Personalized federated learning (PFL) offers a solution to balancing personalization and generalization by conducting federated learning (FL) to guide personalized learning (PL). Little attention has been given to wireless PFL (WPFL), where privacy concerns arise. Performance fairness of PL models is another challenge resulting from communication bottlenecks in WPFL. This paper exploits quantization errors to enhance the privacy of WPFL and proposes a novel quantization-assisted Gaussian differential privacy (DP) mechanism. We analyze the convergence upper bounds of individual PL models by considering the impact of the mechanism (i.e., quantization errors and Gaussian DP noises) and imperfect communication channels on the FL of WPFL. By minimizing the maximum of the bounds, we design an optimal transmission scheduling strategy that yields min-max fairness for WPFL with OFDMA interfaces. This is achieved by revealing the nested structure of this problem to decouple it into subproblems solved sequentially for the client selection, channel allocation, and power control, and for the learning rates and PL-FL weighting coefficients. Experiments validate our analysis and demonstrate that our approach substantially outperforms alternative scheduling strategies by 87.08%, 16.21%, and 38.37% in accuracy, the maximum test loss of participating clients, and fairness (Jain's index), respectively.
High-Performance Computing (HPC) job scheduling involves balancing conflicting objectives such as minimizing makespan, reducing wait times, optimizing resource use, and ensuring fairness. Traditional methods, including heuristic-based (e.g., First-Come-First-Served) or intensive optimization techniques, often lack adaptability to dynamic workloads and heterogeneous HPC systems. To address this, we propose a novel Large Language Model (LLM)-based scheduler using a ReAct-style framework (Reason + Act), enabling iterative, interpretable decision-making. The system incorporates a scratchpad memory to track scheduling history and refine decisions via natural language feedback, while a constraint enforcement module ensures feasibility and safety. We evaluate our approach using OpenAI's O4-Mini and Anthropic's Claude 3.7 across seven real-world HPC workload scenarios, including heterogeneous mixes, bursty patterns, and adversarial cases. Comparisons against FCFS, Shortest Job First, and Google OR-Tools (on 10 to 100 jobs) reveal that LLM-based scheduling effectively balances multiple objectives while offering transparent reasoning through natural language traces. The method excels in constraint satisfaction and adapts to diverse workloads without domain-specific training. However, a trade-off between reasoning quality and computational overhead challenges real-time deployment. This work presents the first comprehensive study of reasoning-capable LLMs for HPC scheduling, demonstrating their potential to handle multiobjective optimization while highlighting limitations in computational efficiency. The findings provide insights into leveraging advanced language models for complex scheduling problems in dynamic HPC environments.
Efficiently serving large language models (LLMs) under dynamic and bursty workloads remains a key challenge for real-world deployment. Existing serving frameworks and static model compression techniques fail to adapt to workload fluctuations, leading to either service-level objective (SLO) violations under full-precision serving or persistent accuracy degradation with static quantization. We present MorphServe, a dynamic, workload-aware LLM serving framework based on morphological adaptation. MorphServe introduces two asynchronous, token-level runtime mechanisms: quantized layer swapping, which selectively replaces less impactful layers with quantized alternatives during high-load periods, and pressure-aware KV cache resizing, which dynamically adjusts KV cache capacity in response to memory pressure. These mechanisms enable state-preserving transitions with minimum runtime overhead and are fully compatible with modern scheduling and attention techniques. Extensive experiments on Vicuna and Llama family models with real-world workloads demonstrate that MorphServe reduces average SLO violations by 92.45 percent and improves the P95 TTFT latency by 2.2x-3.9x compared to full-precision serving, without compromising generation quality. These results establish MorphServe as a practical and elastic solution for LLM deployment in dynamic environments.
LLMs are used predominantly in synchronous communication, where a human user and a model communicate in alternating turns. In contrast, many real-world settings are inherently asynchronous. For example, in group chats, online team meetings, or social games, there is no inherent notion of turns; therefore, the decision of when to speak forms a crucial part of the participant's decision making. In this work, we develop an adaptive asynchronous LLM-agent which, in addition to determining what to say, also decides when to say it. To evaluate our agent, we collect a unique dataset of online Mafia games, including both human participants, as well as our asynchronous agent. Overall, our agent performs on par with human players, both in game performance, as well as in its ability to blend in with the other human players. Our analysis shows that the agent's behavior in deciding when to speak closely mirrors human patterns, although differences emerge in message content. We release all our data and code to support and encourage further research for more realistic asynchronous communication between LLM agents. This work paves the way for integration of LLMs into realistic human group settings, from assistance in team discussions to educational and professional environments where complex social dynamics must be navigated.
The computational sparsity of Mixture-of-Experts (MoE) models enables sub-linear growth in compute cost as model size increases, thus offering a scalable path to training massive neural networks. However, existing implementations suffer from \emph{low GPU utilization}, \emph{significant latency overhead}, and a fundamental \emph{inability to leverage task locality}, primarily due to CPU-managed scheduling, host-initiated communication, and frequent kernel launches. To overcome these limitations, we develop FlashDMoE, a fully GPU-resident MoE operator that fuses expert computation and inter-GPU communication into a \emph{single persistent GPU kernel}. FlashDMoE enables fine-grained pipelining of dispatch, compute, and combine phases, eliminating launch overheads and reducing idle gaps. Unlike existing work, FlashDMoE obviates bulk-synchronous collectives for one-sided, device-initiated, inter-GPU (R)DMA transfers, thus unlocking \emph{payload efficiency}, where we eliminate bloated or redundant network payloads in sparsely activated layers. When evaluated on a single 8-H100 GPU node with MoE models having up to 128 experts and 16K token sequences, FlashDMoE achieves up to \textbf{9}$\times$ higher GPU utilization, \textbf{6}$\times$ lower latency, \textbf{5.7}$\times$ higher throughput, and \textbf{4}$\times$ better overlap efficiency compared to state-of-the-art baselines, despite using FP32 while baselines use FP16. FlashDMoE demonstrates that principled GPU kernel-hardware co-design is key to unlocking the performance ceiling of large-scale distributed ML workloads.
This work presents a novel text-to-vector graphics generation approach, Dream3DVG, allowing for arbitrary viewpoint viewing, progressive detail optimization, and view-dependent occlusion awareness. Our approach is a dual-branch optimization framework, consisting of an auxiliary 3D Gaussian Splatting optimization branch and a 3D vector graphics optimization branch. The introduced 3DGS branch can bridge the domain gaps between text prompts and vector graphics with more consistent guidance. Moreover, 3DGS allows for progressive detail control by scheduling classifier-free guidance, facilitating guiding vector graphics with coarse shapes at the initial stages and finer details at later stages. We also improve the view-dependent occlusions by devising a visibility-awareness rendering module. Extensive results on 3D sketches and 3D iconographies, demonstrate the superiority of the method on different abstraction levels of details, cross-view consistency, and occlusion-aware stroke culling.
Reinforcement learning exhibits potential in enhancing the reasoning abilities of large language models, yet it is hard to scale for the low sample efficiency during the rollout phase. Existing methods attempt to improve efficiency by scheduling problems based on problem difficulties. However, these approaches suffer from unstable and biased estimations of problem difficulty and fail to capture the alignment between model competence and problem difficulty in RL training, leading to suboptimal results. To tackle these limitations, this paper introduces $\textbf{C}$ompetence-$\textbf{D}$ifficulty $\textbf{A}$lignment $\textbf{S}$ampling ($\textbf{CDAS}$), which enables accurate and stable estimation of problem difficulties by aggregating historical performance discrepancies of problems. Then the model competence is quantified to adaptively select problems whose difficulty is in alignment with the model's current competence using a fixed-point system. Experimental results across a range of challenging mathematical benchmarks show that CDAS achieves great improvements in both accuracy and efficiency. CDAS attains the highest average accuracy against baselines and exhibits significant speed advantages compared to Dynamic Sampling, a competitive strategy in DAPO, which is 2.33 times slower than CDAS.
We introduce a novel unsupervised learning method for time series data with latent dynamical structure: the recognition-parametrized Gaussian state space model (RP-GSSM). The RP-GSSM is a probabilistic model that learns Markovian Gaussian latents explaining statistical dependence between observations at different time steps, combining the intuition of contrastive methods with the flexible tools of probabilistic generative models. Unlike contrastive approaches, the RP-GSSM is a valid probabilistic model learned via maximum likelihood. Unlike generative approaches, the RP-GSSM has no need for an explicit network mapping from latents to observations, allowing it to focus model capacity on inference of latents. The model is both tractable and expressive: it admits exact inference thanks to its jointly Gaussian latent prior, while maintaining expressivity with an arbitrarily nonlinear neural network link between observations and latents. These qualities allow the RP-GSSM to learn task-relevant latents without ad-hoc regularization, auxiliary losses, or optimizer scheduling. We show how this approach outperforms alternatives on problems that include learning nonlinear stochastic dynamics from video, with or without background distractors. Our results position the RP-GSSM as a useful foundation model for a variety of downstream applications.
With the rapid growth of live streaming services, Crowdsourced Cloud-edge service Platforms (CCPs) are playing an increasingly important role in meeting the increasing demand. Although stream scheduling plays a critical role in optimizing CCPs' revenue, most optimization strategies struggle to achieve practical results due to various anomalies in unstable CCPs. Additionally, the substantial scale of CCPs magnifies the difficulties of anomaly detection in time-sensitive scheduling. To tackle these challenges, this paper proposes Sentinel, a proactive anomaly detection-based scheduling framework. Sentinel models the scheduling process as a two-stage Pre-Post-Scheduling paradigm: in the pre-scheduling stage, Sentinel conducts anomaly detection and constructs a strategy pool; in the post-scheduling stage, upon request arrival, it triggers an appropriate scheduling based on a pre-generated strategy to implement the scheduling process. Extensive experiments on realistic datasets show that Sentinel significantly reduces anomaly frequency by 70%, improves revenue by 74%, and doubles the scheduling speed.
Existing Large Language Model (LLM) serving systems prioritize maximum throughput. They often neglect Service Level Objectives (SLOs) such as Time to First Token (TTFT) and Time Per Output Token (TPOT), which leads to suboptimal SLO attainment. This paper introduces SCORPIO, an SLO-oriented LLM serving system designed to maximize system goodput and SLO attainment for workloads with heterogeneous SLOs. Our core insight is to exploit SLO heterogeneity for adaptive scheduling across admission control, queue management, and batch selection. SCORPIO features a TTFT Guard, which employs least-deadline-first reordering and rejects unattainable requests, and a TPOT Guard, which utilizes a VBS-based admission control and a novel credit-based batching mechanism. Both guards are supported by a predictive module. Evaluations demonstrate that SCORPIO improves system goodput by up to 14.4X and SLO adherence by up to 46.5% compared to state-of-the-art baselines.
Detection Transformers (DETR) are increasingly adopted in autonomous vehicle (AV) perception systems due to their superior accuracy over convolutional networks. However, concurrently executing multiple DETR tasks presents significant challenges in meeting firm real-time deadlines (R1) and high accuracy requirements (R2), particularly for safety-critical objects, while navigating the inherent latency-accuracy trade-off under resource constraints. Existing real-time DNN scheduling approaches often treat models generically, failing to leverage Transformer-specific properties for efficient resource allocation. To address these challenges, we propose CF-DETR, an integrated system featuring a novel coarse-to-fine Transformer architecture and a dedicated real-time scheduling framework NPFP**. CF-DETR employs three key strategies (A1: coarse-to-fine inference, A2: selective fine inference, A3: multi-level batch inference) that exploit Transformer properties to dynamically adjust patch granularity and attention scope based on object criticality, aiming to satisfy R2. The NPFP** scheduling framework (A4) orchestrates these adaptive mechanisms A1-A3. It partitions each DETR task into a safety-critical coarse subtask for guaranteed critical object detection within its deadline (ensuring R1), and an optional fine subtask for enhanced overall accuracy (R2), while managing individual and batched execution. Our extensive evaluations on server, GPU-enabled embedded platforms, and actual AV platforms demonstrate that CF-DETR, under an NPFP** policy, successfully meets strict timing guarantees for critical operations and achieves significantly higher overall and critical object detection accuracy compared to existing baselines across diverse AV workloads.
It is envisioned that the next generations of wireless communication environment will be characterized with dense traffic demand due to the prediction that there will be large numbers of active users. Hence, it is important to find a solution to deal with such dense numbers of users. This paper investigates optimizing the connectivity and users scheduling to improve the performance of near and far field users in a downlink, multiuser, massive MIMO-NOMA system. For the considered system model, combining NOMA side by side with massive MIMO offers a great opportunity to exploit the available radio resources and boost the overall system efficiency. The paper proposes separate clustering of near field users and far field users. It also proposes using a beamforming scheme to separately serve the users within each cluster. However, NOMA is proposed to be applied among all users to boost resource sharing. In particular, a cognitive-NOMA beamforming scheme and NOMA themed beamforming are proposed to serve the users within each cluster, and they are compared against random beamforming from literature. Simulation results show that both of the proposed beamforming schemes proved their superiority as compared to random beamforming. Several scheduling techniques were also considered in this paper to examine possible solutions for boosting the system performance considered, namely, priority, joint, dynamic, and fairness-based scheduling techniques for both near field and far field users. The paper also proposes a suboptimal, fairness aiming and gradual allocation approach for allocating the transmission power among the users. The results show that user-clustering offers better connectivity and scheduling performance than the case where no clustering is applied.
In recent years, Large Language Models (LLM) such as ChatGPT, CoPilot, and Gemini have been widely adopted in different areas. As the use of LLMs continues to grow, many efforts have focused on reducing the massive training overheads of these models. But it is the environmental impact of handling user requests to LLMs that is increasingly becoming a concern. Recent studies estimate that the costs of operating LLMs in their inference phase can exceed training costs by 25x per year. As LLMs are queried incessantly, the cumulative carbon footprint for the operational phase has been shown to far exceed the footprint during the training phase. Further, estimates indicate that 500 ml of fresh water is expended for every 20-50 requests to LLMs during inference. To address these important sustainability issues with LLMs, we propose a novel framework called SLIT to co-optimize LLM quality of service (time-to-first token), carbon emissions, water usage, and energy costs. The framework utilizes a machine learning (ML) based metaheuristic to enhance the sustainability of LLM hosting across geo-distributed cloud datacenters. Such a framework will become increasingly vital as LLMs proliferate.
We study the problem of assigning operations in a dataflow graph to devices to minimize execution time in a work-conserving system, with emphasis on complex machine learning workloads. Prior learning-based methods often struggle due to three key limitations: (1) reliance on bulk-synchronous systems like TensorFlow, which under-utilize devices due to barrier synchronization; (2) lack of awareness of the scheduling mechanism of underlying systems when designing learning-based methods; and (3) exclusive dependence on reinforcement learning, ignoring the structure of effective heuristics designed by experts. In this paper, we propose \textsc{Doppler}, a three-stage framework for training dual-policy networks consisting of 1) a $\mathsf{SEL}$ policy for selecting operations and 2) a $\mathsf{PLC}$ policy for placing chosen operations on devices. Our experiments show that \textsc{Doppler} outperforms all baseline methods across tasks by reducing system execution time and additionally demonstrates sampling efficiency by reducing per-episode training time.
This paper characterizes the impacts of channel estimation errors and Rician factors on achievable data rate and investigates the user scheduling strategy, combining scheme, power control, and dynamic bandwidth allocation to maximize the sum data rate in the distributed multiple-input-multiple-output (MIMO)-enabled low earth orbit (LEO) satellite networks. However, due to the resource-assignment problem, it is challenging to find the optimal solution for maximizing the sum data rate. To transform this problem into a more tractable form, we first quantify the channel estimation errors based on the minimum mean square error (MMSE) estimator and rigorously derive a closed-form lower bound of the achievable data rate, offering an explicit formulation for resource allocation. Then, to solve the NP-hard problem, we decompose it into three sub-problems, namely, user scheduling strategy, joint combination and power control, and dynamic bandwidth allocation, by using alternative optimization (AO). Specifically, the user scheduling is formulated as a graph coloring problem by iteratively updating an undirected graph based on user requirements, which is then solved using the DSatur algorithm. For the combining weights and power control, the successive convex approximation (SCA) and geometrical programming (GP) are adopted to obtain the sub-optimal solution with lower complexity. Finally, the optimal bandwidth allocation can be achieved by solving the concave problem. Numerical results validate the analytical tightness of the derived bound, especially for large Rician factors, and demonstrate significant performance gains over other benchmarks.
Decision Transformer (DT) has recently demonstrated strong generalizability in dynamic resource allocation within unmanned aerial vehicle (UAV) networks, compared to conventional deep reinforcement learning (DRL). However, its performance is hindered due to zero-padding for varying state dimensions, inability to manage long-term energy constraint, and challenges in acquiring expert samples for few-shot fine-tuning in new scenarios. To overcome these limitations, we propose an attention-enhanced prompt Decision Transformer (APDT) framework to optimize trajectory planning and user scheduling, aiming to minimize the average age of information (AoI) under long-term energy constraint in UAV-assisted Internet of Things (IoT) networks. Specifically, we enhance the convenional DT framework by incorporating an attention mechanism to accommodate varying numbers of terrestrial users, introducing a prompt mechanism based on short trajectory demonstrations for rapid adaptation to new scenarios, and designing a token-assisted method to address the UAV's long-term energy constraint. The APDT framework is first pre-trained on offline datasets and then efficiently generalized to new scenarios. Simulations demonstrate that APDT achieves twice faster in terms of convergence rate and reduces average AoI by $8\%$ compared to conventional DT.
In content moderation for social media platforms, the cost of delaying the review of a content is proportional to its view trajectory, which fluctuates and is apriori unknown. Motivated by such uncertain holding costs, we consider a queueing model where job states evolve based on a Markov chain with state-dependent instantaneous holding costs. We demonstrate that in the presence of such uncertain holding costs, the two canonical algorithmic principles, instantaneous-cost ($c\mu$-rule) and expected-remaining-cost ($c\mu/\theta$-rule), are suboptimal. By viewing each job as a Markovian ski-rental problem, we develop a new index-based algorithm, Opportunity-adjusted Remaining Cost (OaRC), that adjusts to the opportunity of serving jobs in the future when uncertainty partly resolves. We show that the regret of OaRC scales as $\tilde{O}(L^{1.5}\sqrt{N})$, where $L$ is the maximum length of a job's holding cost trajectory and $N$ is the system size. This regret bound shows that OaRC achieves asymptotic optimality when the system size $N$ scales to infinity. Moreover, its regret is independent of the state-space size, which is a desirable property when job states contain contextual information. We corroborate our results with an extensive simulation study based on two holding cost patterns (online ads and user-generated content) that arise in content moderation for social media platforms. Our simulations based on synthetic and real datasets demonstrate that OaRC consistently outperforms existing practice, which is based on the two canonical algorithmic principles.
Generative image editing using diffusion models has become a prevalent application in today's AI cloud services. In production environments, image editing typically involves a mask that specifies the regions of an image template to be edited. The use of masks provides direct control over the editing process and introduces sparsity in the model inference. In this paper, we present InstGenIE, a system that efficiently serves image editing requests. The key insight behind InstGenIE is that image editing only modifies the masked regions of image templates while preserving the original content in the unmasked areas. Driven by this insight, InstGenIE judiciously skips redundant computations associated with the unmasked areas by reusing cached intermediate activations from previous inferences. To mitigate the high cache loading overhead, InstGenIE employs a bubble-free pipeline scheme that overlaps computation with cache loading. Additionally, to reduce queuing latency in online serving while improving the GPU utilization, InstGenIE proposes a novel continuous batching strategy for diffusion model serving, allowing newly arrived requests to join the running batch in just one step of denoising computation, without waiting for the entire batch to complete. As heterogeneous masks induce imbalanced loads, InstGenIE also develops a load balancing strategy that takes into account the loads of both computation and cache loading. Collectively, InstGenIE outperforms state-of-the-art diffusion serving systems for image editing, achieving up to 3x higher throughput and reducing average request latency by up to 14.7x while ensuring image quality.
Speculative decoding accelerates Large Language Model (LLM) inference by employing a small speculative model (SSM) to generate multiple candidate tokens and verify them using the LLM in parallel. This technique has been widely integrated into LLM inference serving systems. However, inference requests typically exhibit uncertain execution time, which poses a significant challenge of efficiently scheduling requests in these systems. Existing work estimates execution time based solely on predicted output length, which could be inaccurate because execution time depends on both output length and token acceptance rate of verification by the LLM. In this paper, we propose a semi-clairvoyant request scheduling algorithm called Least-Attained/Perceived-Service for Speculative Decoding (LAPS-SD). Given a number of inference requests, LAPS-SD can effectively minimize average inference latency by adaptively scheduling requests according to their features during decoding. When the token acceptance rate is dynamic and execution time is difficult to estimate, LAPS-SD maintains multiple priority queues and allows request execution preemption across different queues. Once the token acceptance rate becomes stable, LAPS-SD can accurately estimate the execution time and schedule requests accordingly. Extensive experiments show that LAPS-SD reduces inference latency by approximately 39\% compared to state-of-the-art scheduling methods.
Large language models (LLMs) power many modern applications, but serving them at scale remains costly and resource-intensive. Current server-centric systems overlook consumer-grade GPUs at the edge. We introduce SpecEdge, an edge-assisted inference framework that splits LLM workloads between edge and server GPUs using a speculative decoding scheme, exchanging only token outputs over the network. SpecEdge employs proactive edge drafting to overlap edge token creation with server verification and pipeline-aware scheduling that interleaves multiple user requests to increase server-side throughput. Experiments show SpecEdge enhances overall cost efficiency by 1.91x through achieving 2.22x server throughput, and reduces inter token latency by 11.24% compared to a server-only baseline, introducing a scalable, cost-effective paradigm for LLM serving.