the-shelf large transformer can be decomposed into multiple smaller models for distributed inference, and their intermediate results are aggregated to generate the final output. We formulate an optimization problem to minimize both inference latency and accuracy degradation under heterogeneous hardware constraints. DeBo algorithm is proposed to first solve the optimization problem to derive the decomposition policy, and then progressively calibrate decomposed models to restore performance. We demonstrate the capability to support a wide range of transformer models on heterogeneous edge devices, achieving up to 3.1$\times$ inference speedup with large transformer models. Notably, CoFormer enables the efficient inference of GPT2-XL with 1.6 billion parameters on edge devices, reducing memory requirements by 76.3\%. CoFormer can also reduce energy consumption by approximately 40\% while maintaining satisfactory inference performance.

Ruihang Wang, Minghao Li, Zhiwei Cao et al.

The explosion in artificial intelligence (AI) applications is pushing the development of AI-dedicated data centers (AIDCs), creating management challenges that traditional methods and standalone AI solutions struggle to address. While digital twins are beneficial for AI-based design validation and operational optimization, current AI methods for their creation face limitations. Specifically, physical AI (PhyAI) aims to capture the underlying physical laws, which demands extensive, case-specific customization, and generative AI (GenAI) can produce inaccurate or hallucinated results. We propose Fusion Intelligence, a novel framework synergizing GenAI's automation with PhyAI's domain grounding. In this dual-agent collaboration, GenAI interprets natural language prompts to generate tokenized AIDC digital twins. Subsequently, PhyAI optimizes these generated twins by enforcing physical constraints and assimilating real-time data. Case studies demonstrate the advantages of our framework in automating the creation and validation of AIDC digital twins. These twins deliver predictive analytics to support power usage effectiveness (PUE) optimization in the design stage. With operational data collected, the digital twin accuracy is further improved compared with pure physics-based models developed by human experts. Fusion Intelligence offers a promising pathway to accelerate digital transformation. It enables more reliable and efficient AI-driven digital transformation for a broad range of mission-critical infrastructures.

Ruihang Wang, Zhiwei Cao, Qingang Zhang et al.

Data centers are the backbone of computing capacity. Operating data centers in the tropical regions faces unique challenges due to consistently high ambient temperature and elevated relative humidity throughout the year. These conditions result in increased cooling costs to maintain the reliability of the computing systems. While existing machine learning-based approaches have demonstrated potential to elevate operations to a more proactive and intelligent level, their deployment remains dubious due to concerns about model extrapolation capabilities and associated system safety issues. To address these concerns, this article proposes incorporating the physical characteristics of data centers into traditional data-driven machine learning solutions. We begin by introducing the data center system, including the relevant multiphysics processes and the data-physics availability. Next, we outline the associated modeling and optimization problems and propose an integrated, physics-informed machine learning system to address them. Using the proposed system, we present relevant applications across varying levels of operational intelligence. A case study on an industry-grade tropical data center is provided to demonstrate the effectiveness of our approach. Finally, we discuss key challenges and highlight potential future directions.

Zhiwei Cao, Minghao Li, Feng Lin et al.

Data centers (DCs) as mission-critical infrastructures are pivotal in powering the growth of artificial intelligence (AI) and the digital economy. The evolution from Internet DC to AI DC has introduced new challenges in operating and managing data centers for improved business resilience and reduced total cost of ownership. As a result, new paradigms, beyond the traditional approaches based on best practices, must be in order for future data centers. In this research, we propose and develop a novel Physical AI (PhyAI) framework for advancing DC operations and management. Our system leverages the emerging capabilities of state-of-the-art industrial products and our in-house research and development. Specifically, it presents three core modules, namely: 1) an industry-grade in-house simulation engine to simulate DC operations in a highly accurate manner, 2) an AI engine built upon NVIDIA PhysicsNemo for the training and evaluation of physics-informed machine learning (PIML) models, and 3) a digital twin platform built upon NVIDIA Omniverse for our proposed 5-tier digital twin framework. This system presents a scalable and adaptable solution to digitalize, optimize, and automate future data center operations and management, by enabling real-time digital twins for future data centers. To illustrate its effectiveness, we present a compelling case study on building a surrogate model for predicting the thermal and airflow profiles of a large-scale DC in a real-time manner. Our results demonstrate its superior performance over traditional time-consuming Computational Fluid Dynamics/Heat Transfer (CFD/HT) simulation, with a median absolute temperature prediction error of 0.18 °C. This emerging approach would open doors to several potential research directions for advancing Physical AI in future DC operations.

Qinghao Hu, Peng Sun, Shengen Yan et al.

Modern GPU datacenters are critical for delivering Deep Learning (DL) models and services in both the research community and industry. When operating a datacenter, optimization of resource scheduling and management can bring significant financial benefits. Achieving this goal requires a deep understanding of the job features and user behaviors. We present a comprehensive study about the characteristics of DL jobs and resource management. First, we perform a large-scale analysis of real-world job traces from SenseTime. We uncover some interesting conclusions from the perspectives of clusters, jobs and users, which can facilitate the cluster system designs. Second, we introduce a general-purpose framework, which manages resources based on historical data. As case studies, we design: a Quasi-Shortest-Service-First scheduling service, which can minimize the cluster-wide average job completion time by up to 6.5x; and a Cluster Energy Saving service, which improves overall cluster utilization by up to 13%.

Weiming Zhuang, Yonggang Wen, Shuai Zhang

Person re-identification (ReID) aims to re-identify a person from non-overlapping camera views. Since person ReID data contains sensitive personal information, researchers have adopted federated learning, an emerging distributed training method, to mitigate the privacy leakage risks. However, existing studies rely on data labels that are laborious and time-consuming to obtain. We present FedUReID, a federated unsupervised person ReID system to learn person ReID models without any labels while preserving privacy. FedUReID enables in-situ model training on edges with unlabeled data. A cloud server aggregates models from edges instead of centralizing raw data to preserve data privacy. Moreover, to tackle the problem that edges vary in data volumes and distributions, we personalize training in edges with joint optimization of cloud and edge. Specifically, we propose personalized epoch to reassign computation throughout training, personalized clustering to iteratively predict suitable labels for unlabeled data, and personalized update to adapt the server aggregated model to each edge. Extensive experiments on eight person ReID datasets demonstrate that FedUReID not only achieves higher accuracy but also reduces computation cost by 29%. Our FedUReID system with the joint optimization will shed light on implementing federated learning to more multimedia tasks without data labels.

Weiming Zhuang, Xin Gan, Yonggang Wen et al.

Unsupervised representation learning has achieved outstanding performances using centralized data available on the Internet. However, the increasing awareness of privacy protection limits sharing of decentralized unlabeled image data that grows explosively in multiple parties (e.g., mobile phones and cameras). As such, a natural problem is how to leverage these data to learn visual representations for downstream tasks while preserving data privacy. To address this problem, we propose a novel federated unsupervised learning framework, FedU. In this framework, each party trains models from unlabeled data independently using contrastive learning with an online network and a target network. Then, a central server aggregates trained models and updates clients' models with the aggregated model. It preserves data privacy as each party only has access to its raw data. Decentralized data among multiple parties are normally non-independent and identically distributed (non-IID), leading to performance degradation. To tackle this challenge, we propose two simple but effective methods: 1) We design the communication protocol to upload only the encoders of online networks for server aggregation and update them with the aggregated encoder; 2) We introduce a new module to dynamically decide how to update predictors based on the divergence caused by non-IID. The predictor is the other component of the online network. Extensive experiments and ablations demonstrate the effectiveness and significance of FedU. It outperforms training with only one party by over 5% and other methods by over 14% in linear and semi-supervised evaluation on non-IID data.

Yizheng Huang, Huaizheng Zhang, Yonggang Wen et al.

MLOps is about taking experimental ML models to production, i.e., serving the models to actual users. Unfortunately, existing ML serving systems do not adequately handle the dynamic environments in which online data diverges from offline training data, resulting in tedious model updating and deployment works. This paper implements a lightweight MLOps plugin, termed ModelCI-e (continuous integration and evolution), to address the issue. Specifically, it embraces continual learning (CL) and ML deployment techniques, providing end-to-end supports for model updating and validation without serving engine customization. ModelCI-e includes 1) a model factory that allows CL researchers to prototype and benchmark CL models with ease, 2) a CL backend to automate and orchestrate the model updating efficiently, and 3) a web interface for an ML team to manage CL service collaboratively. Our preliminary results demonstrate the usability of ModelCI-e, and indicate that eliminating the interference between model updating and inference workloads is crucial for higher system efficiency.