PhD Seminars https://phd-seminars-sam.inria.fr Reseach for the PhD candidates by the PhD candidates! Tue, 12 Mar 2024 16:55:34 +0000 en-US hourly 1 https://wordpress.org/?v=5.9.7 https://phd-seminars-sam.inria.fr/files/2019/10/cropped-PhD_Seminars_Logo-32x32.jpg PhD Seminars https://phd-seminars-sam.inria.fr 32 32 PhD Seminar – 18 March 2024 https://phd-seminars-sam.inria.fr/phd-seminar-18-march-2024/ Tue, 12 Mar 2024 16:53:29 +0000 https://phd-seminars-sam.inria.fr/?p=4079 Continue reading

]]>
This time two shorter presentations:

Talk by Emilie Yu (GRAPHDECO team)
Doing a 3-months visit in Canada with MITACS
from plans to final research project

Abstract: Today I will talk about my visit of a research lab in the University of Toronto during three months funded by the MITACS Globalink Research Award. I will give practical information and tips about my experience in this program. Then I will briefly expose the final results from the project, to show how this international collaboration and visit was helpful to my research about VR painting:
The ability to represent artworks as stacks of layers is fundamental to modern graphics design. Despite their ubiquity in 2D painting software, layers have not yet made their way to VR painting, where users paint strokes directly in 3D space by gesturing a 6-degrees-of-freedom controller. So what should 3D layers be?
We propose a practical implementation of 3D-Layers integrated in a VR painting application, and ask professional VR artists to create beautiful paintings with it.

Talk by Nicolas Rosset (GRAPHDECO team)
Interactive design of 2D car profiles with aerodynamic feedback

Nicolas Rosset

Abstract: The design of car shapes requires a delicate balance between aesthetic and performance.
We describe how to train a model on instantaneous, synchronized observations extracted from multiple pre-computed simulations, such that we can visualize and optimize for dynamic flow features, such as vortices. Furthermore, we architectured our model to support gradient-based shape optimization within a learned latent space of car profiles.

When: Monday, Mar 04 at 2pm
Where: Euler Violet

]]>
PhD Seminar – 04 March 2024 https://phd-seminars-sam.inria.fr/phd-seminar-04-march-2024/ Thu, 29 Feb 2024 14:02:33 +0000 https://phd-seminars-sam.inria.fr/?p=4014 Continue reading

]]>
Scientific talk by Jose Daniel Galaz Mora (LEMON team)
Coupling methods of phase resolving coastal wave models

Jose Daniel Galaz Mora

Abstract: Wave propagation has a central role in beach evolution, sediment transport and the impact that natural and artificial structures have on the environment. Modeling these waves with 3D models gives the most accurate results but it is too complex and costly for real applications, so simpler 2D depth-averaged models are used instead. Among these, the Saint-Venant and Boussinesq equations stand out for their complementary ability to capture wave behavior in different conditions. In the last 15 years, combining these into a “hybrid model” by simply switching between them based on wave conditions has showed great promise. Yet, this approach led to mesh-convergence issues, instabilities and spurious waves that put in question its robustness and reliability. In this presentation I will first introduce new coupling methods that we explored aiming to solve these problems. I will show how these methods present similar issues as the hybrid model, which motivates a deeper study of the properties of the model. Then, I will present a mathematical analysis that shows that the linearized model is indeed well-posed in the sense of Hadamard, and that the oscillations observed in these models, at the PDE level, correspond to artificial reflections whose size depends on the dispersiveness of the waves and media. These results contribute to a better understanding for the development of robust depth-averaged water wave models.

When: Monday, Mar 04 at 2pm
Where: Euler Violet

]]>
PhD Seminar – 19 February 2024 https://phd-seminars-sam.inria.fr/phd-seminar-19-february-2024/ Tue, 13 Feb 2024 13:33:39 +0000 https://phd-seminars-sam.inria.fr/?p=3975 Continue reading

]]>
Scientific talk by Nicolas de Almeide Martins (COATI team)
Recent Results in Graph Coloring Games

Abstract: The graph coloring game was first introduced, in the context of graph theory, by Bodlaender in 1991. In such a game there are two players, Alice and Bob. The players are given a graph G and a set of colors C={1,2,…,k}. Starting with Alice, each player, alternately, must choose a vertex v and a color in C to apply to v. The colored vertices must always induce a proper coloring of G. Alice wins if all vertices are colored and Bob wins otherwise. The minimum k such that Alice has a winning strategy with k colors is called the game chromatic number of a graph, denoted by 𝛘_g(G). Boblaender mentioned that the complexity of such a game was an interesting open question.
Recently, Costa et al. proved that, given a k, deciding if 𝛘_g(G) ≤ k is PSPACE-complete, answering the question left open by Bodlaender. After this result, a number of variations of the coloring game were also proved to be PSPACE-complete. In this seminar, we shall present these recent results and introduce a new variation for which the complexity is still an open problem.

When: Monday, Feb 19 at 2pm
Where: Euler Violet

]]>
PhD Seminar – 05 February 2024 https://phd-seminars-sam.inria.fr/phd-seminar-05-february-2024/ Wed, 31 Jan 2024 10:00:38 +0000 https://phd-seminars-sam.inria.fr/?p=3914 Continue reading

]]>
Scientific talk by Louis Ohl (MAASAI team)
Generalised mutual information (GEMINI) –
A constellation of discriminative clustering models

Louis Ohl

Abstract: Clustering is a fundamental learning task which consists in separating data samples into several groups, each called a cluster. The last decade witnessed increasing successes in the field of deep clustering that tries to tackle clustering with neural networks. Often belonging to the family of discriminative models, these methods used mutual information (MI) as an unsupervised objective with increasing regularisations. In this presentation, we focus on the discriminative modelling and question the relevance of MI as a clustering objective. To adress these limitations, we developed a novel objective function: the generalised mutual information (GEMINI). We then show how this objective can be combined with various architectures from deep neural networks to decision trees, even with sparsity constraints for feature selection.

When: Monday, Feb 05 at 2pm
Where: Euler Violet

]]>
PhD Seminar – 22 January 2024 https://phd-seminars-sam.inria.fr/phd-seminar-22-january-2024/ Thu, 18 Jan 2024 13:21:58 +0000 https://phd-seminars-sam.inria.fr/?p=3891 Continue reading

]]>
Scientific talk by Tom Szwagier
Multilevel Machine Learning with Flags

Tom Szwagier

Abstract: In this talk, I will try to democratize a geometrical object that has not been much investigated in Machine Learning yet. A flag is a sequence of nested subspaces of increasing dimension. Flag spaces unify and generalize Grassmannians and Stiefel manifolds which are heavily used in Machine Learning and Computer Vision. Many dimension reduction methods like Principal Component Analysis (PCA) can be rethought as the search for a flag of linear subspaces that best and best approximate the data. This multilevel point of view enriches the previous methods which usually work at a fixed intrinsic dimension that is often unknown or not even well-defined. Alternatively, a flag can be seen as a sequence of mutually orthogonal subspaces and therefore provides a natural parameterization for the eigenspaces of symmetric matrices. In a recent paper, called Stratified PCA, we investigate parsimonious latent variable generative models extending Probabilistic PCA and parameterized with flags. We show that the covariance eigenvectors are not the best features to analyze the data variability when the number of samples is limited and the eigenvalue gaps are small. Instead, we recommend grouping them into flags of eigenspaces.I hope that after this talk you will see flags everywhere in your daily research!

When: Monday, Jan 22 at 2pm
Where: Euler Violet

]]>
PhD Seminar – 18 December 2023 https://phd-seminars-sam.inria.fr/phd-seminar-04-december-2023/ Tue, 28 Nov 2023 14:03:21 +0000 https://phd-seminars-sam.inria.fr/?p=3867 Continue reading

]]>
1. Flash talk – Presentation of the MACBES Team

2. Scientific talk by Riccardo Taiello (EPIONE team)
Let Them Drop: Scalable and Efficient Federated Learning Solutions Agnostic to Client Failures

Riccardo Taiello

Abstract: Secure model aggregation is nowadays recognized as the key component for Federated Learning (FL). It enables the collaborative training of a global machine learning model without leaking any information about FL clients’ local models. It is shown that clients who fail to complete the protocol, referred to as dropped clients, can seriously affect the correct computation of the global machine learning model. While the literature counts numerous fault-tolerant secure aggregation protocols that use secret sharing to reconstruct the inputs of dropped clients, the performance of these solutions decreases with an increase in the dropout rate. In this paper, we propose Eagle, a fault-tolerant, secure aggregation solution that is agnostic to client failures and therefore outperforms existing solutions. Eagle is inherently compatible with realistic FL schemes that implement client selection. Furthermore, existing state-of-the-art solutions usually apply to basic FL settings whereby all clients are synchronized. We propose Owl a secure aggregation solution suitable to the asynchronous setting. We have implemented both solutions and show that: (i) in a cross-device scenario, with realistic dropout rates, Eagle outperforms the best SA solution in SyncFL, namely Flamingo, by x4, approximately (ii) whereas in the asynchronous setting, Owl exhibits the best performance in all scenarios compared to the state-of-the-art solution LightSecAgg (by at least x10). During the final five minutes of the presentation, we will showcase a brief demonstration highlighting a practical implementation of secure aggregation for federated learning within the Fed-BioMed framework.

]]>
PhD Seminars – 27 March 2023 https://phd-seminars-sam.inria.fr/phd-seminars-27-march-2023/ Wed, 22 Feb 2023 08:10:36 +0000 https://phd-seminars-sam.inria.fr/?p=3616 Continue reading

]]>
Scientific talk by Yingyu Yang (EPIONE team)

Flash talk by Enrico Fiasché – Presentation of the ACENTAURI Team!

]]>
PhD seminars – 13 March 2023 https://phd-seminars-sam.inria.fr/phd-seminars-6-march-2023/ Wed, 22 Feb 2023 08:08:57 +0000 https://phd-seminars-sam.inria.fr/?p=3614 Continue reading

]]>
How to give a good scientific presentation! – Dr Arnaud Legout ]]>
PhD seminars – 27 february, 2023 https://phd-seminars-sam.inria.fr/phd-seminars-27-february-2023/ Wed, 22 Feb 2023 08:07:47 +0000 https://phd-seminars-sam.inria.fr/?p=3612 Continue reading

]]>
Flash presentation: Yanis Aeschlimann from Cronos team!

Introducing MOMI – 2023: Smart Environment

Scientific talk – Younes Ben Mazziane (Neo) – Analyzing Count Min Sketch with Conservative Updates

Younes Ben Mazziane

Abstract: Count-Min Sketch with Conservative Updates (CMS-CU) is a popular algorithm to approximately count items’ appearances in a data stream. Despite CMS-CU’s widespread adoption, the theoretical analysis of its performance is still wanting because of its inherent difficulty. In this paper, we propose a novel approach to study CMS-CU and derive new upper bounds on both the expected value and the CCDF of the estimation error under an i.i.d. request process. Our formulas can be successfully employed to derive improved estimates for the precision of heavy-hitter detection methods and improved configuration rules for CMS-CU. The bounds are evaluated both on synthetic and real traces.

]]>
PhD seminars – 6 February, 2023 https://phd-seminars-sam.inria.fr/phd-seminars-6-february-2023/ Wed, 22 Feb 2023 08:07:17 +0000 https://phd-seminars-sam.inria.fr/?p=3610 Continue reading

]]>
Round table: The potential careers at inria after the PhD

We don’t always know what we want to do after our PhD. This round table is meant to present the potential positions you can held at INRIA after your PhD.

First Manuel Serrano, from the Comité d’Evaluation (CE), will present the competitive examinations to become a researcher or engineer at inria and the purpose of the CE. In short, the CE is:

  • in charge of organizing Inria’s teams evaluation
  • in charge of running all the national recruiting and promotion juries.
  • participates in all local recruiting campaigns.

Then, Anthony Schoof, the Head of Transfer, Innovation and Partnerships will present the Inria Start-up Studio program and its benefits!

Finally, we have also invited several new engineers, researchers and members of the start-up studio to share their experiences with us:

  • Jenny Kartsaki: Research and Development Engineer at the NeuroMod Institute, Université Cote d’Azur in collaboration with the SED (Service d’Expérimentation et de Développement) Inria
  • Côme Lebreton: Engineer in ABS team
  • Romain Tetley: Engineer transfert and Innovation
  • Francesca Casagli: Permanent researcher in the BIOCORE team
  • Antonia Machlouzarides-Shalit: previous member of INRIA start-up studio and CEO of NeuroPin

2020_0330-Inria_Startup_Studio (2)

]]>