What is web-based interactive research paper presentation?¶

This is a extended web version of the journal article titled Combined neural network and adaptive DSP training for long-haul optical communications to be published in Journal of Lightwave Technology with DOI:10.1109/JLT.2021.3111437.

Here we have open-sourced the data we obtained from our experimental setup and you can run all the (well-documented) codes to reproduce the results presented in the paper. You can also select and display different results and view it from different angles.

What’s the problem with regular research papers presented in static pdfs? Why bother making a web-based interactive version?¶

Despite the time-tested robustness of PDF-based article, there are opinions and complains about this traditional media:

much information is lost in the final results, which might lead to misinterpretions, and much time is wasted just to understand its content properly unless one is very familar with that topic.
to respect “Code as a Research Object” initiative, it’d be great to see a direct data-computation-results reproduction within the article itself and unfortunately static papers just do not serve the purpose.
lack of continous revision (versioning) and citation including revison information.

Some other disciplines such as Math and Physics have long tried to extended their papers to use other multimedia to present the computation process. The so-called notebook interface is one of such alternatives which was invented 30 years ago and will soon become the standard interface of Mathematica!

Nevertheless, not everyone has a copy of Mathematica but web-browser is universal. There was a time when web-based presentation is bad for presenting research content due to its blur image-based equations, informal typesetting, lack of cross-referencing and other basic formatting functions. However, over the past few years, there are many open-sourced communities trying to make a difference in specific aspects:

equations: MathJax, KaTex
figure: Plotly, Altair, Bokeh
formating: Sphinx, MyST

You might never hear these names but they have been partially used in many different areas. In the recent Jupyter Book project, all these components are well integrated as part of Jupyter Book ecosystem which allow easy-to-code, effective, professional and interactive presentation based on all of these enabling tools.

While it is obvious that Optics and Photonics involve a lot more experimental work that are not meant to be ‘reproducible’ by someone else looking at the screen on the other side of the planet, open-sourcing the obtained data, enabling readers to understand their processing methodologies and try it themselves to improve reproducibility and readability are becoming a required component when presenting one’s work in all research disciplines. Therefore, it is important for us in Optics to also adopt this increasingly common trend in Science and Engineering research for the benefit of our own work and beyond. If the media to convey research ideas and results research paper is seen as a “communication channel”, we are trying our best to maximize its “capacity” by using new and effective tools.

In addition to the basic formating with classic PDF article, you can expect more features from this material:

with the source codes written as Jupyter notebooks, you can reproduce the results via cloud runtime (needs Colab access)
additional Authors’ remarks are present alongslide the main content
in-page reader comments/responses
interactive figures
better cross-references
equations with LaTex source codes
continious improvements/revision of the content

Source codes¶

At source notebook page, you can execute the source codes directly in Colab by hovering on at the top bar and choose Colab. You may also download the notebook by hovering on and click .ipynb so that you can run them locally.

To improve readability, common codes are placed in the gdbp package, which are refered by the source notebooks.

Datasets¶

The data access is part of the source codes, users don’t have to download the dataset seperately.

Such convenience is achieved by our efficient data APIs, please refer to the dataset repo for more information.

Figures¶

Most figures in this article are reproduced into interactive graphs. Below is an example.

You can:

drag to zoom in and double click to resume
move the cursor around to see the number indicators
single click on a legend item to hide that trace, double click to isolate other traces
see more options in the image toolbar

Equations¶

Equations rendering is done by MathJax, which allows checking \(\LaTeX\) commands in the right click menu of each equation.

Feedbacks¶

You are welcome to comment via:

selecting and annotating the contents (click the highlighted to reveal the existed comments). All the public annotations are visible by default, you may hide them by clicking button at the top right of each page. You need an account from Hyposhsis to leave comments.
raising issue or commenting in the discussion panel in the Github repository ( link at the top).

Cite¶

@article{fan2021combined,
  title={Combined neural network and adaptive DSP training for long-haul optical communications},
  author={Fan, Qirui and Lu, Chao and Lau, Alan Pak Tao},
  journal={Journal of Lightwave Technology},
  year={2021},
  publisher={IEEE}
}