lixus7/Time-Series-Works-Conferences: A Task-Grouped Reading List for Time-Series Papers

What the repository actually is

This is a Markdown paper list, not software. The README describes it as a "Time-Series Work Summary in CS Top Conferences (NIPS, ICML, ICLR, KDD, AAAI, WWW, IJCAI, CIKM, ICDM, ICDE, etc.)", and the primary language field in the repository metadata is unknown, which is consistent with a collection that is mostly prose and links. There is no package to install, no API to call, and no release has been published. The MIT licence covers whatever text and code exist in the repository, but the papers it points to carry their own licences and terms, and the material here says nothing about those.

The author identifies himself as Du Yin, a Ph.D. student in the School of Computer Science and Engineering at UNSW Sydney, co-supervised by Flora Salim and Hao Xue. That matters for how you read the list. It is one researcher's working index, kept for his own reading and shared publicly, not a committee-reviewed survey. The topics list attached to the repository (accident-detection, anomaly-detection, demand-forecasting, multivariate-timeseries, spatio-temporal-prediction, traffic-prediction, travel-time-prediction and others) reads like the set of areas a spatio-temporal and time-series researcher would track, which is a narrower and more coherent scope than a general machine learning paper dump.

The problem it solves: venue-first reading lists are hard to use

Conference proceedings are organised by venue and year. If you are working on time-series imputation, you do not want to walk ICLR, KDD, AAAI and ICDM year by year filtering for the subset that touches missing values. This repository inverts that order. The README's table of contents lists task headings directly: Multivariat Time Series Forecasting, Multivariat Probabilistic Time Series Forecasting, Time Series Imputation, Time Series Anomaly Detection, Demand Prediction, Time Series Generation, Travel Time Estimation, and Traffic Location Prediction, with the last entry truncated in the supplied text. Each heading is an anchor, so the intended navigation is a jump from the contents list to the task section.

The intended user is a graduate student or applied researcher doing a targeted literature scan. That is a real need. A task-first index shortens the time between deciding to read about imputation and having a list of candidate papers, and it makes gaps visible: if a task section is thin, that is information about the state of the field as captured by this list, not just about the list.

How the index is organised: task sections plus a commented-out shorthand table

The visible structure is a single README with anchor links. The top-level contents list points to task sections, and the metadata shows a second anchor, #Conferences, that the README refers to in a commented-out block as "the Conferences page with more conference information". Because that block is commented out, the rendered README does not expose it, and the supplied material does not show what the Conferences page contains.

The same commented-out region holds a two-column abbreviation table mapping full names to short forms used in the list: Adaptive GNN to AGNN, Attention to Attn, AutoRegression(RNN,GRU,LSTM) to AR, Controlled Differential Equations to CDE, Contrastive Learning to CL, Encoder Decoder to EncDec, Ensemble to Ens, Feature Decomposed to FeaD, Federated Learning to FL, Generative Adversarial Network to GAN, Graph Convolutional Network to GCN, Hour, Day, Week, Month to HA, Heterogeneous GNN to HGNN, Multiple Graph to MGNN, Memory to Mem, Meta Learning to MetaL, MultiTask to MulT, Network Architechture Search to NAS (the README's spelling), Ordinary Differential Equations to ODE, Statistic to Stat, TCN (WaveNet) to TCN, Temporal Graph Network to TGN, Transformer to Trans, Transfer Learning to TransL, and Variational Auto-Encoder to VAE. The README states that these abbreviations exist "to reduce repetition" and warns that "some terms may not represent general interpretations and apply only to this repository".

That warning is the most important sentence in the file for anyone who wants to read the list quickly. A tag like Trans or FeaD is a local shorthand, not a standard taxonomy. If you skim entries by abbreviation without reading the table, you will misread the method labels. Because the table sits inside an HTML comment in the version supplied, a reader who only looks at the rendered page may never see the definitions at all. That is a documentation defect worth knowing about before you trust the tags.

Getting it running: clone, read, and open issues

There is nothing to build. The workflow is a clone and a text search. To get a local copy:

git clone https://github.com/lixus7/Time-Series-Works-Conferences.git cd Time-Series-Works-Conferences

From there, grep is the practical interface. Searching for a task anchor or a method abbreviation in the README is faster than scrolling, and searching for a venue name such as ICLR or KDD lets you cross-cut the task grouping when you need the venue view instead. The repository has no homepage configured, so the README is the entry point.

The README also documents a contribution path rather than a configuration path. A commented-out note states that the task section "has been completed" while the methodology section will "continue to be updated", and asks readers who find missing resources or errors to "open an issue or make a pull request". The repository's issue tracker is linked from the badge block at the top, and the README also links to a GitHub Pages view and to a set of related repositories (xiyuanzh/time-series-papers, qingsongedu/awesome-AI-for-time-series-papers, xuehaouwa/Awesome-Trajectory-Prediction, and the author's own starred list). External mirrors of the paper collection are mentioned for OneDrive and Google Drive, with the README noting that a VPN may be required for the latter. Those mirrors are outside the repository, so their contents are not covered by the MIT licence and are not versioned alongside the list.

Where a curated list stops being enough

The main limitation is stated by the project itself. The methodology section is incomplete, and the abbreviation table that would make the method tags legible is commented out in the supplied README. A reader who arrives at the rendered page sees task groupings but not the definitions behind the short labels. That is a genuine usability gap, not a nitpick: the list's value depends on the tags being readable.

There is a second, structural limitation. A curated list is a snapshot of one person's reading. There is no automated ingestion from conference proceedings, no schema, and no release cadence. The repository metadata shows a last push of 2026-08-12 and no releases, so the only version you can cite is a commit hash. If you reference the list in a paper or a report, cite the commit you actually read, because main can move without a version marker.

Finally, the list does not ship code, datasets, or reported numbers. Nothing in the supplied material suggests it evaluates methods or reproduces results. If you need to compare two forecasting models on a shared dataset, this repository will point you at papers and stop there. The README's own framing is a summary of works, and the contribution instructions are about adding papers, not about running anything.

Alternatives and the difference in approach

The README itself names three sibling repositories, which makes the comparison concrete. xuehaouwa/Awesome-Trajectory-Prediction is the closest in scope but narrower: it covers trajectory prediction specifically, whereas this list spans forecasting, imputation, anomaly detection, generation, demand prediction, travel time estimation and traffic location prediction. If your question is trajectory prediction, the narrower list will have more depth per topic. If your question crosses tasks, this one is broader.

qingsongedu/awesome-AI-for-time-series-papers takes the general "awesome" approach: a broad AI-for-time-series collection. The difference is curation style rather than topic. This repository groups by task first and uses a local abbreviation scheme, which makes it faster to scan by problem but harder to read without the key. The general list is easier to read cold but less opinionated about how the material is sliced.

There is also a category difference worth naming. Paper lists like this one are not the same tool as a survey article or a benchmark suite. A survey gives you synthesis and comparison; a benchmark suite gives you numbers on shared data. This repository gives you neither. It gives you addresses, grouped by task, maintained by one person. That is a legitimate and useful thing, but it is a different thing, and treating it as a benchmark because it mentions KDD and ICML would be a category error.

Maintenance, licence, and what to verify before you cite it

Maintenance is manual and personal. The README's backlog section is marked "To do", the methodology section is described as ongoing, and the contribution model is issues and pull requests from readers. There is no stated review process, no CI, and no release. For a paper list that is normal, but it means the freshness of any given task section is unknown from the README alone. You cannot tell from the supplied material how far back the coverage runs or how consistently each task section is updated.

The licence is MIT, which is permissive for the repository contents. That does not extend to the linked papers, the OneDrive and Google Drive mirrors, or the sibling repositories, each of which carries its own terms. Nothing here is legal advice; if you plan to redistribute the list or the mirrors, check the source terms yourself.

Before citing or relying on an entry, verify three things: the paper exists at the venue stated, the entry's method tag matches the abbreviation table (and remember the table warns that terms are local to this repository), and any code link resolves. Because there are no releases, pin a commit when you cite the list. The repository is a starting point for a literature review, and its accuracy is only as good as the last person who opened an issue about a broken link.

Editorial conclusion

Use this repository if you are starting a literature review in multivariate forecasting, imputation, anomaly detection, demand prediction, generation, travel time estimation or traffic location prediction, and you want the entry points grouped by task instead of scattered across venue proceedings. Do not use it as a benchmark harness, a dataset registry, or a source of reported numbers; nothing in the material suggests it ships code, and the README states the methodology section is still being completed. Before relying on any entry, open the linked paper, confirm the venue line in the README matches the published record, and check whether the entry carries a code link. The repository is MIT licensed and has no releases, so pin the commit you read rather than tracking main.

Official sources

1. Issues
2. License: MIT
3. lixus7/Time-Series-Works-Conferences on GitHub
4. README