IPOL MLBriefs Workshop
Welcome to the 4th edition of the IPOL MLBriefs workshop
May 27th — May 31st, 2024.
A BPI Sesame OVDSaas project, sponsored by the DataIA Institute
In many fields of applied mathematics and computer science (computer vision, machine learning, image/signal processing, …), new methods and applications are appearing at a vertiginous pace, to the point it is hard to keep up with all new methods. Yet, being able to understand and try these methods oneself is crucial for research.
To bridge this gap and promote reproducible research, during the IPOL MLBriefs workshop we will help you use the IPOL journal and demo system to create and publish an online demo of a method with a companion technical article describing the method, experimentally highlighting its use cases and limitations, and providing new insight onto the method.
This workshop will be also be a great opportunity to meet fellow researchers and attend plenary talks from invited speakers.
News: Recordings are live! Please see the tutorials and invited talks sections, or our YouTube channel.
Important Dates
The workshop was held between May 27th and May 31st 2024.
The submission deadline is June 30th, 2024.
Program
The conference program will be announced later.
Talks and tutorials took place in the morning, Monday to Thursday. Please find below the program. On afternoons and on Friday, May 31st, participants gathered in the workshop rooms and work on their demos and articles.) Please see the tutorials and invited talks sections, or our YouTube channel for the recordings.
- Monday, May 27th
- Introduction to the workshop
- Invited presentation:Dr. Christian Sandor, head of team ARAI (LISN, Université Paris-Saclay), researcher in Augmented Reality.
- Tutorial: How to create your first IPOL demo?
- Tuesday, May 28th (9.30 - 12.30, Amphithéâtre Hodgkins)
- Invited presentation: Dr. Tomasz Bednarz, Director of Strategic Researcher Engagement at NVIDIA Corporation
- Tutorial: Turn your code into a full-fledged online demo
- Wednesday, May 29th (9.30 - 12.30, Amphithéâtre Hodgkins)
- Invited presentation: Dr. Gaël Varoquaux, head of team INRIA Soda, director of the scikit-learn operations at Inria Foundation, researcher in AI, statistics and health
- Thursday, May 30th (9.30 - 12.30, room Amphithéâtre Hodgkins)
- Tutorial by Pablo Musé, Managing Editor of IPOL: writing an IPOL article on reproducible research
- Questions and answers
- How to submit your work to IPOL?
- Demo of the participants’ works
Tutorials
Invited speakers
We are happy to announce our three invited speakers:
Christian Sandor: On the Convergence of Augmented Reality and Artificial Intelligence
Abstract
Augmented Reality (AR) embeds spatially-registered computer graphics into a user’s view of the real world. During the last 23 years, my research focus has been on AR. Since 2018, I have been studying the applications of AI, particularly Deep Learning, to Visual Computing. In April 2024, I started a new research team to investigate the convergence of AR and AI. This talk consists of two parts. First, I present my previous work on AR and AI. Second, I introduce our current thoughts on why AR+AI will change the world forever.
Biography
Dr. Christian Sandor is a Professor at Université Paris-Saclay and the leader of the ARAI team at LISN, CNRS (Centre National de la Recherche Scientifique). Since the year 2000, his foremost research interest is Augmented Reality, as he believes that it will have a profound impact on the future of mankind.
In 2005, he obtained a doctorate in Computer Science from the Technische Universität München, Germany under the supervision of Prof. Gudrun Klinker and Prof. Steven Feiner. He decided to explore the research world in the spirit of Alexander von Humboldt and has lived outside of Germany ever since to work with leading research groups at institutions including: The University of Tokyo (Japan), Nara Institute of Science and Technology (Japan), Columbia University (New York, USA), Canon’s Leading-Edge Technology Research Headquarters (Tokyo, Japan), Graz University of Technology (Austria), University of Stuttgart (Germany), University of South Australia, City University of Hong Kong, and Tohoku University (Japan).
Together with his students, he won multiple awards at the premier Augmented Reality conference, IEEE International Symposium on Mixed and Augmented Reality (IEEE ISMAR): best demo (2011, 2016) and best poster honorable mention (2012, 2013). Further awards include: best short paper at ACM SIGGRAPH International Conference on Virtual-Reality Continuum and its Applications in Industry (2018), best paper at ACM Symposium on Spatial User Interaction (2018), and best demo (audience vote) at ACM SIGGRAPH Asia XR (2023).
He has presented several keynotes and has acquired over 3 million USD funding. In 2012, he was awarded in Samsung’s Global Research Outreach Program. In 2014, he received a Google Faculty Award for creating an Augmented Reality X-Ray system for Google Glass. In October 2020, he was appointed Augmented Reality Evangelist at the Guangzhou Greater Bay Area Virtual Reality Research Institute. In 2021, Dr. Sandor was named “Associate Editor of the Year” by IEEE Transactions on Visualization and Computer Graphics.
He serves as an editorial board member for IEEE Transactions on Visualization and Computer Graphics and as a steering committee member for ACM Symposium on Spatial User Interaction. He has been program chair for numerous conferences, including IEEE ISMAR, ACM SIGGRAPH Asia XR, and ACM SIGGRAPH Asia Symposium On Mobile Graphics And Interactive Applications.
Tomasz Bednarz: Simulation, Modeling and (Generative) AI as Tools to Accelerate Scientific Discovery
Abstract
Science and simulation are inherently creative processes, and artificial intelligence (AI) has been with us for quite a long-time assisting us in effective and creative knowledge discovery and generative assets creation. AI has the capacity to educate and augment humans, and to make them better at their jobs and activities, including carrying out more effective scientific discovery or enhancing cultural experiences. This talk will highlight and show examples of how scalable computational, graphics, and generative methods plus AI-driven simulation techniques can generate new theories, uncover new knowledge, create art, reveal invisible, and assist e.g., in 3D reconstruction that can be applied across industries.
Biography
Dr. Tomasz Bednarz is a Director of Strategic Researcher Engagement at NVIDIA Corporation. His team at NVIDIA actively engages with top researchers and premier research institutions that do compelling, computationally intense work to solve some of the world’s most challenging scientific problems across research domains.
Earlier he was a Director and Head of Visualization at the Expanded Perception & Interaction Centre (EPICentre) at the UNSW, and Leader in Simulation and Modeling at CSIRO’s - Australian National Science Agency.
Over the last couple of years, he has been involved in wide range of projects in area of immersive visualization, human-computer interaction, computational imaging, image analysis and processing, accelerated computing, simulation, modeling, computer graphics, computer games, computational fluid dynamics, machine learning, artificial intelligence, and multi-sensors assimilation.
In 2019, he served as SIGGRAPH Asia 2019 conference Chair.
Gaël Varoquaux: Model evaluation, a machine-learning bottleneck
Abstract
While machine-learning papers always come with amazing promises, these do not always carry over to wider settings. One of the reasons for such shortcomings is that our evaluation is often detached from reality. As machine-learners love to go metric chasing, it is crucial that these metrics are chosen to reflect the application settings. In addition, conclusions of evaluation must also be informed the associated uncertainty on each model evaluation. In my talk, I will expose some seldom-discussed considerations on choice of metrics for simple classification problems, and on evaluation uncertainty, as well as why these considerations are important.
Biography
Dr. Gaël Varoquaux is a research director working on data science at INRIA (French computer science national research) where he leads the Soda Team.
Varoquaux’s research covers fundamentals of artificial intelligence, statistical learning, natural language processing, causal inference, as well as applications to health, with a current focus on public health and epidemiology. He also creates technology: he co-funded scikit-learn, one of the reference machine-learning toolboxes, and helped build various central tools for data analysis in Python.
Varoquaux has worked at UC Berkeley, McGill, and university of Florence. He did a PhD in quantum physics supervised by Alain Aspect and is a graduate from École Normale Superieure, Paris.
Call for Papers
Submissions consist of an online demo of a method, with a paper describing and analysing the method. Starting from a working code, you will be taught how to create a demo during the workshop. The companion paper describes the method and experiments on it.
- Contrary to many workshops, you are invited to prepare your submission during the workshop. You will be given time to submit the paper after the workshop ends. It is not expected that you submit the paper before the workshop starts.
- IPOL and MLBriefs publications usually act as companions to more traditional publications. It is allowed, and even encouraged, to submit an article covering an already-published method. Please check our authorship and double submission policy below for more information.
- Due to the growing number of submissions, all authors may be asked to review several papers.
Submission areas
Areas of submission are broad; the IPOL journal mainly focuses on (but is not limited to):
- Image and video processing
- Computer vision
- Machine learning
- Signal processing
Tracks
Papers can be submitted to different tracks. The IPOL tracks are open year-long (Check the IPOL website for more information), while the MLBriefs track is open until June 30th, 2024.
MLBriefs track
In this track, authors submit an online demo accompanied by a technical paper analysing and experimenting on the method. At the very least, the paper is expected to contain a description of the implemented method and its parameters, as well as experiments (both qualitative and quantitative) highlighting its strengths and shortcomings. The code that is used for the demo does not necessarily needs to be the authors’ code, as long as the licence of the code is respected.
Even though the method does not need to be new, we expect the submitted paper to be exciting to read in addition to, or instead of, the paper describing the original method. The contributions of the submitted article towards the method should be discussed within the introduction of the paper. Such contribution can include, but is not limited to:
- Survey: Rather than working on a single method, create demos for an array of different methods. The paper can then analyse how the methods differ, and how it can affect the results. The qualitative and quantitative experiments should be done comparatively, to explain the differing results, use cases, and limitations. If possible, those results should be linked to the theoretical analysis to explain why the results differ. The choice of compared methods should also be discussed. Full exhaustivity is of course not required — some methods have no published code, or the licence prevents use in IPOL —, but there should be some reasoning behind the choice of methods (e. g. comparing methods that follow a specific approach, or recent methods to further easy experimentation on them, etc)
- Simplification: Some published methods are very interesting, but also highly technical and complicated to read, either by themselves or due to low accessibility without specific prior knowledge. Yet, these methods can often be insightful for people from other domains, that do not have the required knowledge to read the method. Therefore, articles that take a complicated method and explain it in a more simple way, or from a different point of view, can be very interesting to many people.
- Analysis: On the other hand, some methods that are published work very well, but it is not understood why. Theoretical and experimental analysis of where and why the method works, to better specify the use cases and limitation of the method, are welcome.
- Method improvements and domain adaptation: When implementing a method, you can sometimes think of small changes or improvements that lead to better results or usability of the method in new domains.
IPOL track
The IPOL track is the traditional track of IPOL submissions, and exists independently of the MLBriefs workshop. The description of the method is expected to be exhaustive, and supported by pseudo-code descriptions. Emphasis is placed on the reproducibility of the work:
- If a model is trained, all information related to the data and the training should be provided in the article.
- The article should contain enough information to recode the method from scratch, producing the exact same results as the submitted code.
- In addition to the paper and online demo, the code will be peer-reviewed to ensure that the submitted code exactly matches the article description and pseudo-code.
The code is usually the authors’ work as well, although this is not a strict requirement.
The article is complemented by a rigorous experimental analysis highlighting the use cases, strengths and limitations of the method. Papers can also explain the method from an alternative viewpoint than the original article, provide a rigorous theoretical analysis explaining why some of the strengths and limitations appear, and/or improve the method or adapt it to new domains.
Due to this code peer-review requirement, note that IPOL submissions can only be considered if the code is submitted in Python, C, C++, Matlab or GNU Octave. If you want to submit code in other languages, please contact us first: submission may still be possible depending on the available reviewers.
IPOL dataset track
The success of algorithms depends on the quality of the data they process and on the reliable information about how these data were obtained. The algorithms evaluation should be systematically performed on high quality data shared among many researchers. Their properties have to be certified in the best possible way. For this reason, IPOL welcomes the submission of datasets, with companion articles explaining thoroughly how the data have been obtained, how they can be used and with all additional information that can help researchers to use them properly. The submissions will be evaluated by referees by the same procedure as for other IPOL articles.
Although not a method, an IPOL dataset submission can be complemented by a demo as well. Depending on the dataset, this demo might help explore the dataset, or help generate new data in the case of synthetic dataset.
Code and article licence and access
Articles published on IPOL are distributed under a Creative Commons CC-BY-NC-SA 3.0 licence. Users are free to copy, distribute, transmit and adapt the articles for non-commercial purposes if they attribute the work to the authors and maintain this license.
The codes are published along the demos under a free software licence chosen by the authors, usually GPL, AGPL, LGPL or BSD. As an exception, implementations of patented algorithms may be distributed with ad-hoc licencing terms.
There are no licencing fees or article processing charges for publishing to IPOL: publishing is free.
Dual submissions and authorship
IPOL authors do not necessarily need to be the inventors of the algorithms they publish. IPOL authors are the authors of the online demo, a technical description of the algorithm, experiments, and possibly of the code and implementation. IPOL publishes new, but also basic and well-known algorithms. IPOL therefore does not guarantee the novelty of the published algorithms, nor the novelty or validity of their underlying theory. When the submitted algorithm is new, IPOL encourages a parallel submission by the authors to classic journals. A submitted algorithm and code may use several other algorithms and computer programs. In that case, the authors must specify the part which is submitted to IPOL. Only this part will be evaluated.
IPOL welcomes submissions of algorithms based on theories submitted or already published in journal papers. In particular, joint submissions to IPOL and SIIMS (SIAM Journal of Imaging Science) are encouraged. Reviews are separate and independent, and acceptance/rejection of the submissions for one journal is not conditioned in any form either on the submission or acceptance/rejection on the other.
Upon acceptance, cross links between both electronic articles will be placed, so readers will be able to automatically navigate between the SIIMS and IPOL complementary materials.
Page limits
There is no set minimum or hard limit to the number of pages. Authors are free to write as long a paper as they deem reasonable. Note the following point:
- Longer papers are perfectly acceptable when the contributions of the paper are extensive enough to justify it, however papers should not be needlessly long. Note that very long papers can be difficult to read for the audiance, and will also result in longer review times. When the paper is long, it becomes even more crucial that it be well-organized. Question the relevance of each part. Relevant but minor results can also be placed in an annex to simplified reading.
- Papers are expected to provide a significant contribution. There can be excellent short papers, but papers without a meaningful contribution shall be rejected
- Because there is no hard limit to the number of pages, there is no reason to tamper with the margins, rephrase the sentences to make them shorter at the cost of them being illegible, or reduce the figure sizes. Please keep the paper legible.
Submission
Submissions for all tracks happen through the IPOL journal management system. You will need to create an account, do not use a gmail address for this account or you might not receive the emails. You can follow the tutorial below:
Organizing Committees
General MLBriefs organization
Quentin Bammey, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
Program Committee
They invite plenary speakers and organize the workshop program
- Gabriele Facciolo, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, Gif-sur-Yvette, France
- Pablo Musé, Universidad de la República, Montevideo, Uruguay
- Jean-Michel Morel, City University of Hong Kong (CityU), Hong Kong
- Quentin Bammey, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, Gif-sur-Yvette, France
Editorial board
They manage the reviewing process for the submitted papers
MLBriefs track
- Quentin Bammey, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
- Gabriele Facciolo, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, Gif-sur-Yvette, France
- Pablo Musé, Universidad de la República, Montevideo, Uruguay
IPOL track
- Jose-Luis Lisani, Universitat de les Illes Balears, Palma de Mallorca, Spain
- The IPOL editorial board
Technical Committee
They develop the IPOL demo system and provide technical help to the participants on demo creation
- Jérémy Anger, Kayrros SAS, France / Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
- Amyn Jaafar, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
- Héctor Macías Casado, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
- Quentin Bammey, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
- Miguel Colom, Centre Borelli, École Normale Supérieure Paris-Saclay, Université Paris-Saclay, France
Venue and Travel
The conference was held at the ENS Paris-Saclay, Gif-sur-Yvette, France.
All sessions will be transmitted by Zoom as well.
The ENS is easily accessible from the Paris city centre.
Coming from Paris
From the Paris city centre, take the RER B towards Saint-Rémy-lès-Chevreuse, get off at Le Guichet, then take the bus 9 towards Campus HEC / Gare de Jouy-en-Josas, and get off at Moulon, in front of the ENS. You can also get off the RER B at Massy-Palaiseau, take bus 91.06 or 91.10, and get off at Moulon.
Count between one hour and one hour and half from Châtelet les Halles to the ENS.
Coming by train
The ENS is best deserved by the Massy-Palaiseau TGV station, buses 91.06 and 91.10 go from the Massy-Palaiseau bus station (which is next to the train station) to Moulon (in front of the ENS) in 20 to 30 minutes.
If you arrive at any of the other Paris stations, first reach the RER B, then follow the instructions from the previous section.
Coming by plane
Paris is served by two major airports: Paris Charles de Gaulle and Orly.
The Charles de Gaulle airport is located directly on the RER B, from there you can follow the instructions of the first section “Coming from Paris”. The trip will take around 2 hours. Note that in May, the RER B will not operate between Châtelet les Halles and the airport in the evening, starting at 10.45 PM. If you arrive after this hour, you can take the RoissyBus shuttle, which operates between the airport and Paris.
From the Orly airport, take the RER C towards Massy - Palaiseau, stop there and take bus 91.06 or 91.10 to get off at Moulon. The journey will take around one hour.
Coming by car
There are several free car parks near the ENS, the closest being the “Parking Digiteo”. Note that as per university policy, we cannot grant you entry in the ENS car park, which is reserved for employees.
Sponsors
We are happy to announce the Institut DataIA as a sponsor of the fourth IPOL MLBriefs workshop. Thank you!
The IPOL MLBriefs workshop is a great place to meet young researchers. We expect more than 100 attendees from over the world.
If you would like to sponsor future editions of the IPOL MLBriefs workshop, please contact us.
Sponsor benefits include:
- Your name and logo on the MLBriefs website, as well as thanks during the introduction session.
- The possibility to give a plenary talk to the attendees, about the scientific work of your company,
- The possibility to distribute goodies and/or leaflets to the attendees,
Your sponsorship will help funding the organization of the workshop, possibly travel grants for international attendees, and foster the overall development of the IPOL platform.
Contact us
You can contact us at the following address (remove the AT, DOT and underscores): contact_AT_mlbriefs_DOT_com
