Why a new conference format?

What is "Deep Learning Update (DLU)"?

DLU is a new format for an online-only conference focusing on Deep Learning.
You do not only submit your paper, but also your presentation video.
Both your paper AND your talk will be reviewed by two independent reviewers.
The goal is not only to filter for very good papers,
but also to encourage submissions that provide an excellent presentation of their work in a talk.
If your paper AND your talk are of good to excellent quality, both will be published
on the DLU website in a digital lecture hall.

How did the idea for a new conference format come about?

Call for Papers

The DLU conference has a focus on work from
the field of Deep Learning.

We invite researchers from all areas of Deep Learning
to submit their work to DLU.

Important Dates

Submission Deadline for Paper/Videos:
15.01.2021 (Anywhere-On-Earth)

Notification of Acceptance: 15.02.2021

Opening of the Online-Conference: 15.03.2021

Paper/Talk Submission

You are interested in participating
in this new conference format?
You have an interesting contribution
in the area of "Deep Learning"?
Then read how to ...

FAQ

Do you have a question that is not answered
by the description of the submission process?

Here we compile frequently asked questions
regarding this new conference and
the corresponding answers.

Why a new conference?

Some time ago we (former PhD students in the field of machine learning; now in industry and on the road as professors) sat together in a bar and discussed how the current pandemic forces many conferences to make the respective meetings virtual.

We came to the conclusion that many conferences unfortunately make a big mistake here. They see the online version as an emergency solution and copy the old conference format in presence to the online format 1:1. Instead of live presentations at a conference venue, there are live presentations in Zoom and Co.

We believe that an online conference - if designed correctly - is not necessarily an emergency solution, but can offer many advantages that a classic conference does not.

How often have you ever been in a crowded conference room and realized that you are not interested in the presentation or that it is of low quality? But getting up now and leaving the lecture room could seem very rude. So you do waste your time and often wonder afterwards whether your participation in the conference was really efficient in terms of time and content.

How can we imagine a better approach?

Once you have entered a digital lecture hall you can choose between different topics (sessions): e.g.

"Deep Learning for Computer Vision"
"Deep Learning for Audio Processing"
"Deep Learning for Natural Language Processing"
"Deep Learning for Robotics"
etc.

You go through one of these portals to the respective session and come to an overview page of all talks that are thematically related to the session. For each presentation there is a 15-30 seconds short video by the authors. Are you interested? Then you can proceed to the actual talk. Not interested? Go to the short video of the next authors.

In our opinion, the actual video talks should be short nevertheless. Therefore they will be limited to 10min. The authors should limit themselves to the core ideas and core statements in their talk. The paper will also be made available here and will allow the audience to read the details and dive into the work properly. In future versions of DLU we plan to provide a discussion page for each paper and presentation.

We think that this will allow you to quickly and easily select presentations on exactly the new Deep Learning topics that you are interested in. So you can get your personal and individual "Deep Learning Update" on the latest research in your Deep Learning area.

Call for Papers

DLU (Deep Learning Update) is a new format for an online-only conference focusing on Deep Learning. Authors submit not only a paper but also a video of their talk. Both, the paper and the video, will be reviewed to filter for valuable content and a clear presentation of the corresponding research.

We invite researchers from all areas of Deep Learning to submit their contributions, for example contributions in the field of:

  • Deep Learning for Computer Vision
  • Deep Learning for Audio and Speech Processing
  • Deep Learning for Natural Language Processing (NLP)
  • Deep Learning for Robotics
  • Deep Learning for Time Series Analysis
  • Deep Reinforcement Learning
  • Deep Learning in Medicine and Bioinformatics
  • Deep Learning for Advanced Driver Assistance Systems and Autonomous Driving
  • Bio-Inspired Deep Learning Approaches
  • Recurrent Networks
  • Generative Models (e.g., Image generation, Text generation, etc.)
  • Cognitive Science and Neuroscience
  • Computational models of the brain
  • Deep Learning and Information Theory
  • Unsupervised Learning with Neural Networks
  • New Learning Approaches for Neural Networks
  • Optimization Techniques for Neural Networks
  • Compression of Neural Networks
  • Uncertainty Estimation for Neural Networks
  • Reservoir and Deep Reservoir Computing
  • Random Neural Networks
  • Binary Neural Networks
  • Hardware for acceleration of Neural Networks
  • Energy Efficient Neural Networks

Paper/Talk Submission

In the following we will describe the whole process of the conference from submission to publication. We have tried to keep the process as simple and transparent as possible:

Step 1: Submission

A valid submission for the DLU Conference consists of three parts:

1. Your paper paper.pdf with a minimum of 4 pages, maximum 10 pages. Maximum file size: 10 MB
2. A short video short_video.mp4 of minimum 15 seconds, maximum 30 seconds length. Maximum file size: 40 MB
3. A long video long_video.mp4 of at least 5 minutes, maximum 10 minutes length. Maximum file size: 400 MB

The reviewing process will be double-blind: You do not know the reviewers and the reviewers do not know you (you cannot see each other, for this, "double-blind"). For this, make sure, there is no textual information in your paper or in the videos that gives the reviewers a hint who you are (i.e. no author names shall be visible in the paper and in the videos).

Paper: In your paper you describe your research work as usual. Your paper must be created with the help of the following simple LaTeX template to guarantee a uniform appearance of all papers at the conference: latex_template_for_dlu_paper_submission.zip

Videos: If you prepare the videos make sure to turn-off your camera so that you cannot be seen. Your voice in the video, on the other hand, we regard as largely unproblematic for the reviewing process, i.e. you are allowed to give your own speech in the video as a speaker. It can be assumed that you are usually not identifiable to a reviewer because of your speaking voice.

The short video is about your main ideas and your research contribution in a very condensed way. It will be used later to help participants of the online conference to decide whether they want to read your paper and watch your talk.

In the long video, you will have a maximum of 10 minutes to present your research in an easily understandable way like in a face-to-face conference. Concentrate here on an easy-to-understand presentation that presents the core ideas of your work, your method or approach and your results. Participants may then read more details in your paper if they are interested.

Deadline: the deadline for submission is 15.01.2021 (AoE time - Anywhere on Earth)

What is the current "Anywhere on Earth Time? Here is an AoE clock.

How to submit the two videos and the paper?

We have tried various conference submission software. Our result was that classic conference submission software is not equipped to receive large files like videos as it is necessary for this conference. Therefore, we are now going down an uncomplicated path and use a proven service ("WeTransfer") to transfer larger files:

1. Pack your paper, your short and your long video into a single zip file submission.zip
The file size of this single zip file has to be equal or less than 450 MB. If it is larger, we will reject the submission.

2. Go to the website WeTransfer. No account will be needed.

3. Send us this zip-file via WeTransfer (choose the "email transfer" option, not the "link transfer" option) to the email address
dlu@deep-learning-update.org
and enter the contact email address of the contact author for your submission in the "sender address" field.

4. The DLU organizers will then receive a link that allows us to download this zip file. The DLU organizers will send you a submission ID (~1-2 days later) to the eMail address you have specified as the contact eMail-address. If you contact us on any matter in the following, always specify this submission ID in the subject field of your eMail.

Step 2: Reviewing Process

Each valid and complete submission will be reviewed. Two reviewers will be appointed to independently review all 3 entries: Your paper, your short video and your long video.

Each reviewer will send the program committee three scores on a scale of 1-10 (1=worst grade, 10=best grade): one score for your paper, one score for your short video and one score for your long video. All three entries (paper, short video, long video) will be scored individually.

So the program committee will finally receive 3 scores from each of the two reviewer:

How the final scores p3, s3, l3 are computed
Scores by Reviewer #1
(1-10)
Scores by Reviewer #2
(1-10)
Final score
(1-10)
Paper p1 p2 p3:=(p1+p2)/2
Short video s1 s2 s3:=(s1+s2)/2
Long video l1 l2 l3:=(l1+l2)/2
Your contribution is accepted for publication if and only if: p3>=7 AND s3 >=7 AND l3>=7.

So you have to make sure that not only your paper is good (7) to excellent (10), but also your short presentation in the short video, as well as your longer presentation in the long video.

Reviewing results with a final decision of the program committee regarding acceptance or rejection for each contribution will then be sent to all authors on 15.02.2021.

Step 3: Registration and Camera Ready Paper

Your contribution has been rejected? Sorry! Try a new submission for "Deep Learning Update 2" in the next year!
Your contribution has been accepted? Congrats! Now there are only a few steps left to publish it. You need to

1. Pay the registration fee ($300 per contribution; if there are several authors, it remains $300)
The conference fee is comparatively moderate compared to usual live-conferences, which take place at a conference location. Here we save the costs for a conference venue. Nevertheless many costs arise. The conference fee is mainly used to 1. finance the costs for the development of a professional interactive website through an agency, 2. finance the video post-processing costs (encoding), 3. finance the running server costs (webspace, etc.) However, since the proportionate costs of creating the conference website, video processing and maintaining the website are the same, regardless of whether the contribution is by one or more authors, only one of the authors needs to pay the registration fee.
2. Send us a signed written confirmation that the authors hold all copyrights in both the paper and the videos. We need the signatures of all the authors.
3. Send us a signed written permission that the paper and the two videos may be published on the conference website. Again, we need the signatures of all the authors.
4. Send us a camera ready paper camera_ready_paper.pdf file with all the authors specified in the paper.

Simply send us both the documents (2,3) and your paper (4) per eMail to dlu@deep-learning-update.org and again, do not forget to specify your submission ID in the subject field of your eMail.

Details about the payment process (1) and standard forms for 2 and 3 will be published exactly here soon.

Step 4: Opening of the conference

The online conference will be openend on 15.03.2021.

Then all contributions will be published in a digital conference hall (i.e., for each contribution both videos and the paper). Your two presentation videos will be available on the conference website for at least two years, completely free for everyone (OpenAccess). The papers will also be stored for a longer period of time and will remain freely available to the public (OpenAccess) on the DLU website.

FAQ

Q001: If the paper is excellent and gets e.g. 10 points in the review, but the video less than 7 points, is the submission really rejected?
A001: Yes, we would then reject the submission anyway. Even if it were the other way around: A submission with an excellent talk but bad paper would also be rejected. The reason is simply the concept for this new conference format: We do not only want to publish very good papers here, but also those that are very good at conveying this work in a nice talk.

Q002: So you would have rejected Einstein's theory of relativity as a submission, if his talk/video would have been difficult to understand?
A002: Yes, exactly! If Albert's talk was bad, we would have rejected his submission. There are also other reasons why we would not accept such a work at the DLU: the topic has nothing to do with Deep Learning. Nevertheless: Great paper, Albert!

Q003: Can I submit my paper first and if the paper is good according to the reviewers (>=7 points), then record the presentation video and send it to you?
A003: No, the review process should be as simple as possible and should therefore be a one-step process. It is also important for the reviewers to have both the paper and the presentation available. It allows the reviewers to determine whether the presentation really summarizes the main ideas and results of the paper.

Q004: How shall I record my video / my talk?
A004: With any software you like. We recommend OBS (Open Broadcaster Software) which "is a free and open-source cross-platform streaming and recording program built with Qt and maintained by the OBS Project." (source: Wikipedia).

Q005: I have a technical question regarding OBS. Can you help me?
A005: No, sorry! We try to concentrate on the organization of this conference. But you can find many sources in the WWW, e.g. YouTube video tutorials, which discuss how to use OBS for recording your screen and your voice.

Q006: How shall the videos be structured?
A006: We have discussed this in the organisation team whether we would like to make specifications here. Our conclusion: We deliberately do not want to make any specifications here so that the presentations are also different and varied. So you can give a classic slide presentation and say something about each slide in the video or you can choose another way of presentation, e.g., create a drawing step by step directly in the video to explain your approach, your experiments etc. Maybe you want to show animations, video results etc. You are also welcome to give your presentation as a rap if you think that you can present the problem, your contribution, your approach and your results well with such a type of a talk.

Q007: What shall be the resolution of the videos?
A007: We recommend using a resolution of 2560x1440. But you are free to use other video resolutions as well. Just use a resolution that allows your presentation to be easily comprehensible.

Q008: Which container and encoding format shall be used for the videos?
A008: Please use .mp4 as a container format for the videos. For encoding of the video please use the H.265 video compression standard.

Q009: What is the maximum file size for the file submission.zip that shall be submitted via WeTransfer?
A009: Please make sure that your paper.pdf has a size of less than 10 MB, your short_talk.mp4 video file has a size of less than 40 MB and your long_talk.mp4 video file has a size of less than 400 MB. The maximum file size of the submission file submission.zip which contains the paper and the two videos is therefore 450 MB. Submissions which are larger will be rejected automatically.