Research Papers
Research Papers news — Monday 19 October 2026
29 stories from 1 sources, every one in English. Each links to its publisher.
-
BaGuaLu-2: Taming Dynamic Memory Fluctuation for Efficient MoE Model Training over 10,000 GPUs
arXiv cs.DC (Distributed)
arXiv:2610.23064v1 Announce Type: new Abstract: Fine-grained Mixture-of-Experts (MoE) is the de facto architecture for scaling large language models. However, dynamic expert load imbalance causes severe memory fluctuations that degrade…
-
Understanding KV Residency in Production-Scale Agentic LLM Workloads
arXiv cs.DC (Distributed)
arXiv:2610.22863v1 Announce Type: new Abstract: Agentic large language model (LLM) applications turn one user request into a sequence of model and tool calls over a growing context. Retaining this context's key-value (KV) state in GPU…
-
ScaleRAN: A Runtime for Elastic GPU Resource Management in AI-RAN
arXiv cs.DC (Distributed)
arXiv:2610.23072v1 Announce Type: cross Abstract: AI-RAN is driving the convergence of radio access network (RAN) processing and artificial intelligence (AI) workloads on GPU-enabled computing platforms, ranging from GPU-accelerated base…
-
Accelerated Execution of Bayesian Neural Networks using a Single Probabilistic Forward Pass and Code Generation
arXiv cs.DC (Distributed)
arXiv:2511.23440v2 Announce Type: replace-cross Abstract: Machine learning models perform well across domains such as diagnostics, weather forecasting, NLP, and autonomous driving, but their limited uncertainty handling restricts use in…
-
One Communicator, Three Vendors: Extending the MoA Predictive Theory to a Real NVIDIA+AMD+CPU MPI Ensemble
arXiv cs.DC (Distributed)
arXiv:2610.22769v1 Announce Type: new Abstract: The companion MPI-extension paper validated the first communication-bearing machine-shape ensemble: four NVIDIA A100 ranks and four CPU ranks on one NCSA Delta node under one Open MPI job…
-
DIFFLET: A Fast Inference Engine for Diffusion Transformer Models on AWS Trainium
arXiv cs.DC (Distributed)
arXiv:2610.23047v1 Announce Type: new Abstract: On AWS Trainium2, every component of a diffusion model runs as a program compiled ahead of time for fixed shapes, while pipelines differ across models, prompt lengths across requests, and…
-
Federated k-Core Decomposition: A Secure Distributed Approach
arXiv cs.DC (Distributed)
arXiv:2410.02544v3 Announce Type: replace Abstract: As one of the most well-studied cohesive subgraph models, the $k$-core is widely used to find graph nodes that are ``central'' or ``important'' in many applications, such as biological…
-
On the (In)stability of Adaptive Fee Mechanisms in Multi-Chain Systems
arXiv cs.DC (Distributed)
arXiv:2610.23305v1 Announce Type: cross Abstract: EIP-1559 has become the dominant transaction fee mechanism in permissionless blockchains, regulating scarce block space through an adaptive base fee. The design balances a trade-off between…
-
TALLOP: Joint Optimization of Local Training and Communication Compression for Time-Constrained Federated Learning
arXiv cs.DC (Distributed)
arXiv:2610.22996v1 Announce Type: cross Abstract: Minimizing the wall-clock time required to reach a target accuracy is crucial in federated learning (FL), where many applications require frequent model updates as data distributions evolve…
-
Encodings and Node Architecture: Lessons from the Fault-Tolerant Distributed Transversal Fanout
arXiv cs.DC (Distributed)
arXiv:2610.22966v1 Announce Type: cross Abstract: An encoding choice specifies the quantum error-correcting code used to encode logical qubits at a given node. Different encoding choices may differ in the number of physical qubits, the…
-
From Border-Agnosticism to Causal Rewrite Systems
arXiv cs.DC (Distributed)
arXiv:2610.23351v1 Announce Type: cross Abstract: Causal Graph Rewriting was recently introduced as an evolution of Causal Graph Dynamics to allow for a controlled form of asynchronism in these dynamical systems with dynamical graphs. In…
-
QuarkShard: Decomposing Cross-Shard Requests into Single-Input Transactions
arXiv cs.DC (Distributed)
arXiv:2610.23014v1 Announce Type: new Abstract: Sharding blockchains enhance scalability by distributing the workload across multiple shards. Within sharding blockchains, securely and efficiently processing cross-shard user requests…
-
Scalable Levenberg-Marquardt Training Framework for Deep Learning
arXiv cs.DC (Distributed)
arXiv:2610.22925v1 Announce Type: cross Abstract: The Levenberg-Marquardt (LMA) algorithm is a widely used nonlinear least-squares solver known for its rapid convergence. For small-scale neural networks, LMA provides better convergence and…
-
Distributed Stochastic Momentum Tracking with Local Updates: Achieving Optimal Communication Complexities
arXiv cs.DC (Distributed)
arXiv:2510.24155v4 Announce Type: replace-cross Abstract: We propose Local Momentum Tracking (LMT), a novel distributed stochastic gradient method for solving distributed optimization problems over networks. To reduce communication…
-
TRACE: Transformer Residual Activation Codebook Encoding for Communication-efficient Distributed Edge Inference
arXiv cs.DC (Distributed)
arXiv:2610.22435v1 Announce Type: cross Abstract: Distributed Transformer inference across co-located edge devices is fundamentally communication-bound, as exchanging activations between sequence-parallel partitions dominates inference…
-
FederatedBytes: Tokenizer-Free, Byte-Level Small Language Models for Efficient Multimodal Federated Learning
arXiv cs.DC (Distributed)
arXiv:2610.22906v1 Announce Type: cross Abstract: Federated Learning (FL) is a powerful machine learning framework that allows building a global model from the contributions of multiple, distributed clients, while ensuring the privacy of…
-
The LLM-Attacker: A Unified Red-Team Framework for Privacy Evaluation of Split and Distributed LLM Systems
arXiv cs.DC (Distributed)
arXiv:2610.22722v1 Announce Type: cross Abstract: Split language models keep input text on a client but send intermediate representations to a server. Attacks can reconstruct text from these messages. Comparing privacy evaluations is…
-
TCDA: Tunable Coded Distributed Arrays for Data Availability Sampling
arXiv cs.DC (Distributed)
arXiv:2610.22508v1 Announce Type: new Abstract: Data availability sampling (DAS) lets blockchain nodes verify a block's data availability by downloading only a few random pieces of it. Its network layer must disseminate each block in time…
-
Fast Distributed $(\Delta + 1)$-Edge Coloring in Bipartite Graphs
arXiv cs.DC (Distributed)
arXiv:2610.22690v1 Announce Type: new Abstract: We show that, on bipartite $n$-vertex graphs in the LOCAL model of distributed computation, $(\Delta + 1)$-edge coloring can be solved deterministically in $O(f(\Delta) \log n)$ rounds for…
-
MailRail: Mitigating Incast Congestion with User-Space, Node-Based Flow Control
arXiv cs.DC (Distributed)
arXiv:2610.22890v1 Announce Type: new Abstract: Irregular workloads, such as those stemming from Graph Neural Networks (GNNs) and Mixture-of-Experts (MoE) models, frequently induce severe incast congestion in large-scale parallel…
-
From Pilots to Production: Lessons in Cross-Institutional Federated Training and Artificial Intelligence for Science
arXiv cs.DC (Distributed)
arXiv:2609.39803v2 Announce Type: replace Abstract: Many of the most valuable scientific datasets cannot be centralized: they are proprietary, export-controlled, classified, or bound by data-sovereignty restrictions. This inverts the usual…
-
On the Power of Alternating Local Certification with Constant-Size Certificates
arXiv cs.DC (Distributed)
arXiv:2610.23694v1 Announce Type: new Abstract: We study alternation hierarchies for distributed decision in anonymous networks with constant-size certificates. A prover and a disprover alternately assign certificates to nodes over a…
-
Redundancy Meets Synergy: Dependency-aware Expert Selection for MoE via Submodular Optimization
arXiv cs.DC (Distributed)
arXiv:2610.00558v2 Announce Type: replace-cross Abstract: While Mixture-of-Experts (MoE) models effectively scale model capacity through sparse activation, their deployment is often bottlenecked by prohibitive memory requirements…
-
Agentic AI for Heterogeneous System Design Across Chip, Scale-Up, and Scale-Out Domains
arXiv cs.DC (Distributed)
arXiv:2610.23724v1 Announce Type: cross Abstract: The design space of modern AI/HPC systems has expanded rapidly across compute architectures, including GPUs, dataflow processors, and RISC-V accelerators; memory tech- nologies, including…
-
Revisiting Cauchy-Schwarz Norm Bounds for SDC Detection in LLM Inference
arXiv cs.DC (Distributed)
arXiv:2610.18557v2 Announce Type: replace Abstract: Large Language Models (LLMs) execute billions of floating-point operations per output token across many hardware accelerators. At this scale, hardware-induced faults can bypass…
-
Flash RNN-T: Memory-Efficient Joint Network and Transducer Loss Computation
arXiv cs.DC (Distributed)
arXiv:2610.23409v1 Announce Type: cross Abstract: The recurrent neural network transducer (RNN-T) remains a practical choice for streaming speech recognition, but training requires large intermediate tensors. We present Flash RNN-T, a…
-
Optimistic, Signature-Free Reliable Broadcast and Its Applications
arXiv cs.DC (Distributed)
arXiv:2505.02761v5 Announce Type: replace Abstract: Reliable broadcast (RBC) is a key primitive in fault-tolerant distributed systems, and improving its efficiency can benefit a wide range of applications. This work focuses on…
-
RoutePack: Replay-Guided Expert Placement and Rollout Trajectory Scheduling for MoE Reinforcement Learning
arXiv cs.DC (Distributed)
arXiv:2608.12146v2 Announce Type: replace Abstract: Reinforcement learning (RL) post-training of Mixture-of-Experts (MoE) models couples two load-balancing problems. Rollout-trajectory composition determines attention and dense…
-
Accelerated Dataflow: scaling inference in streaming data analytics with efficiency
arXiv cs.DC (Distributed)
arXiv:2610.22731v1 Announce Type: new Abstract: Streaming data analytics helps businesses identify opportunities and spot risks in a timely fashion. As large language model (LLM) usage grows, people are exploring incorporating agents or…