FSL-Mate: A Paper Tracker and a PaddlePaddle Library in One Repository
FSL-Mate: A collection of resources for few-shot learning (FSL).
At a glance
- What is it?
- FSL-Mate bundles a continuously updated few-shot learning bibliography with PaddleFSL, a PaddlePaddle-based Python library. The paper list is the part with a clear maintenance record; the library is the part you must verify against your own stack before committing.
- Who is it for?
- Adopt FSL-Mate if you need a single place to scan few-shot learning papers across CVPR, ICML, NeurIPS, ACL and adjacent venues, or if your training stack is already PaddlePaddle and you want FSL utilities that match it. Do not adopt it as a framework-agnostic benchmark suite: the library half is tied to PaddlePaddle, and the repository ships no licence file, so redistribution terms are unresolved until the maintainers add one.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 151 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
Two Projects Sharing One Repository
FSL-Mate is not a single tool. The README describes it as a collection of resources for few-shot learning, and it currently contains two distinct things. The first is FewShotPapers, a paper list tracking research advances in FSL. The second is PaddleFSL, described in the README as a PaddlePaddle-based Python library for FSL. These serve different audiences. Someone writing a literature review wants the paper list. Someone training a prototypical network on PaddlePaddle wants the library. The repository puts both behind one entry point, which means the top-level README tells you almost nothing about either in detail. The substantive documentation lives in the two subdirectories, and anyone evaluating the project should start there rather than at the root.
The split is worth naming because it shapes how you should judge the project. A paper list is judged on coverage and currency. A library is judged on API stability, dependencies and test coverage. FSL-Mate has a visible track record on the first and, from the material available, no visible track record on the second.
What FewShotPapers Actually Tracks
The news entries in the README are the clearest evidence of what the maintainers do. The most recent entry, dated 2026-04-18, adds FSL papers published at EMNLP 2025, ICCV 2025, ICML 2025, WWW 2026 and AAAI 2026. An earlier entry from 2025-10-14 covers CVPR 2025, ACL 2025 and IJCAI 2025. A third, from 2025-07-17, is broader: CVPR 2024, ICML 2024, IJCAI 2024, ACL 2024, NeurIPS 2024, EMNLP 2024, ICCV 2024, ICLR 2025, WWW 2024-2025, KDD 2024-2025, AAAI 2024-2025, NAACL 2024-2025 and SIGIR 2024-2025.
That pattern tells you the list is organised by venue and refreshed in batches rather than continuously. If you are working in computer vision or NLP and want to see what appeared at the major conferences in the last two cycles, the coverage is broad across those families. What the README does not state is how papers are selected, whether there is a taxonomy, or whether entries carry annotations. Those are the questions that decide whether a paper list saves you time or just moves the search elsewhere. The repository topics include deep-learning, few-shot, meta-learning, one-shot-learning and low-shot, which suggests the scope is the standard FSL family rather than a narrow subfield.
PaddleFSL and the PaddlePaddle Constraint
PaddleFSL is the part of FSL-Mate that runs code. The README gives one sentence for it: a PaddlePaddle-based Python library for FSL. That single adjective is the most consequential fact about the library. PaddlePaddle is Baidu's deep learning framework, and a library built on it inherits its tensor API, its data loading conventions and its installation path. If your existing training code is PyTorch, adopting PaddleFSL means either running two frameworks side by side or porting your pipeline.
The repository topics include paddlepaddle alongside the FSL terms, which is consistent with the README rather than adding to it. Beyond that, the supplied material does not describe the library's module layout, its supported algorithms, its dataset loaders or its evaluation protocol. The README points to the PaddleFSL subdirectory for details, and that is where any serious evaluation has to happen. Treat the top-level description as a pointer, not a specification.
Getting It Running Means Reading the Subdirectory
The top-level README contains no installation command, no pip invocation and no configuration keys. What it does contain is two links into the repository: github.com/tata1661/FSL-Mate/tree/master/FewShotPapers and github.com/tata1661/FSL-Mate/tree/master/PaddleFSL. The practical route is to clone the repository and read the README inside the subdirectory you care about:
git clone https://github.com/tata1661/FSL-Mate.git cd FSL-Mate/PaddleFSL
The default branch is master, so a clone without arguments lands on the right branch. If you only want the bibliography, the FewShotPapers directory is a document collection and needs no environment at all. If you want the library, expect to install PaddlePaddle first, since PaddleFSL depends on it. The top-level README does not state a supported Python version, a PaddlePaddle version floor, or a package name for installation. Those details, if they exist, are in the PaddleFSL README, and you should confirm them before writing any setup script.
No Licence File in the Repository
The licence is listed as unknown in the repository metadata, and the README does not state one. That is a real constraint, not a formality. Without a licence file, the default copyright position applies: others have no granted right to copy, modify or redistribute the code, regardless of it being publicly visible. For a paper list, the practical risk is low. For PaddleFSL, it matters if you plan to vendor the code into a product, ship it in a container image, or fork it for internal use.
This is a description of the repository state, not legal advice. If your organisation has a policy on dependencies without declared licences, FSL-Mate will trip it. The remedy is an issue on the repository asking the maintainers to add a licence file, which is the same channel the README names for feedback.
Where FSL-Mate Is the Wrong Choice
If you need a few-shot learning library that runs on PyTorch, FSL-Mate's library half is the wrong tool, and the mismatch is structural rather than a matter of missing features. The PaddlePaddle dependency is stated in the README itself, so there is no ambiguity to resolve. You would be adopting a second deep learning framework to get FSL utilities.
The second case is benchmarking. The README describes a paper list and a library. It does not describe a benchmark suite, a leaderboard, or a set of standardised evaluation splits with published numbers. If your goal is to compare methods under a fixed protocol, FSL-Mate does not offer that, at least not in anything the supplied material shows. A paper list tells you what was published; it does not tell you which method wins on a given split.
The third case is production dependency. With no declared licence and no release artefacts retrieved, pinning FSL-Mate to a version is not possible in the way it is for a package published to an index. You would be tracking a branch.
The Alternative: learn2learn and the PyTorch Ecosystem
The obvious alternative for the library half is learn2learn, a PyTorch-based meta-learning library. The difference in approach is not feature parity, it is framework gravity. learn2learn targets PyTorch, so it composes with the optimisers, data loaders and model definitions that most FSL research code already uses. PaddleFSL targets PaddlePaddle, so it composes with that stack instead. Choosing between them is mostly choosing which framework you are already committed to.
For the paper list half, the alternative is not a library at all. It is a maintained survey or an arXiv listing filtered by category. The trade-off is that a venue-organised list like FewShotPapers gives you conference coverage in one place, while an arXiv feed gives you recency without curation. FSL-Mate's list sits between those poles: curated by venue, updated in batches, with the batch dates visible in the README.
There is also a citation angle. The README asks users to cite Wang, Yao, Kwok and Ni, "Generalizing from a few examples: A survey on few-shot learning", ACM Computing Surveys 53(3), 2020. If you use the paper list in a survey of your own, that citation is the one the maintainers request.
Maintenance Cadence and What to Verify First
The README's news section is the maintenance record. Three dated entries appear in the supplied material, spanning 2025-07-17, 2025-10-14 and 2026-04-18. The gaps are roughly three months and six months. That is a plausible cadence for a venue-driven paper list, since conference notification cycles cluster. It tells you nothing about the library's update frequency, because none of the news entries mention PaddleFSL changes.
That asymmetry is the thing to verify. Before depending on PaddleFSL, check the commit history of the PaddleFSL subdirectory specifically, not the repository as a whole, because paper-list commits will dominate the top-level activity feed. Check whether the library's README documents a supported PaddlePaddle version and whether that version is still current. Check whether there are tests. And check whether a licence file has been added since this review, since that single file changes what you are allowed to do with the code. The contact route the README gives is an issue on the repository or an email to Yaqing Wang at wangyaqing@bimsa.cn.
Editorial conclusion
Adopt FSL-Mate if you need a single place to scan few-shot learning papers across CVPR, ICML, NeurIPS, ACL and adjacent venues, or if your training stack is already PaddlePaddle and you want FSL utilities that match it. Do not adopt it as a framework-agnostic benchmark suite: the library half is tied to PaddlePaddle, and the repository ships no licence file, so redistribution terms are unresolved until the maintainers add one. Before you build on PaddleFSL, clone the repository, read the subdirectory README, and check whether the API surface you need is documented there rather than in the top-level file, which describes the project only at the level of two bullet points.
Community notes