6.1.1 Cyanure

• Name:
  Cyanure: An Open-Source Toolbox for Empirical Risk Minimization
• Keyword:
  Machine learning
• Functional Description:
  Cyanure is an open-source C++ software package with a Python interface. The goal of Arsenic is to provide state-of-the-art solvers for learning linear models, based on stochastic variance-reduced stochastic optimization with acceleration mechanisms and Quasi-Newton principles. Arsenic can handle a large variety of loss functions (logistic, square, squared hinge, multinomial logistic) and regularization functions (l2, l1, elastic-net, fused Lasso, multi-task group Lasso). It provides a simple Python API, which is very close to that of scikit-learn, which should be extended to other languages such as R or Matlab in a near future.
• Release Contributions:
  version initiale
• URL:
  http://­thoth.­inrialpes.­fr/­people/­mairal/­arsenic/­welcome.­html
• Contact:
  Julien Mairal
• Participant:
  Julien Mairal

6.1.2 OTK

• Name:
  Optimal Transport Kernel Embedding
• Keyword:
  Machine learning
• Functional Description:

  Source code reproducing the results of the following paper

  Grégoire Mialon*, Dexiong Chen*, Alexandre d'Aspremont, Julien Mairal. A Trainable Optimal Transport Embedding for Feature Aggregation and its Relationship to Attention. ICLR 2021.

• URL:
  https://­github.­com/­claying/­OTK/
• Contact:
  Julien Mairal

6.1.3 T3SC

• Name:
  A Trainable Spectral-Spatial Sparse Coding Model for Hyperspectral Image Restoration
• Keyword:
  Machine learning
• Functional Description:

  This is the source code for reproducing the results of the paper

  A Trainable Spectral-Spatial Sparse Coding Model for Hyperspectral Image Restoration (Neurips 2021)

• URL:
  https://­github.­com/­inria-thoth/­T3SC/
• Contact:
  Théo Bodrito

6.1.4 GraphIT

• Name:
  GraphiT: Encoding Graph Structure in Transformers
• Keyword:
  Machine learning
• Functional Description:

  This is the source code for reproducing the results of the paper

  Grégoire Mialon*, Dexiong Chen*, Margot Selosse*, Julien Mairal. GraphiT: Encoding Graph Structure in Transformers.

• URL:
  https://­github.­com/­inria-thoth/­GraphiT/
• Contact:
  Grégoire Mialon

6.1.5 cog-eval

• Name:
  The ImageNet-CoG Benchmark
• Keywords:
  Computer vision, Evaluation
• Functional Description:
  Code repository for the ImageNet-CoG Benchmark introduced in the paper "Concept Generalization in Visual Representation Learning" (ICCV 2021). It contains code for reproducing all the experiments reported in the paper, as well as instructions on how to evaluate any custom model on the ImageNet-CoG Benchmark.
• URL:
  https://­github.­com/­naver/­cog
• Contact:
  Karteek Alahari

6.1.6 Multi-modal transformer

• Name:
  Multi-modal Transformer for Video Retrieval
• Keywords:
  Computer vision, Video retrieval, Transformer
• Functional Description:
  The task of retrieving video content relevant to natural language queries plays a critical role in effectively handling internet-scale datasets. Most of the existing methods for this caption-to-video retrieval problem do not fully exploit cross-modal cues present in video. Furthermore, they aggregate per-frame visual features with limited or no temporal information. In this paper, we present a multi-modal transformer to jointly encode the different modalities in video, which allows each of them to attend to the others. The transformer architecture is also leveraged to encode and model the temporal information. On the natural language side, we investigate the best practices to jointly optimize the language embedding together with the multi-modal transformer. This is an implementation of our novel framework, which was presented at ECCV 2020.
• URL:
  http://­thoth.­inrialpes.­fr/­research/­mmt/
• Contact:
  Karteek Alahari
• Partner:
  Google

Emerging Properties in Self-Supervised Vision Transformers

Participants: Mathilde Caron, Hugo Touvron, Ishan Mishra, Hervé Jégou, Piotr Bojanowski, Julien Mairal.

In 6, we question if self-supervised learning provides new properties to Vision Transformer (ViT) that stand out compared to convolutional networks (convnets). Beyond the fact that adapting self-supervised methods to this architecture works particularly well, we make the following observations: first, self-supervised ViT features contain explicit information about the semantic segmentation of an image, which does not emerge as clearly with supervised ViTs, nor with convnets. Second, these features are also excellent k-NN classifiers, reaching 78.3% top-1 on ImageNet with a small ViT. Our study also underlines the importance of momentum encoder, multi-crop training, and the use of small patches with ViTs. We implement our findings into a simple self-supervised method, called DINO, which we interpret as a form of self-distillation with no labels. We show the synergy between DINO and ViTs by achieving 80.1% top-1 on ImageNet in linear evaluation with ViT-Base.

DINO's results.

Concept generalization in visual representation learning

Participants: Mert Bulent Sariyildiz, Yannis Kalantidis, Diane Larlus, Karteek Alahari.

In this work 20, we study concept generalization, i.e., the extent to which models trained on a set of (seen) visual concepts can be used to recognize a new set of (unseen) concepts. Although this paradigm is a popular way of evaluating visual representations, the choice of which unseen concepts to use is usually made arbitrarily, and independently from the seen concepts used to train representations, thus ignoring any semantic relationships between the two. We argue that semantic relationships between seen and unseen concepts affect generalization performance and propose ImageNet-CoG (see Figure 2), a novel benchmark on the ImageNet dataset that enables measuring concept generalization in a principled way. Our benchmark leverages expert knowledge that comes from WordNet in order to define a sequence of unseen ImageNet concept sets that are semantically more and more distant from the ImageNet-1K subset, a ubiquitous training set. Under the prism of concept generalization, we analyse a variety of supervised, semi-supervised and self-supervised models with different backbone architectures, such as Convolutional Neural Networks, Vision Transformers or architectures found by Neural Architecture Search models. Our large-scale study shows that our benchmark is able to uncover a number of interesting insights.

Overview of our ImageNet Concept Generalization Benchmark

Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond

Participants: Đ. Khuê Lê-Huu, Karteek Alahari.

In  11, we introduce regularized Frank-Wolfe, a general and effective algorithm for inference and learning of dense conditional random fields (CRFs). The algorithm optimizes a nonconvex continuous relaxation of the CRF inference problem using vanilla Frank-Wolfe with approximate updates, which are equivalent to minimizing a regularized energy function. Our proposed method is a generalization of existing algorithms such as mean field or concave-convex procedure. This perspective not only offers a unified analysis of these algorithms, but also allows an easy way of exploring different variants that potentially yield better performance. We illustrate this in our empirical results on standard semantic segmentation datasets, where several instantiations of our regularized Frank-Wolfe outperform mean field inference, both as a standalone component and as an end-to-end trainable layer in a neural network (Figure 3). We also show that dense CRFs, coupled with our new algorithms, produce significant improvements over strong CNN baselines.

CRF as an end-to-end trainable component in a neural network

Modélisation Parcimonieuse de CNNs avec des Paquets d'Ondelettes Dual-Tree

Participants: Hubert Leterme, Kévin Polisano, Valérie Perrier, Karteek Alahari.

We propose in 13 to improve the mathematical interpretability of convolutional neural networks (CNNs) for image classification. In this purpose, we replace the first layers of existing models such as AlexNet or ResNet by an operator containing the dual-tree wavelet packet transform, i.e., a redundant decomposition using complex and oriented waveforms. Figure 4 provides a schematic representation of one such architecture. Our experiments show that these modified networks behave very similarly to the original models once trained. The goal is then to study this operator from a theoretical point of view and to identify potential optimizations. We want to analyze its main properties such as directional selectivity, stability with respect to small shifts and rotations, thus retaining discriminant information while decreasing intra-class variability. This work is a step toward a more complete description of CNNs using well-defined mathematical operators, characterized by a small number of arbitrary parameters, making them easier to interpret.

Our two-layer model.

LiDARTouch: Monocular metric depth estimation with a few-beam LiDAR

Participants: Florent Bartoccioni, Eloi Zablocki, Patrick Pérez, Matthieu Cord, Karteek Alahari.

In this paper 23, we address the task of monocular depth prediction, a key component of many autonomous systems, by self-supervised deep learning. Existing methods are either fully-supervised with an additional expensive LiDAR (32 or 64 beams) as input or self-supervised with camera-only methods, much cheaper, but suffering from scale ambiguity and infinite depth problems. In contrast, we introduce LiDARTouch, a novel method combining a monocular camera with a cheap minimal 4-beam LiDAR input, typical of laser scanners currently used in the automotive industry. We introduce a new self-supervision scheme to leverage this very sparse LiDAR input at three complementary levels. While being extremely sparse, we show that the use of a few-beam LiDAR alleviate the scaling ambiguity and infinite depth problems that camera-only methods suffer from. We also reach competitive performances with respect to fully-supervised depth completion methods while being significantly cheaper and more annotation friendly. Our method can be trained on any domain with no modification, and it can thus bring accurate and metric depth estimation at a vehicle fleet scale. In Figure 5, we present three examples along with selected close-ups highlighting the infinite-depth problem. For example, on the leftmost column, we observe a typical ‘hole’ in the depth map where previous `Depth Estimation` method estimates a vehicle three times as far as its true distance. In contrast, by leveraging small touches of LIDAR we disambiguate the prediction and can accurately and safely handle moving objects with no relative motion, typical of cars in fluid traffic.

Masking Modalities for Cross-modal Video Retrieval

Participants: Valentin Gabeur, Arsha Nagrani, Chen Sun, Kateek Alahari, Cordelia Schmid.

Pre-training on large scale unlabelled datasets has shown impressive performance improvements in the fields of computer vision and natural language processing. Given the advent of large-scale instructional video datasets, a common strategy for pre-training video encoders is to use the accompanying speech as weak supervision. However, as speech is used to supervise the pre-training, it is never seen by the video encoder, which does not learn to process that modality. We address this drawback of current pre-training methods, which fail to exploit the rich cues in spoken language, in 21. Our proposal is to pre-train a video encoder using all the available video modalities as supervision, namely, appearance, sound, and transcribed speech. We mask an entire modality in the input and predict it using the other two modalities. This encourages each modality to collaborate with the others, and our video encoder learns to process appearance and audio as well as speech 6. We show the superior performance of our "modality masking" pre-training approach for video retrieval on the How2R, YouCook2 and Condensed Movies datasets.

Illustration of our approach

A Contextual Bandit Bake-off

Participants: Alberto Bietti [Microsoft Research], Alekh Agarwal [Microsoft Research], John Langford [Microsoft Research].

Beyond Tikhonov: Faster Learning with Self-Concordant Losses via Iterative Regularization

Participants: Gaspard Beugnot, Julien Mairal, Alessandro Rudi.

KALE Flow: A Relaxed KL Gradient Flow for Probabilities with Disjoint Support

Participants: Pierre Glaser, Michael Arbel, Arthur Gretton.

KALE's results.

Amortized implicit differentiation for stochastic bilevel optimization

Participants: Michael Arbel, Julien Mairal.

AMIGO's results.

Counterfactual Learning of Stochastic Policies with Continuous Actions

Participants: Houssam Zenati, Alberto Bietti, Matthieu Martin, Eustache Diemert, Julien Mairal.

Counterfactual reasoning from logged data has become increasingly important for many applications such as web advertising or healthcare. In this paper 26, we address the problem of counterfactual learning of stochastic policies with continuous actions, which raises difficult challenges about (i) data modelization, (ii) optimization, and (iii) evaluation on real data.

First, we introduce a modeling strategy based on a joint kernel embedding of contexts and actions, which overcomes the shortcomings of previous discretization strategies as shown in 9. Second, we empirically show that the optimization aspect of counterfactual learning is more important than previously thought, and we demonstrate the benefits of proximal point algorithms and differentiable estimators. Finally, we propose an evaluation protocol for offline policies in real-world logged systems, which is challenging since policies cannot be replayed on test data, and we release a new large-scale dataset along with multiple synthetic, yet realistic, evaluation setups.

pooling mechanism

A Continuized View on Nesterov Acceleration for Stochastic Gradient Descent and Randomized Gossip

Participants: Mathieu Even, Raphaël Berthier, Francis Bach, Nicolas Flammarion, Pierre Gaillard, Hadrien Hendrikx, Laurent Massoulié, Adrien Taylor.

In 7, we introduce the continuized Nesterov acceleration, a close variant of Nesterov acceleration whose variables are indexed by a continuous time parameter. The two variables continuously mix following a linear ordinary differential equation and take gradient steps at random times. This continuized variant benefits from the best of the continuous and the discrete frameworks: as a continuous process, one can use differential calculus to analyze convergence and obtain analytical expressions for the parameters; and a discretization of the continuized process can be computed exactly with convergence rates similar to those of Nesterov original acceleration. We show that the discretization has the same structure as Nesterov acceleration, but with random parameters. We provide continuized Nesterov acceleration under deterministic as well as stochastic gradients, with either additive or multiplicative noise. Finally, using our continuized framework and expressing the gossip averaging problem as the stochastic minimization of a certain energy function, we provide the first rigorous acceleration of asynchronous gossip algorithms.

Convergence of the

Mixability made efficient: Fast online multiclass logistic regression

Participants: Rémi Jézéquel, Pierre Gaillard, Alessandro Rudi.

Mixability has been shown to be a powerful tool to obtain algorithms with optimal regret. However, the resulting methods often suffer from high computational complexity which has reduced their practical applicability. For example, in the case of multiclass logistic regression, the aggregating forecaster (Foster et al., 2018) achieves a regret of $O(log(Bn\left)\right)$ whereas Online Newton Step achieves $O(e^{B}log\left(n\right))$ obtaining a double exponential gain in $B$ (a bound on the norm of comparative functions). However, this high statistical performance is at the price of a prohibitive computational complexity $O\left(n^{37}\right)$.

In 10, we use quadratic surrogates to make aggregating forecasters more efficient. We show that the resulting algorithm has still high statistical performance for a large class of losses. In particular, we derive an algorithm for multi-class logistic regression with a regret bounded by $O(Blog(n\left)\right)$ and a computational complexity of only $O\left(n^4\right)$.

pooling mechanism

Online Sign Identification: Minimization of the Number of Errors in Thresholding Bandits

Participants: Réda Ouhamma, Rémy Degenne, Pierre Gaillard, Vianney Perchet.

In the fixed budget thresholding bandit problem, an algorithm sequentially allocates a budgeted number of samples to different distributions. It then predicts whether the mean of each distribution is larger or lower than a given threshold. In 17, we introduce a large family of algorithms (containing most existing relevant ones), inspired by the Frank-Wolfe algorithm, and provide a thorough yet generic analysis of their performance. This allowed us to construct new explicit algorithms, for a broad class of problems, whose losses are within a small constant factor of the non-adaptive oracle ones. Quite interestingly, we observed that adaptive methods empirically greatly out-perform non-adaptive oracles, an uncommon behavior in standard online learning settings, such as regret minimization. We explain this surprising phenomenon on an insightful toy problem.

Sampling distributions

Dueling Bandits with Adversarial Sleeping

Participants: Aadirupa Saha, Pierre Gaillard.

In 19, we introduce the problem of sleeping dueling bandits with stochastic preferences and adversarial availabilities (DB-SPAA). In almost all dueling bandit applications, the decision space often changes over time; e.g., retail store management, online shopping, restaurant recommendation, search engine optimization, etc. Surprisingly, this `sleeping aspect' of dueling bandits has never been studied in the literature. Like dueling bandits, the goal is to compete with the best arm by sequentially querying the preference feedback of item pairs. The non-triviality however results due to the non-stationary item spaces that allow any arbitrary subsets items to go unavailable every round. The goal is to find an optimal `no-regret' policy that can identify the best available item at each round, as opposed to the standard `fixed best-arm regret objective' of dueling bandits. We first derive an instance-specific lower bound for DB-SPAA $\Omega \left({\sum }_{i=1}^{K-1}{\sum }_{j=i+1}^K\frac{logT}{\Delta (i,j)}\right)$, where $K$ is the number of items and $\Delta (i,j)$ is the gap between items $i$ and $j$. This indicates that the sleeping problem with preference feedback is inherently more difficult than that for classical multi-armed bandits (MAB). We then propose two algorithms, with near optimal regret guarantees. Our results are corroborated empirically.

Regret

A Trainable Optimal Transport Embedding for Feature Aggregation and its Relationship to Attention

Participants: Grégoire Mialon, Dexiong Chen, Alexandre d'Aspremont, Julien Mairal.

pooling mechanism

GraphiT: Encoding Graph Structure in Transformers.

Participants: Grégoire Mialon, Dexiong Chen, Alexandre d'Aspremont, Julien Mairal.

Lucas-Kanade Reloaded: End-to-End Super-Resolution from Raw Image Bursts

.

Participants: Bruno Lecouat, Jean Ponce, Julien Mairal.

super resolution results.

A Trainable Spectral-Spatial Sparse Coding Model for Hyperspectral Image Restoration

Participants: Théo Bodrito, Alexandre Zouaoui, Jocelyn Chanussot, Julien Mairal.

T3SC architecture

Tactical Optimism and Pessimism for Deep Reinforcement Learning

Participants: Ted Moskovitz, Jack Parker-Holder, Aldo Pacchiano, Michael Arbel, Michael I. Jordan.

TOP's results.

Towards an Understanding of Default Policies in Multitask Policy Optimization

Participants: Ted Moskovitz, Michael Arbel, Jack Parker-Holder.

DefaultPolicy's results.

Memory-Augmented Reinforcement Learning for Image-Goal Navigation

Participants: Lina Mezghani, Sainbayar Sukhbaatar, Thibaut Lavril, Oleksandr Maksymets, Dhruv Batra, Piotr Bojanowski, Karteek Alahari.

In this work 24, we address the problem of image-goal navigation in the context of visually-realistic 3D environments. This task involves navigating to a location indicated by a target image in a previously unseen environment. Earlier attempts, including RL-based and SLAM-based approaches, have either shown poor generalization performance, or are heavily reliant on pose/depth sensors. We present a novel method, shown in Figure 19, that leverages a cross-episode memory to learn to navigate. We first train a state-embedding network in a self-supervised fashion, and then use it to embed previously-visited states into an agent’s memory. In order to avoid overfitting, we propose to use data augmentation on the RGB input during training. We validate our approach through extensive evaluations, showing that our data-augmented memory-based model establishes a new state of the art on the image-goal navigation task in the challenging Gibson dataset. We obtain this competitive performance from RGB input only, without access to additional sensors such as position or depth.

Illustration of our approach

Episodic Transformer for Vision-and-Language Navigation

Participants: Alexander Pashevich, Cordelia Schmid, Chen Sun.

Residual Reinforcement Learning from Demonstrations

Participants: Minttu Alakuijala, Gabriel Dulac-Arnold, Julien Mairal, Jean Ponce, Cordelia Schmid.

Extracting Universal Representations of Cognition across Brain-Imaging Studies

Participants: Arthur Mensch, Julien Mairal, Bertrand Thirion, Gael Varoquaux.

We show in 2 how to extract shared brain representations that predict mental processes across many cognitive neuroimaging studies. Focused cognitive-neuroimaging experiments study precise mental processes with carefully-designed cognitive paradigms; however the cost of imaging limits their statistical power. On the other hand, large-scale databasing efforts increase considerably the sample sizes, but cannot ask precise cognitive questions. To address this tension, we develop new methods that turn the heterogeneous cognitive information held in different task-fMRI studies into common-universal-cognitive models. Our approach does not assume any prior knowledge of the commonalities shared by the studies in the corpus; those are inferred during model training. The method uses deep-learning techniques to extract representations - task-optimized networks - that form a set of basis cognitive dimensions relevant to the psychological manipulations, as illustrated in Figure 25. In this sense, it forms a novel kind of functional atlas, optimized to capture mental state across many functional-imaging experiments. As it bridges information on the neural support of mental processes, this representation improves decoding performance for 80% of the 35 widely-different functional imaging studies that we consider. Our approach opens new ways of extracting information from brain maps, increasing statistical power even for focused cognitive neuroimaging studies, in particular for those with few subjects.

Visualization of task-optimized networks

9.1.1 Associate Teams in the framework of an Inria International Lab or in the framework of an Inria International Program

9.2.1 Visits of international scientists

9.3.1 ERC Starting Grant SOLARIS

Participants: Julien Mairal.

The main paradigm used in the project is that of kernel methods and consist of building functional spaces where deep learning models live. By doing so, we want to derive theoretical properties of deep learning models that may explain their success, and also obtain new tools with better stability properties. Another work package of the project is focused on large-scale optimization, which is a key to obtain fast learning algorithms.

9.4.1 ANR Project AVENUE

Participants: Karteek Alahari.

9.5.1 3IA MIAI chair: Towards More Data Efficiency in Machine Learning

Participants: Julien Mairal, Karteek Alahari, Massih-Reza Amini, Margot Selosse, Juliette Marrie, Romain Ménégaux.

10.1.1 Scientific events: organisation

10.1.2 Scientific events: selection

10.1.3 Journal

10.1.4 Invited talks

• M. Arbel: invited seminar at Instituto Superior Técnico (online).
• M. Arbel: invited talk at "Les journées MAS, Institut Denis Poisson"
• M. Arbel: invited seminar at Flatiron Institute/NYU (online).
• M. Arbel: invited seminar at Ecole Normale Supérieure, Lyon.
• M. Arbel: invited seminar at Laboratoire Jean Kuntzmann, Grenoble.
• H. Leterme: CollaboTICS - International Collaborative Workshop of RUB-UGA-UT, 3rd Edition (online).
• J. Mairal: invited talk at the French-German Machine Learning Symposium, Munich (online).
• J. Mairal: invited seminar at Flatiron Institute/NYU (online).
• J. Mairal: invited seminar at KU Leuven (online).
• J. Mairal: invited talk at the Prairie workshop. Paris.
• J. Mairal: keynote speaker at the Mascot NUM conference. (online).
• P. Gaillard: invited seminar at LJK. Grenoble.
• P. Gaillard: invited seminar at the Aptikal research seminar. Grenoble.

10.1.5 Scientific expertise

• K. Alahari: reviewer for ANR.
• J. Mairal: reviewer for the Swiss Data Science Center.

10.2.1 Teaching

• Doctorat: K. Alahari, Lecturer at the Machine Learning Summer School (MLSS), 3h eqTD, Taipei, Taiwan (online).
• Doctorat: K. Alahari, Lecturer at the CVIT summer school on machine learning, 3h eqTD, IIIT Hyderabad, India (online).
• Master: K. Alahari, Machine Learning for Multimodal Data, 11.25h eqTD, M2, UGA, Grenoble.
• Master: K. Alahari, Understanding Big Visual Data, 13.5h eqTD, M2, Grenoble INP.
• Master: K. Alahari, Graphical Models Inference and Learning, 18h eqTD, M2, CentraleSupelec, Paris.
• Master: K. Alahari, Introduction to computer vision, 9h eqTD, M1, ENS Paris.
• Master: H. Leterme, Analyse pour l'ingénieur, 31.5h eqTD, Grenoble INP Ensimag.
• Master: H. Leterme, Traitement d'images, 12h eqTD, Grenoble INP Ensimag.
• Master: J. Mairal, Kernel methods for statistical learning, 17h eqTD, M2, Ecole Normale Supérieure, Cachan, France.
• Master: J. Mairal, Kernel methods for statistical learning, 27h eqTD, M2, UGA, Grenoble.
• Master: P. Gaillard, Sequential Learning, 27h eqTD, M2, Ecole Normale Supérieure, Cachan, France.

10.2.2 Supervision (PhD defenses)

• PhD: Vladyslav Sydorov: Techniques spatiales pour la compréhension vidéo. 10/05/2021. Dir: Karteek Alahari, Cordelia Schmid.
• PhD: Mathilde Caron: Apprentissage auto-supervisé de représentations visuelles avec des réseaux de neurones profonds. 09/12/2021. Dir: Julien Mairal.
• PhD: Alexander Pashevich: Des robots qui voient : Apprentissage de comportements guidés par la vision. Univ. Grenoble Alpes. 29/09/2021. Dir: Cordelia Schmid.
• PhD: Raphaël Berthier: Analysis and Acceleration of Gradient Descents and Gossip Algorithms. 25/09/2021. Dir: Pierre Gaillard.

10.2.3 Juries

• K. Alahari: Reviewer for the PhD thesis of Enzo Battistella, Université Paris-Saclay.
• K. Alahari: Jury member of the thesis defense committee of Allison Del Giorno, Carnegie Mellon University.
• J. Mairal: Reviewer for the PhD thesis of Tran Khanh Hung, Université Paris-Saclay.
• J. Mairal: Reviewer for the PhD thesis of Hamza Cherkaoui, Université Paris-Saclay.
• J. Mairal: Reviewer for the PhD thesis of Thomas Eboli, PSL-Sorbonne Université.
• J. Mairal: Jury member for the HdR of Chaohui Wang, Université Paris-Est.
• J. Mairal: Jury member for the HdR of Franck Iutzeler, Université Grenoble-Alpes.
• J. Mairal: CSI member for the PhD of Tayeb Zarrouk. Univ. Grenoble Alpes.

10.3.1 Interventions

• J. Mairal was interviewed for a podcast from IHP called “l'Oreille Mathématique”.

International journals

• 1 articleA.Alberto Bietti, A.Alekh Agarwal and J.John Langford. A Contextual Bandit Bake-off.Journal of Machine Learning Research221332021, 1-49
  HALback to text
• 2 articleA.Arthur Mensch, J.Julien Mairal, B.Bertrand Thirion and G.Gaël Varoquaux. Extracting representations of cognition across neuroimaging studies improves brain decoding.PLoS Computational Biology175May 2021, e1008795:1-20
  HALDOIback to text

International peer-reviewed conferences

• 3 inproceedingsM.Michael Arbel and J.Julien Mairal. Amortized implicit differentiation for stochastic bilevel optimization.The Tenth International Conference on Learning RepresentationsOnline, France2022
  HALback to text
• 4 inproceedingsG.Gaspard Beugnot, J.Julien Mairal and A.Alessandro Rudi. Beyond Tikhonov: Faster Learning with Self-Concordant Losses via Iterative Regularization.NeurIPS 2021 – 35th Annual Conference on Neural Information Processing SystemsAdvances in Neural Information Processing Systems 34Virtual, FranceDecember 2021, 1-37
  HALback to text
• 5 inproceedingsT.Théo Bodrito, A.Alexandre Zouaoui, J.Jocelyn Chanussot and J.Julien Mairal. A Trainable Spectral-Spatial Sparse Coding Model for Hyperspectral Image Restoration.NeurIPS 2021 – 35th Annual Conference on Neural Information Processing SystemsSydney, AustraliaDecember 2021, 1-19
  HALback to text
• 6 inproceedingsM.Mathilde Caron, H.Hugo Touvron, I.Ishan Misra, H.Hervé Jegou, J.Julien Mairal, P.Piotr Bojanowski and A.Armand Joulin. Emerging Properties in Self-Supervised Vision Transformers.ICCV 2021 - International Conference on Computer VisionVirtual, FranceOctober 2021, 1-21
  HALback to text
• 7 inproceedingsM.Mathieu Even, R.Raphaël Berthier, F.Francis Bach, N.Nicolas Flammarion, P.Pierre Gaillard, H.Hadrien Hendrikx, L.Laurent Massoulié and A.Adrien Taylor. A Continuized View on Nesterov Acceleration for Stochastic Gradient Descent and Randomized Gossip.Advances in Neural Information Processing Systems 34NeurIPS 2021 - 35th Conference on Neural Information Processing SystemsSydney (virtual), AustraliaMorgan Kaufmann PublishersDecember 2021, 1-32
  HALback to text
• 8 inproceedingsC.Camila Fernández, C.Chung Shue Chen, P.Pierre Gaillard and A.Alonso Silva. Experimental Comparison of Semi-parametric, Parametric, and Machine Learning Methods for Time-to-Event Analysis Through the IPEC Score.SFdS 2020 - 52èmes Journées de Statistiques de la Société Française de StatistiqueNice, FranceJune 2021, 1-6
  HAL
• 9 inproceedingsP.Pierre Glaser, M.Michael Arbel and A.Arthur Gretton. KALE Flow: A Relaxed KL Gradient Flow for Probabilities with Disjoint Support.NeurIPS 2021 -Thirty-Fifth Annual Conference on Neural Information Processing SystemsAdvances in Neural Information Processing SystemsOnline, France2021, 1-29
  HALback to text
• 10 inproceedingsR.Rémi Jézéquel, P.Pierre Gaillard and A.Alessandro Rudi. Mixability made efficient: Fast online multiclass logistic regression.NeurIPS 2021. Thirty-fifth Conference on Neural Information Processing SystemsOnline, FranceDecember 2021
  HALback to text
• 11 inproceedingsÐ.Ð.Khuê Lê-Huu and K.Karteek Alahari. Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond.NeurIPS 2021 - 35th Annual Conference on Neural Information Processing SystemsVirtual-only Conference, AustraliaDecember 2021, 1-35
  HALback to text
• 12 inproceedingsB.Bruno Lecouat, J.Jean Ponce and J.Julien Mairal. Lucas-Kanade Reloaded: End-to-End Super-Resolution from Raw Image Bursts.ICCV 2021 - International Conference on Computer VisionVirtual, France2021, 1-16
  HALback to text
• 13 inproceedingsH.Hubert Leterme, K.Kévin Polisano, V.Valérie Perrier and K.Karteek Alahari. Modélisation Parcimonieuse de CNNs avec des Paquets d'Ondelettes Dual-Tree.ORASIS 2021 - Journées francophones des jeunes chercheurs en vision par ordinateurSaint Ferréol, FranceSeptember 2021, 1-9
  HALback to text
• 14 inproceedingsG.Grégoire Mialon, D.Dexiong Chen, A.Alexandre D'Aspremont and J.Julien Mairal. A Trainable Optimal Transport Embedding for Feature Aggregation and its Relationship to Attention.ICLR 2021 - The Ninth International Conference on Learning RepresentationsVirtual, FranceMay 2021
  HALback to text
• 15 inproceedingsT.Ted Moskovitz, M.Michael Arbel, J.Jack Parker-Holder and A.Aldo Pacchiano. Towards an Understanding of Default Policies in Multitask Policy Optimization.25th International Conference on Artificial Intelligence and StatisticsVolume 130: International Conference on Artificial Intelligence and StatisticsOnline, France2022
  HALback to text
• 16 inproceedingsT.Ted Moskovitz, J.Jack Parker-Holder, A.Aldo Pacchiano, M.Michael Arbel and M. I.Michael I Jordan. Tactical Optimism and Pessimism for Deep Reinforcement Learning.NeurIPS 2021 - Thirty-fifth Annual Conference on Neural Information Processing SystemsAdvances in Neural Information Processing SystemsOnline, France2021, 1-15
  HALback to text
• 17 inproceedingsR.Reda Ouhamma, R.Rémy Degenne, P.Pierre Gaillard and V.Vianney Perchet. Online Sign Identification: Minimization of the Number of Errors in Thresholding Bandits.NeurIPS 2021 - 35th International Conference on Neural Information Processing SystemsNeurIPS 2021 - 35th International Conference on Neural Information Processing SystemsVirtual, Canada2021, 1-25
  HALback to text
• 18 inproceedingsA.Alexander Pashevich, C.Cordelia Schmid and C.Chen Sun. Episodic Transformer for Vision-and-Language Navigation.ICCV 2021 - International Conference on Computer VisionVirtual, United StatesOctober 2021, 1-18
  HALback to text
• 19 inproceedingsA.Aadirupa Saha and P.Pierre Gaillard. Dueling Bandits with Adversarial Sleeping.NeurIPS 2021 - 35th International Conference on Neural Information Processing SystemsVirtual, CanadaDecember 2021, 1-25
  HALback to text
• 20 inproceedingsM. B.Mert Bulent Sariyildiz, Y.Yannis Kalantidis, D.Diane Larlus and K.Karteek Alahari. Concept Generalization in Visual Representation Learning.ICCV 2021 - International Conference on Computer VisionVirtual, CanadaOctober 2021
  HALback to text

Conferences without proceedings

• 21 inproceedingsV.Valentin Gabeur, A.Arsha Nagrani, C.Chen Sun, K.Karteek Alahari and C.Cordelia Schmid. Masking Modalities for Cross-modal Video Retrieval.WACV 2022 - Winter Conference on Applications of Computer VisionWaikoloa, United StatesJanuary 2022, 1-10
  HALback to text

Reports & preprints

• 22 miscM.Minttu Alakuijala, G.Gabriel Dulac-Arnold, J.Julien Mairal, J.Jean Ponce and C.Cordelia Schmid. Residual Reinforcement Learning from Demonstrations.June 2021
  HALback to text
• 23 miscF.Florent Bartoccioni, É.Éloi Zablocki, P.Patrick Pérez, M.Matthieu Cord and K.Karteek Alahari. LiDARTouch: Monocular metric depth estimation with a few-beam LiDAR.September 2021
  HALback to text
• 24 miscL.Lina Mezghani, S.Sainbayar Sukhbaatar, T.Thibaut Lavril, O.Oleksandr Maksymets, D.Dhruv Batra, P.Piotr Bojanowski and K.Karteek Alahari. Memory-Augmented Reinforcement Learning for Image-Goal Navigation.January 2021
  HALback to text
• 25 miscG.Grégoire Mialon, D.Dexiong Chen, M.Margot Selosse and J.Julien Mairal. GraphiT: Encoding Graph Structure in Transformers.June 2021
  HALback to text
• 26 miscH.Houssam Zenati, A.Alberto Bietti, M.Matthieu Martin, E.Eustache Diemert and J.Julien Mairal. Counterfactual Learning of Stochastic Policies with Continuous Actions: from Models to Offline Evaluation.August 2021
  HALback to text