Product Information
Description
An annual survey of the cs.DC arXiv category, covering research published from September 2025 – August 2026. Built on a close-reading set of 132 papers and organised into 37 chapters, it maps the year by the field's research sub-directions.
Book framework (share of the year's output):
- Communication-Efficient Parallel Training and Inference (13.4%) — Optimizing distributed training and inference systems by minimizing communication overhead, memory usage, and energy consumption through efficient parallelism and compression techniques.
- Agentic AI Systems and HPC Workflow Orchestration (12.7%) — Designing frameworks for agentic AI and hybrid human-AI collaboration within high-performance computing environments to optimize complex, compute-intensive workflows.
- Distributed Quantum Computing and Graph Algorithms (11.7%) — Developing distributed algorithms for quantum computing, graph processing, and fault-tolerant systems to enhance scalability and security in decentralized networks.
- Multi-Agent Consensus and Resource Scheduling (10.5%) — Addressing scheduling, consensus, and resource allocation challenges in multi-agent systems, multi-tenant environments, and serverless architectures.
- Privacy-Preserving Federated Learning Systems (10.3%) — Enhancing federated learning frameworks with privacy-preserving mechanisms, energy efficiency, and robust client-level disagreement resolution for decentralized data.
- High-Performance Computing Architecture and Analysis (8.4%) — Analyzing and optimizing high-performance computing architectures, including GPU kernels, many-core platforms, and performance diagnostics for exascale systems.
- GPU Cluster Management and Accelerated Computing (8.3%) — Optimizing GPU cluster orchestration, multi-GPU simulation, and heterogeneous resource allocation to maximize throughput and reliability in accelerated computing.
- Disaggregated LLM Serving and KV Cache Optimization (7.4%) — Improving large language model serving efficiency through disaggregated architectures, KV cache management, speculative decoding, and SLO-aware scheduling.
- Edge-Cloud Continuum and Adaptive Resource Management (7.4%) — Managing adaptive resource allocation and task scheduling across the edge-cloud continuum to balance latency, reliability, and cost in distributed applications.
- Large-Scale LLM Training and Infrastructure Optimization (6.8%) — Optimizing infrastructure and scheduling for large-scale language model training and inference, including simulation, policy repair, and distributed optimization.
- Efficient LLM Inference and Speculative Decoding (3.1%) — Reducing latency and energy costs in LLM inference via speculative decoding, MoE optimization, and hardware-aware memory management on edge and superchips.
Each sub-direction is covered across three chapters — how it framed its questions, the strength of its key findings, and its open debates and frontiers — so the reader gets a traceable picture of what this part of the field actually produced over the year.
