Digital Group

@conference{guerard_1835,

title = {Tourists Profiling by Interest Analysis},

author = {Guillaume Guérard and Sonia Djebali and Quentin Gabot},

url = {https://www.madics.fr/event/symposium-madics-4/},

year  = {2022},

date = {2022-07-01},

booktitle = {MaDICS 2022},

address = {Lyon, France},

abstract = {I'm a research student in MSc in Computer Science, my research study concern the behaviors of tourists using digital traces they leave during their travels on social medias. The studies conducted on diverse aspects of tourism focus on quantitative aspects of digital traces to reach its conclusions. In my works, I suggest a study focused on both qualitative and quantitative aspect of digital traces to understand the dynamics governing tourist behavior, especially those concerning attractions networks.},

keywords = {},

pubstate = {published},

tppubtype = {conference}

}

@conference{bobbia_1773,

title = {Donsker's Theorem for local Data-Driven Empirical Measure},

author = {Benjamin Bobbia},

url = {x},

year  = {2022},

date = {2022-02-01},

booktitle = {Séminaire du laboratoire Modal X},

address = {Nanterre, France},

abstract = {Nowadays, empirical processes are well known objects. A reason that push forward theirs studies is that, in many models, we can write the estimators as images of empirical measures. In this work we investigate the case of local empirical measures built over a sub-sample with data conditioned to be in a certain area. There exists numerous results about such question, but what can we say if the area of interest is data driven ? In the present work we present a general framework which allows to derive asymptotic results for this particular empirical measures (with "low" cost in terms of technicality and assumptions). Then, we use this approach in the framework of extreme values theory deriving tail and quantile estimations.},

keywords = {},

pubstate = {published},

tppubtype = {conference}

}

@conference{bobbia_1772,

title = {Donsker's Theorem for local data-driven empirical measures},

author = {Benjamin Bobbia},

url = {https://femmes-et-maths.fr/2021/06/28/forum-2021-des-jeunes-mathematien-nes/},

year  = {2021},

date = {2021-12-01},

booktitle = {Forum Jeunes Mathématiciens},

address = {Besançon, france},

abstract = {Nowadays, empirical processes are well known object. A reason that push forward theirs studies is that, in many modelisations, we can write the estimators as images of empirical measures. In this work we investigate the case of local empirical measures, that is the empirical measure built over a subsample with data conditionned to be in a certain area. There exist numerous results about such result, but what can we say if the previous area is data driven ? We can handle this situation with high technical cost and additional assumptions. The main aim of the present work is to present a general framework which allows to derive asymptotic results for this particular empirical measures with lower cost (in terms of technicality and assumptions). We give a Donsker-type theorem for such measures and present some domains of application as extreme values theory or multivariate regular variations},

keywords = {},

pubstate = {published},

tppubtype = {conference}

}

@conference{i-hsien_kuo_1764,

title = {Learning to Continually Learn Rapidly from Few and Noisy Data},

author = {Nicholas I-Hsien Kuo and Mehrtash Harandi and Nicolas Fourrier and Christian Walder and Gabriela Ferraro and Hanna Suominen},

url = {https://proceedings.mlr.press/v140/kuo21a.html},

year  = {2021},

date = {2021-04-01},

booktitle = {AAAI Workshop on Meta-Learning and MetaDL Challenge},

pages = {140:65-76},

address = {online},

abstract = {Neural networks suffer from catastrophic forgetting and are unable to sequentially learn new tasks without guaranteed stationarity in data distribution. Continual learning could be achieved via replay - by concurrently training externally stored old data while learning a new task. However, replay becomes less effective when each past task is allocated with less memory. To overcome this difficulty, we supplemented replay mechanics with meta-learning for rapid knowledge acquisition. By employing a meta-learner, which learns a learning rate per parameter per past task, we found that base learners produced strong results when less memory was available. Additionally, our approach inherited several meta-learning advantages for continual learning: it demonstrated strong robustness to continually learn under the presence of noises and yielded base learners to higher accuracy in less updates.},

keywords = {},

pubstate = {published},

tppubtype = {conference}

}

@conference{i-hsien_kuo_1765,

title = {M2SGD: Learning to Learn Important Weights},

author = {Nicholas I-Hsien Kuo and Mehrtash Harandi and Nicolas Fourrier and Christian Walder and Gabriela Ferraro and Hanna Suominen},

url = {https://ieeexplore.ieee.org/document/9150875},

year  = {2020},

date = {2020-06-01},

booktitle = {IEEE Conference on Computer Vision and Pattern Recognition Workshops},

address = {Seattle, USA},

abstract = {Meta-learning concerns rapid knowledge acquisition. One popular approach cast optimisation as a learning problem and it has been shown that learnt neural optimisers updated base learners more quickly than their handcrafted counterparts. In this paper, we learn an optimisation rule that sparsely updates the learner parameters and removes redundant weights. We present Masked Meta-SGD (M 2 SGD), a neural optimiser which is not only capable of updating learners quickly, but also capable of removing 83.71% weights for ResNet20s. We release our codes at https://github.com/Nic5472K/CLVISION2020_CVPR_M2SGD},

keywords = {},

pubstate = {published},

tppubtype = {conference}

}