Eric Arazo (@ArazoEric) header
Eric Arazo (@ArazoEric) profile picture
Joined August 2016
Photo by Eric Arazo (@ArazoEric)
Our paper: "Is your noise correction noisy? PLS: Robustness to label noise with two stage detection" was accepted at #WACV2023 ! We detect incorrect true label guesses using our pseudo-loss to combat label noise in image classification datasets arxiv.org/abs/2210.04578
3
1
14
Very glad to announce that our label noise paper was accepted #ECCV2022. We observe a linear separation between OOD and ID samples in web crawled datasets using unsupervised contrastive learning arxiv.org/abs/2207.01573. Team: @ArazoEric @kevinmcguinness @oconnorn @insight_centre
4
19
I presented my work on using drones to predict herbage quality at the #AgricultureVision workshop @CVPR. A research collab in the @VistaMilk SFI centre between @teagasc and @insight_centre funded by @scienceirel. Great to attend in person again ! Paper: shorturl.at/cqzUZGreat
2
4
27
Excited to present Thursday our #BMVC21 paper on training CNNs under budget constraints. Check it out in the oral session 8 or poster session 4! Great teamwork @Diego_OrtegoH @oconnorn @kevinmcguinness @insight_centre BMVC:bmvc2021-virtualconference.c… arXiV:arxiv.org/abs/2110.14283
1
4
11
Replying to @Diego_OrtegoH
@Diego_OrtegoH "Objective Interpolation Training for Robustness to Label Noise" accepted #CVPR2021 Synthetic & real-world noise in image classification @ArazoEric Paul Albert @oconnorn @kevinmcguinness @DCU: lnkd.in/eabGH_y Code: git.io/JI40X #DeepLearning
2
7
Insight researcher @Diego_OrtegoH will be presenting "Towards Robust Learning with Different Label Noise Distributions" Jan. 12th, T1.2 at @icpr2020milan @ArazoEric @oconnorn @kevinmcguinness @insight_centre #ICPR2020 #AI Paper: arxiv.org/abs/1912.08741 git.io/JJ0PV
5
17
Happy to announce that our paper "Towards Robust Learning with Different Label Noise Distributions" has been accepted as an oral presentation at @icpr2020milan. @ArazoEric @oconnorn @kevinmcguinness @insight_centre Paper: arxiv.org/abs/1912.08741 Code: git.io/JJ0PV
6
16
Interested in training DNNs with noisy labels? Check out our new work on Multi-Objective Interpolation Training to deal with synthetic and web noise. @ArazoEric Paul Albert @oconnorn @kevinmcguinness @insight_centre Paper: arxiv.org/abs/2012.04462 Code: git.io/JI40X
7
17
🔊New paper out! "Unsupervised Contrastive Learning of Sound Event Representations". Work done with @Diego_OrtegoH We learn audio representations by contrasting differently augmented views of sound events. paper: arxiv.org/abs/2011.07616 code (soon): github.com/edufonseca/uclser… (1/5)
2
13
2
69
We are happy to present our paper "Pseudo-Labeling and Confirmation Bias in Deep Semi-Supervised Learning" at #IJCNN202 at #WCCI2020 WCCI20. Come along (or e-mail us) if you want to know about some of the challenges of SSL (this evening from 8pm). Code: git.io/fjQsC
2
5
That's a good way of sharing papers @alexhdezgcia! I am glad of participating as a volunteer in #ICML2020 , I had the chance to help together with hundreds of volunteers in a great online event. I will leave here some of the papers I have been looking at.
At #ICML2020 I have decided to focus on a small number of papers and really try to understand them and engage with the authors and other participants in the presentations. As opposed to trying to see a lot of posters. I will be posting here the most interesting papers I find.
2
3
"Data-Efficient Image Recognition with Contrastive Predictive Coding" Nice insights on CPC and interesting experiments (especially with all the self-supervised approaches that are coming out these days).
1
"Training Neural Networks for and by Interpolation" Very clear explanation in the presentation: they propose a way of adapting the learning rate through training. I am looking forward to reading the paper more deeply.
If you are attending #ICML2019 and are interested in learning in the presence of label noise, come see our poster on "Unsupervised Label Noise Modeling and Loss Correction" arxiv.org/abs/1904.11238 this evening from 6:30pm. @oconnorn @ArazoEric @insight_centre
4
15
Our upcoming #ICML2019 paper "Unsupervised Label Noise Modeling and Loss Correction" is now on arXiv arxiv.org/abs/1904.11238. New SoTA in CIFAR10/100 under high-levels of noise using Beta mixtures to model & correct losses. Code: git.io/fjsvE @oconnorn @kevinmcguinness
1
3
Eric Arazo retweeted
It is official now: @BCN_ai is born. Long life to #BCNai !
3
13
38
Eric Arazo retweeted
Super delicious diagram. And cool interactive tool too: evogeneao.com/explore/tree-o…
17
367
41
1,052