# GNN-Communication-Networks is a reading list, not a library > This repository holds no code: it is a curated bibliography of graph-based deep learning applied to communication networks, with a survey.md that indexes the surveyed studies by scenario and a monthly-updated literature list. Its most useful single artifact is a separate prototyping tool, IGNNITION, which is where you go when the reading list turns into an experiment. **jwwthu/GNN-Communication-Networks** — This is the repository for the collection of Graph-based Deep Learning for Communication Networks. - Repository: https://github.com/jwwthu/GNN-Communication-Networks - Stars: 607 · Forks: 89 - Language: Unknown - License: not declared - Published: 2026-10-22 · Updated: 2026-10-22 · Language: en - Canonical page: https://hysenlabs.com/projects/jwwthu-gnn-communication-networks ## Two files, and that is the whole repository The top-level listing of this repository contains two files and nothing else: README.md and survey.md. There is no source directory, no build configuration, no test suite and no continuous integration. The primary language is not even recorded, which is consistent with a repository that holds prose. What the README states at the top is that it is the repository for the collection of Graph-based Deep Learning for Communication Networks. That is the entire promise, and the rest of the file is a bibliography organised by category. The categories are worth reading as a map of the field rather than as a random list. There is a citation section, a section of other surveys, a competition section, a tool section, and a literature section that is split by year and then by publication type, with a 2026 subsection that begins with journal papers. The literature list says so in one line: ```bash The list would be updated monthly. ``` The 2026 entries are the visible portion of that maintenance. A single year of journal papers in this area arrives at a rate of several per month, so the monthly claim is a substantial editorial commitment rather than a formality, and it means the README is a living index rather than a historical record. For a reader who wants to know what a paper did, this repository will not tell you. It is a table of contents with author lists, venues, years, volume and issue numbers where available, and links to publisher pages. ## survey.md is the part that turns a list into a map The citation block at the top of the README tells you which works the author wants acknowledged, and the first of them points at the companion file. The survey on graph-based deep learning for communication networks is cited as a Computer Communications paper from 2022, volume 185, pages 40 to 54, and immediately under that citation there is a note: for the surveyed studies in different scenarios, you may check survey.md. That is the key navigational hint. survey.md is a scenario index, so instead of reading a flat list of papers you can look up the situation your problem resembles and find the studies that addressed it. For a field where the applications are scattered across routing, resource allocation, network management, localisation, IoT and 6G, that index is the difference between a usable reference and an overwhelming one. The other two works in the citation block are a federated learning paper for network attack detection using attention-based graph neural networks, published in Scientific Reports in 2024, and a paper on graph neural networks for routing optimisation, published in Sustainability in 2024. The author asks readers to cite the relevant work if the repository proves helpful, which is a request for academic credit rather than a technical dependency, and it tells you the repository is maintained by an academic who also publishes in this area. There is also a promotional line inviting readers to follow a WeChat public account and a Zhihu account, both named for networks and communications, which the author says push conference deadline summaries and open source code and paper collections. It is worth knowing the repository is one artefact of a wider personal publication effort, because that shapes what it contains: reading lists and pointers rather than maintained tools. ## The tool is where the reading list turns into an experiment Under the tool heading, the README points to IGGNITION, described as a fast prototyping tool for graph neural networks for communication networks, from a SIGCOMM 2021 poster and demo session paper, and it is the one entry in the whole repository that leads to code. A companion IEEE Network paper describes IGNNITION as bridging the gap between graph neural networks and networking systems, and the link for the code points to a separate site at ignnition.org. The important point for an engineer is that the tool is not in this repository. This repository curates and points; the implementation lives elsewhere and is its own project with its own lifecycle. The tool entry also illustrates a pattern you will see across the whole list. A graph neural network paper in a networking venue usually ships as code that is hard to rerun, because the datasets are telecom operator data that cannot be redistributed. A prototyping tool that works on generic network topologies is what makes the rest of the literature reproducible at all, and it is why the author lists it separately under a tool heading rather than among the papers. If you are evaluating whether graph learning is applicable to your network problem, the realistic workflow is to read survey.md to find the scenario, read the two or three papers that match it, and then use the prototyping tool to build your own graph representation of the network and test the idea. The repository supports the first two steps and points at the third. ## The 2026 entries cluster around resource allocation and graph reinforcement learning The visible 2026 journal list is the most current signal of where the field is going, and the clustering is clear. Resource allocation with graph methods dominates. There is an adaptive graph reinforcement learning method for load balancing in multi-edge collaboration in wireless metropolitan area networks, a multi-agent graph reinforcement learning method for decentralised topology robustness optimisation in the Internet of Things, a graph neural network and deep reinforcement learning approach for V2V resource allocation, a max-min fair resource allocation scheme for rate-splitting multiple access built on graph neural networks, a graph attention network approach to data detection and resource allocation in distributed two-way cell-free networks, and a graph neural network enabled power allocation scheme for NOMA. Reinforcement learning appears as often as supervised learning, and usually in the multi-agent form, which is a natural fit because a network is a graph of interacting agents and the actions of one node change the payoffs of its neighbours. Two entries make that explicit in their names, MARLA-TGN for dynamic privacy-preserving virtual network function auctions in space-ground integrated 6G networks, and GT-MARL for graph-and-transformer-enhanced multi-agent reinforcement learning for cloud-edge collaborative scheduling. Other clusters are visible too: knowledge-driven work such as a physics-informed heterogeneous graph attention network with KAN regression for indoor localisation, a meta-learning enhanced graph neural network approach for signal compression in uplink cell-free MIMO, and graph neural networks for optimising data analytics workflow scheduling in cloud-fog computing. A careful reader will notice that most of these are 2026 IEEE Transactions papers on mobile computing, vehicular technology, the Internet of Things Journal and China Communications. The list is therefore a snapshot of what a particular research community is publishing, and it is not a neutral survey of everything published on the topic. ## Related surveys and a competition framework The other surveys section widens the frame beyond the author's own work, and it is where a reader finds the orientation pieces. There is an overview on the application of graph neural networks in wireless networks in the IEEE Open Journal of the Communications Society from 2021, and a paper on graph neural networks for communication networks covering context, use cases and opportunities in IEEE Network from 2022, which reads from its title like the broadest of the group. A survey on graph neural networks for intelligent modelling in network management and orchestration sits in Electronics from 2022, and a paper on graph-based resource allocation for integrated space and terrestrial communications in Sensors from 2022. A 2022 IEEE Wireless Communications piece covers graph neural networks meeting wireless communications with motivation, applications and future directions. Two of the entries point to code as well as to the paper. The graph-powered learning methods in the Internet of Things survey in Machine Learning with Applications from 2023 is a literature piece, and the graph neural networks in IoT survey in ACM Transactions on Sensor Networks from 2023 links to a separate repository called GNN4IoT, which is the one other entry in the list that is a codebase rather than a paper. The competition section is a different kind of pointer. It cites a paper about the graph neural networking challenge, described as a worldwide competition for education in AI and ML for networks, published in ACM SIGCOMM Computer Communication Review in 2021, plus the arXiv preprint on scaling graph-based deep learning models to larger networks. The competition framing is worth knowing about, because a public challenge with an evaluation dataset is one of the few ways to compare graph methods for networks on a common footing without operator data. ## What you cannot get from this repository Three limitations are worth stating plainly, because the repository type invites confusion. The first is that there is no code and no licence. The recorded licence is unknown, and the primary language is unknown too. For a collection of links and prose, the missing licence is unlikely to block research use, but if you want to reuse the survey text, the tables or the organisation of the list in your own paper or tooling, you have no clear permission and no contact path in the repository to ask for one. The second is that the list is a pointer, not a verified summary. Each entry gives authors, title, venue, year and a link, and the entries are grouped by type and year, but nothing in the repository tells you what a paper found, whether its result was replicated, or how the datasets were obtained. A citation list is a table of contents: it saves you search time and it does not save you reading time. The third is that the repository does not implement or benchmark anything, so a claim in a paper cannot be checked against the original code from here, and the practical work of turning a paper into a result lands on you, with the IGNNITION tool as the only pointer toward a starting point. A smaller detail worth noting for anyone who searches by topic: the repository topic list spells communications with three m characters, so a search for the correctly spelled topic will not find it. ## Conclusion This repository fits a researcher or graduate student entering graph learning for networking who needs a map of the literature and a scenario index rather than a maintained code base, and it does not fit an engineer who needs an implementation, because the repository contains only a README and survey.md with no licence declared. Before you rely on it, read survey.md for the scenario that matches your problem, note that the author asks you to cite the survey and the routing optimisation paper if the collection helps you, and treat every entry as a pointer to a paper rather than as a validated result. The last push was on 2026-08-31 and the repository publishes no releases. ## FAQ ### What is in the GNN-Communication-Networks repository? It holds two files, README.md and survey.md, and no code. The README is a curated bibliography of graph-based deep learning for communication networks, with sections for citation, other surveys, a competition, a tool and a literature list split by year, and survey.md indexes the surveyed studies by scenario. ### How often is the literature list updated? The README states that the list would be updated monthly, and the visible 2026 section contains journal papers from venues including IEEE Transactions on Mobile Computing, IEEE Transactions on Vehicular Technology, the IEEE Internet of Things Journal and China Communications. ### Which tool does the repository recommend for building graph models for networks? It lists IGGNITION, described as a fast prototyping tool for graph neural networks for communication networks, with a SIGCOMM 2021 poster and demo paper and an IEEE Network paper on bridging the gap between graph neural networks and networking systems. The code link points to a separate site rather than to this repository. ### What licence applies to the GNN-Communication-Networks repository? No licence is recorded for the repository, and the primary language is not recorded either. The contents are a README and a survey file of references, with no source code, so there is nothing to install or build. ## Sources - [Issues](https://github.com/jwwthu/GNN-Communication-Networks/issues) - [jwwthu/GNN-Communication-Networks on GitHub](https://github.com/jwwthu/GNN-Communication-Networks) - [README](https://github.com/jwwthu/GNN-Communication-Networks/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/jwwthu-gnn-communication-networks