
Senior AI Engineer | Applied Scientist @SonyAI_global | PhD from @EPFL đšđ
Lausanne, Switzerland
Joined July 2020
- Tweets20
- Following167
- Followers125
- Likes114
Pinned Tweet
Our work has been published in @Nature ! đđ
FHIBE redefines ethical data collection and pushes fairness benchmarks forward in human-centric computer vision.
A massive congrats to the whole team! đ
Paper: nature.com/articles/s41586-0âŠ
#FHIBE #Nature
4
71
Apostolos Modas retweeted
Nature research paper: Fair human-centric image dataset for ethical AI benchmarking
go.nature.com/4hSUagA
Apostolos Modas retweeted
Glad to have two papers accepted to #EMNLP2024 as main! đ
1) "Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes" arxiv.org/abs/2407.03623 w/ Jerone Andrews, @dorazhao9, @SciOrestis, @amodas_, @YutaNakashima4, @alicexiang
đ§”1/2
Do you want to learn more about ethical and responsible image data collection? Then come find us at #NeurIPS2023 đ€
Oral: Thu. 10:30 CST (Oral 5B) đ
Poster #1602: Thu. 10:45 CST (Poster session 5) đšâđŒ
Our paper raises ethical considerations in visual data collection, and offers perspectives on how to think about them from the start.
Work led by Jerone Andrews! With @dorazhao9 , myself, @amodas_ ,@SciOrestis ,@alicexiang
Paper: arxiv.org/abs/2302.03629
#NeurIPS2023
6
306
Apostolos Modas retweeted
Are you amazed by weight interpolation techniques to edit large pre-trained models? đ€
Then youâre going to love our work đ: âTask Arithmetic in the Tangent Spaceâđ which explains why they work and gives a simple way to make them better! đ€
đ: rb.gy/q9djm
đ§”1/9
Apostolos Modas retweeted
Are you amazed by NeRFs, SIRENs or FFNs but donât know why they work? đ€
Then youâre going to love our work đ: âA structured dictionary perspective on implicit neural representationsâ @CVPR which explains how INRs manage to look so crisp! đ€
đ: https://xitter.cf/t.co/29uNJrKap8
đ§” 1/9
Apostolos Modas retweeted
If you can donate money to support Ukraine now, here are some vetted funds đ
Apostolos Modas retweeted
"Deep nets are commonly believed to always converge to the simplest possible solution. This study elegantly demonstrates that they sometimes prefer complex solutions to simple ones, depending on the inductive biases in their architecture" ~Review by Anon rentry.co/rev_fs_0002
Apostolos Modas retweeted
When there exist multiple valid solutions for a given task, do neural networks always go for the âsimplestâ one?
with @gortizji, @ItamarSalazar, @modasapo and @pafrossard.
Happy to hear your thoughts on that :)
A neural anisotropic view of underspecification in deep learning
deepai.org/publication/a-neuâŠ
by Guillermo Ortiz-Jiménez et al. including @pafrossard
#MachineLearning #DeepLearning
Apparently adversarial robustness improves transfer learning even in non-trivial classification tasks!đ§
Check our new paper "Improving filling level classification with adversarial training" đ·đ„
đarXiv: https://xitter.cf/t.co/OhQCLyrWEB
@pafrossard @smartcameras @aXompi @RicSanMat
Hours of thorough and hard work, now published in the Proceedings of IEEE đŁđ„ł
Many credits to my collaborators @gortizji @smoosavid @pafrossard! đȘ
Interested in adv. robustness beyond security, or how adv. examples can help us understand deep learning? đ€
Check our review paper âOptimism in the face of adversityâ, just published in the Proc. IEEE đ
đXplore: https://xitter.cf/t.co/4WaL6OZIuT đarXiv: https://xitter.cf/t.co/x8BnM17mle
A huge DenseNet vs an old-fashioned Logistic Regression, solving a linearly separable problem. Who is gonna win??
Why do state-of-the-art NNs only generalise on a few linearly separable problems? đ€
Figure it out at our poster today at 6-8pm CET (9-11am PST) @ Town E1, spot D3 #NeurIPS2020
đĄ Gather town: https://xitter.cf/t.co/DlnnZsU4Lu
Our framework provides a geometric connection between adversarial examples and data features! Find us today 6-8pm CET (9-11am PST) @ Town A0, spot A0.
#NeurIPS2020
Want to know what happens when you train a CNN on frequency reversed samples and what this has to do with robustness? Then join us at our poster in #NeurIPS2020 đ€
âHold me tight! Influence of discriminative features on deep network boundariesâ
đ: https://xitter.cf/t.co/tApRCvijDB
Apostolos Modas retweeted
Nicolas Flammarion is looking for #PhD student. Find out more about his #research epfl.ch/labs/tml/, and learn more about our @EPFL #EDIC #computerscience PhD program: go.epfl.ch/phd-edic
Our new "tutorial" on what we've learned from adversarial robustness beyond security is online!
Still a pre-print, so any comment/criticism is more than welcome! :)
With @gortizji, @smoosavid, @pafrossard
Seeing the glass as half fullđ„Optimism in the Face of Adversity arxiv.org/pdf/2010.09624v1.pâŠ
In this pre-print we talk about what adv. examples reveal about the geometry of deep nets and how other topics in ML can benefit from such insights.
With @gortizji , @modasapo, @pafrossard
6
Apostolos Modas retweeted
Very excited with the new blog post of our NeurIPS 2020 paper âNeural Anisotropy Directionsâ where we study the role of architecture in deep learning.
Co-authors: @modasapo, @smoosavid and @pafrossard.
đ» Blog: https://xitter.cf/t.co/eC1pRyWzkV
đ Paper: https://xitter.cf/t.co/o7RrW7n8OY
1/n
Apostolos Modas retweeted
If you are interested to know how âRedundant features can hurt robustness to distribution shiftâ check out our virtual poster at ICML'20 UDL Workshop.
This is joint work with @ApostolosModas, @smoosavid and @pafrossard.
đ Paper: https://xitter.cf/t.co/XDqVZatZQy
(1/11)



















