Research Papers
Research Papers news — Wednesday 30 September 2026
38 stories from 2 sources, every one in English. Each links to its publisher.
-
vSkipper: Translating Dynamic Layer Skipping into LLM Serving Gains
arXiv cs.DC (Distributed)
arXiv:2609.37062v1 Announce Type: new Abstract: Dynamic layer skipping reduces LLM computation by allowing each token to execute only a subset of the model's layers. However, existing skippers rely on specialized generation loops and do…
-
DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification
arXiv cs.DC (Distributed)
arXiv:2609.37532v1 Announce Type: new Abstract: Growing large language model applications demand efficient inference. At high concurrency, block-diffusion speculative decoding suffers from verification padding, rejected candidates, and…
-
CF-LoRA: Decoupled Factor Aggregation and Adaptation-Aware Client Clustering for Federated LoRA Fine-Tuning
arXiv cs.DC (Distributed)
arXiv:2609.36986v1 Announce Type: cross Abstract: Federated LoRA fine-tuning enables parameter-efficient adaptation of pre-trained models without sharing private data, but suffers from two fundamental mismatches under heterogeneous client…
-
SHORTCUT: In-Collective Topology Reconfiguration for Low-Latency AllReduce
arXiv cs.DC (Distributed)
arXiv:2510.03491v2 Announce Type: replace-cross Abstract: Distributed ML training relies on efficient AllReduce communication to aggregate data across nodes. In this setting, reconfigurable optical interconnects offer high-bandwidth…
-
Mixture-of-Kittens: MoE Megakernel for NVL72s
arXiv cs.DC (Distributed)
arXiv:2609.36070v1 Announce Type: new Abstract: AI accelerator systems are rapidly consolidating into scale-up architectures, where tens to thousands of GPUs communicate over high-bandwidth, single-hop fabrics. We find that existing…
-
Encoding and Node Choices in Transversal Fault-Tolerant Distributed Quantum Computations: An Initial Study
arXiv cs.DC (Distributed)
arXiv:2609.37210v1 Announce Type: cross Abstract: We compare and study different Bivariate-Bicycle (BB) encodings and node choices for distributed quantum operations such as transversal non-local CNOTs. We observe that while some encodings…
-
Joint Effects of GPU Server Topology, Parallelism, and Congestion Control on MoE Inference: A Controlled Simulation Study
arXiv cs.DC (Distributed)
arXiv:2609.37828v1 Announce Type: new Abstract: Mixture-of-experts (MoE) models expand capacity via sparse activation, but inference across GPUs introduces tensor-parallel (TP) collectives and expert-parallel (EP) dispatch and combine…
-
RLX: A Unified Multi-Backend Tensor Compiler and Distributed Runtime in Rust
arXiv cs.DC (Distributed)
arXiv:2609.37916v1 Announce Type: new Abstract: Production machine learning (ML) stacks often split graph compilation and kernel execution across different layers and languages, making backend behavior, deployment guarantees, and…
-
Quantifying Teleportation Overhead in Distributed Unitary Coupled-Cluster Ans\"atze
arXiv cs.DC (Distributed)
arXiv:2609.36266v1 Announce Type: cross Abstract: Distributed quantum computing (DQC) has been proposed as a way to scale quantum algorithms for practical applications beyond monolithic quantum processor architectures. Among these…
-
ASCEND: Personal AI Agents for Autonomous Scientific Computing Across HPC Clusters and GPU Workstations
arXiv cs.DC (Distributed)
arXiv:2609.32868v2 Announce Type: replace Abstract: Traditional scientific computing requires researchers to translate computational intent into environment configuration, resource requests, and executable jobs, then diagnose failures from…
-
Windowed and Quantized Group-Based ADMM for Distributed Optimization in Heterogeneous Edge Networks
arXiv cs.DC (Distributed)
arXiv:2609.37069v1 Announce Type: new Abstract: Distributed optimization in edge networks is constrained by heterogeneous client computing capabilities and limited communication resources. We propose the Windowed and Quantized Group-Based…
-
Federated Clustering with Unknown Local and Global Cluster Cardinalities
arXiv cs.DC (Distributed)
arXiv:2609.36762v1 Announce Type: cross Abstract: Federated clustering methods that do not require the global number of clusters $K$ still assume that each client knows its local number $K_g$. This assumption is hard to justify when…
-
Purlin: Separating Orchestration from the Datapath of Collectives
arXiv cs.DC (Distributed)
arXiv:2609.36954v1 Announce Type: new Abstract: Distributed inference depends on GPU collective communication that must keep pace with evolving hardware and specialized workloads. However, existing collective implementations often couple…
-
Scepsy: Serving Agentic Workflows Using Aggregate LLM Pipelines
arXiv cs.DC (Distributed)
arXiv:2604.15186v3 Announce Type: replace Abstract: Agentic workflows carry out complex tasks by orchestrating multiple large language models (LLMs) and tools. Serving them at a target throughput with low latency is hard because they are…
-
Cobalt: Leveraging Expert Co-activation for Efficient Distributed MoE Training
arXiv cs.DC (Distributed)
arXiv:2609.36959v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) has increasingly become a mainstream approach for scaling large language models, as it expands model capacity while keeping computation cost nearly constant. Training…
-
SPLASH: Switching Parallel Layouts of Attention with Seamless Handoff for LLM Serving
arXiv cs.DC (Distributed)
arXiv:2609.37626v1 Announce Type: new Abstract: No single way of parallelizing attention serves large language models well under all loads. Low concurrency favors tensor parallelism, many independent requests favor data-parallel attention…
-
Throughput-Optimized Networks at Scale
arXiv cs.DC (Distributed)
arXiv:2605.27963v2 Announce Type: replace-cross Abstract: Data center network design plays a critical role in AI training by supporting scaling to thousands of accelerators. An open problem, designing a near-optimal throughput-oriented…
-
ARGOS: Reinforcement Learning-Driven Multidimensional Elasticity for Service Orchestration in the Computing Continuum
arXiv cs.DC (Distributed)
arXiv:2609.37085v1 Announce Type: new Abstract: Data-intensive services in the Computing Continuum must balance analytics quality, resource usage, and cost across heterogeneous nodes with limited and uneven capacity. This balance becomes…
-
ParaAnya: Accelerating Parallel Diffusion Sampling with Plug-and-Play Output Caching
arXiv cs.DC (Distributed)
arXiv:2609.36522v1 Announce Type: new Abstract: Diffusion models have achieved remarkable success in generative tasks, but their inherently sequential sampling process introduces a severe computational bottleneck. Recent Parallel-in-Time…
-
Byzantine Causal Reliable Broadcast with Constant Metadata Overhead
arXiv cs.DC (Distributed)
arXiv:2609.37913v1 Announce Type: new Abstract: Asynchronous Byzantine Reliable Broadcast (BRB) is a fundamental primitive that guarantees agreement and validity in distributed systems subject to Byzantine faults, but it lacks ordering…
-
MEDEM: Multi-Engine DL Accelerator Design Methodology
arXiv cs.DC (Distributed)
arXiv:2609.37399v1 Announce Type: cross Abstract: Multi-engine deep learning (DL) accelerators are becoming increasingly prevalent as they address the heterogeneity and growing complexity of modern DL workloads. To efficiently process…
-
Reshaping Rollout Workloads for Asynchronous RL Post-Training on Heterogeneous Accelerators
arXiv cs.DC (Distributed)
arXiv:2609.36899v1 Announce Type: new Abstract: Reinforcement learning (RL) post-training increasingly relies on long-horizon, multi-turn rollouts. As post-training jobs outgrow a single cluster, rollout pools assembled across clusters…
-
Accelerator Choice Is Not Enough: AlphaFold2 Inference on Cloud TPUs
arXiv cs.DC (Distributed)
arXiv:2609.34818v2 Announce Type: replace Abstract: AlphaFold2 is written in JAX, so the same inference code compiles and runs unchanged on CPUs, GPUs and Google Cloud TPUs. That portability makes the accelerator look like the main…
-
Hardware-Attributed Operator Profiling for PyTorch
arXiv cs.DC (Distributed)
arXiv:2609.11938v2 Announce Type: replace-cross Abstract: Framework profilers expose operator timing without hardware counters; GPU profilers expose hardware counters without operator attribution. Bridging this gap manually is error-prone…
-
Joule-Profiler: Profiling the Energy Consumption of Build Automation Tools Made Easy
arXiv cs.DC (Distributed)
arXiv:2609.31228v1 Announce Type: cross Abstract: Build pipelines are integral to modern software development, yet their energy footprint remains largely invisible to practitioners. Existing CI energy tools either rely on model-based…
-
Recolorable Graph Exploration by an Oblivious Agent with Fewer Colors
arXiv cs.DC (Distributed)
arXiv:2505.02789v2 Announce Type: replace Abstract: Recently, B\"ockenhauer, Frei, Unger, and Wehner (SIROCCO 2023) introduced a novel variant of the graph exploration problem in which a single memoryless agent must visit all nodes of an…
-
FP64 Is All You Want, INT8 Is All You Need, FP4/6/8 Is All You Have
arXiv cs.DC (Distributed)
arXiv:2609.37693v1 Announce Type: new Abstract: Ozaki scheme II emulates FP64 matrix products with INT8 ones through residues modulo pairwise coprime moduli, and variants for FP8 and FP4 have followed. We treat these schemes as one family…
-
Efficient Agentic LLM Serving over SSD-based Sparse KV Storage
arXiv cs.DC (Distributed)
arXiv:2609.36938v1 Announce Type: new Abstract: Agentic sessions driven by Large language models (LLMs) often alternate between model inference and tool use, accumulating long histories across successive rounds. Serving these sessions…
-
Economical and ecological impact of sector coupling applied to computing clusters
arXiv cs.DC (Distributed)
arXiv:2604.25540v2 Announce Type: replace Abstract: The rising share of abundant renewable energy inevitably increases volatility in the electricity production. The concept of sector coupling means that the volatility of electricity…
-
Consensus Time in 3-Majority and 2-Choices Is Determined by the Maximum Initial Opinion Density
arXiv cs.DC (Distributed)
arXiv:2606.11778v3 Announce Type: replace Abstract: We establish the correct parameter governing the convergence time of the 3-Majority and 2-Choices dynamics on the complete graph in the synchronous model. Recent work [Shimizu and…
-
Encoder-Sharing Hierarchical Federated Multi-Task Learning for VANETs
arXiv cs.DC (Distributed)
arXiv:2609.36157v1 Announce Type: cross Abstract: Most federated learning frameworks for vehicular ad hoc networks assume that all vehicles collaboratively train a single model for a common task. This assumption limits their applicability…
-
Hierarchical Secure Distributed Linearly Separable Computation with Arbitrary Heterogeneous Data Assignment
arXiv cs.DC (Distributed)
arXiv:2609.33092v2 Announce Type: replace Abstract: This paper studies secure distributed linearly separable computation over a three-layer hierarchical network, where clustered users communicate with a central server through relays. The…
-
Replay the Curvature: Accurate and Scalable NVFP4 Quantization for Large Language Model Inference
arXiv cs.DC (Distributed)
arXiv:2609.36654v1 Announce Type: cross Abstract: Large language models make weight storage and memory traffic major inference costs, motivating low-precision formats that represent each weight with only a few bits. Such formats use a…
-
Designing Dense Satellite Clusters for Distributed Space-based Datacenters
arXiv cs.DC (Distributed)
arXiv:2605.15335v2 Announce Type: replace Abstract: Recent proposals for datacenters in sun-synchronous Low Earth Orbit (LEO) rely on a large number of compute satellites formation-flying in dense clusters. Designing such satellite…
-
Heddle: Learning Structural Templates for Parallelism Planning on Heterogeneous GPU Clusters
arXiv cs.DC (Distributed)
arXiv:2609.34244v2 Announce Type: replace Abstract: Training large machine learning models on shared GPU infrastructures faces two challenges: (1) GPU availability shifts dynamically with varying resource demands from tenants, and (2)…
-
Making Cross-Continental Federated Learning Repeatable with FLIP: a Multi-Application Study
arXiv cs.DC (Distributed)
arXiv:2609.36001v1 Announce Type: cross Abstract: Federated learning (FL) in healthcare remains challenging, as the overhead of rebuilding governance guarantees for every collaboration stops most projects at the proof-of-concept stage…
-
Restoring hearing in people with a rare form of deafness
Nature Technology
-
Elena, Aris, Marcus: AI-generated ‘ghosts’ are polluting the scientific literature
Nature Technology