Vincent Schellekens

81 posts

Vincent Schellekens banner
Vincent Schellekens

Vincent Schellekens

@schellekensv

Postdoc: signal processing, machine learning, quantization and stuff.

Katılım Mart 2018
98 Takip Edilen67 Takipçiler
Vincent Schellekens retweetledi
Michael P. Sheehan
Michael P. Sheehan@mpsheehan1995·
Delighted to say our paper "A Sketching Framework for Reduced Data Transfer in Photon Counting Lidar" has been accepted for publication in the IEEE Trans. on Computational Imaging. Thanks to all the reviewers for their suggestions. See the arXiv version - arxiv.org/abs/2102.08732
English
1
2
6
0
Vincent Schellekens retweetledi
Martin Genzel
Martin Genzel@MartinGenzel·
"Under what conditions can a random neural network make two class sets with positive distance linearly separable?" In a recent joint work with Sjoerd Dirksen, @jacquesdurden & Alex Stollenwerk, we take a geometric approach to this problem. 1/5
Martin Genzel tweet media
English
4
37
130
0
Chat Sceptique
Chat Sceptique@ChatSceptique·
...et c'est finalement Benjamin Chiêm de l'@UCLouvain_be que nous avons choisi 🥳 Il a aussi obtenu le prix des professionnels de la vulgarisation scientifique, nous laissant penser qu'on a bien choisi 😅
Chat Sceptique tweet media
Français
2
0
4
0
Chat Sceptique
Chat Sceptique@ChatSceptique·
Hier n'a pas été une journée comme les autres pour moi 😅 Qui parmi ces 15 chercheurs belges envoyer à Paris pour le concours international de *Ma thèse en 180 secondes* ? Dans le jury du concours national à @UniversiteLiege, un certain Chat Sceptique 😇 Je vous raconte ça ?
Chat Sceptique tweet media
Français
2
3
43
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
... and finally, the last tweet about our "Breaking the waves" paper: it is now published in "Information and Inference" 🙂you can find it here 👉 shorturl.at/bpuS1 (the ArXiv version is prettier though)
Vincent Schellekens tweet media
Vincent Schellekens@schellekensv

I'm very happy to announce a big update on "Breaking the waves" with @jacquesdurden! The key messages are now much clearer, following the nice suggestions from our top-quality reviewers (seriously, huge shoutout to them for their very constructive work) 😀

English
0
2
12
0
Vincent Schellekens retweetledi
Low-rank Jack (go mathstodon)
Low-rank Jack (go mathstodon)@jacquesdurden·
Preprint time: very happy of this work of Stephanie Guerit (who’s doing her viva in 2w) made in collaboration with @finding_sid, J. Lee, and Hervé Rigneault, on “Compressive lensless endoscopy with partial speckle scanning,“ with follow-up to come soon arxiv.org/abs/2104.10959
English
0
5
14
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
Numerical experiments also back up our theoretical results. As a proof-of-concept, we tackled feature extraction for audio event classification: as can be seen here, when we go from full-precision (blue) to quantized (green), we reach the same accuracy with reduced resources 😀
Vincent Schellekens tweet media
English
0
0
2
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
We already studied this specific case before (arxiv.org/abs/1804.10109), but the main novelty of this work is that we provide sound theoretical guarantees which explain exactly *why* (and when) this "asymmetric compressive learning" idea works 🧑‍🏫
English
1
0
1
0
Vincent Schellekens retweetledi
Michael P. Sheehan
Michael P. Sheehan@mpsheehan1995·
New preprint out: "A Sketching Framework for Single Photon Counting Lidar" with Julián Tachella and Mike Davies - arxiv.org/abs/2102.08732. We propose a sketching method to compress the time-of-flight data without any significant loss of information. 1/8
English
1
6
12
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
@vlanx0 Par contre, si ces recommendations "vidéos similaires" sont symétriques (si A recommande B alors B recommande A), c'est peut-être pour le mieux 🤔
Français
1
0
4
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
Among others, we added a new experiment highlighting the practical gain of the "semi-quantized" random Fourier features scheme studied in the paper. Main takeaway: with this scheme (green), we can heavily reduce the bitrate of an RFF-based classifier with little loss of accuracy.
Vincent Schellekens tweet media
English
0
1
6
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
I'm very happy to announce a big update on "Breaking the waves" with @jacquesdurden! The key messages are now much clearer, following the nice suggestions from our top-quality reviewers (seriously, huge shoutout to them for their very constructive work) 😀
Vincent Schellekens@schellekensv

New preprint out: "Breaking the waves: asymmetric random periodic features for low-bitrate kernel machines", with @jacquesdurden ! What happens when you take features (such as random Fourier features) but with a different feature map for both signals? arxiv.org/abs/2004.06560

English
1
2
9
0
Vincent Schellekens retweetledi
ISPGroupUCL (go Mastodon)
ISPGroupUCL (go Mastodon)@ISPGroupUCL·
Dr Amirafshar Moshtaghpour receives the ICTEAM Best Thesis Award 2020 from the ICTEAM institute of UCLouvain for his thesis entitled "Computational Interferometry for Hyperspectral Imaging". Congratulations @amirafsharm!!
English
0
3
6
0
Vincent Schellekens
Vincent Schellekens@schellekensv·
@colobas_ @mariotelfig The paper title caught my attention when I was browsing through the program, I'm looking forward to discover it in "virtual Bruges" 😄
English
0
0
0
0

Keşfet

@mpsheehan1995 @ieeeICASSP @TachellaJulian @IEEEspm @RemiGribonval @jacquesdurden @ChatSceptique @UniversiteLiege