Janis Postels

24 posts

Janis Postels

@JanisPostels

Katılım Eylül 2019

221 Takip Edilen107 Takipçiler

Sabitlenmiş Tweet

Janis Postels@JanisPostels·16 Tem

Will be presenting our work "On the Practicality of Deterministic Epistemic Uncertainty" next week at #ICML2022 in Baltimore. If you are interested checkout our paper or visit us at our poster on Tue Jul 19 between 06:30 PM & 08:30 PM @ Hall E 506 arxiv.org/abs/2107.00649

English

Janis Postels@JanisPostels·28 Kas

Looking forward to this :)

Bertrand Charpentier@Bertrand_Charp

Jakob Heiss will present his ICML paper "NOMU - Neural Optimization-based Model Uncertainty" at the Uncertainty in AI reading group today at 5:30pm (Berlin time). uncertainty-reading-group.github.io/2021-11-28-tal… Co-author: @JWeissteiner, Hanna Wutte, @SvenSeuken, Josef Teichmann

English

Janis Postels@JanisPostels·15 Eyl

@mentzer_f Congrats 🎉

English

Fabian Mentzer@mentzer_f·15 Eyl

Will be presented at #NeurIPS! 🎉🎉🎉

Fabian Mentzer@mentzer_f

Very happy to finally show our latest work: VCT, a transformer for video compression 🤖 We radically simplify the video compression setup by getting rid of motion prediction, warping, and residuals. Instead, a transformer learns to model temporal redundancies purely from data:

English

Janis Postels@JanisPostels·12 Eyl

Will be presenting our work "ManiFlow: Implicitly Representing Manifolds with Normalizing Flows" at #3DV tomorrow! We show how to sample from manifolds (e.g. Point Clouds) with normalizing flows! @MDanelljan Luv Van Gool @fedassa Paper: arxiv.org/abs/2208.08932

English

Janis Postels@JanisPostels·12 Eyl

@lilianweng Great summary! Would have saved me a lot of time if I'd seen it a little earlier :-D

English

Lilian Weng@lilianweng·12 Eyl

🧮 I finally spent some time learning what exactly Neural Tangent Kernel (NTK) is and went through some mathematical proof. Hopefully after reading this, you will not feel all the math behind NTK is that scaring, but rather, quite intuitive. lilianweng.github.io/posts/2022-09-…

English

173

Janis Postels@JanisPostels·4 Ağu

Further, we invite you to share your recent/ongoing work in this area. There will be a total prize pool of up to 1500$ distributed among the best submissions. openreview.net/group?id=IROS.…)

English

Janis Postels@JanisPostels·4 Ağu

This WS brings together experts from both fields - probabilistic methods in robotics and deep learning - to provide a forum for this urgent topic. @wolfram_burgard @contactrika @nikoSuenderhauf Fabio Ramos Teresa Vidal @yaringal Ali Agha @balajiln @padsmagt @IngmarPosner

English

Janis Postels@JanisPostels·4 Ağu

Very happy to announce our workshop on Probabilistic Robotics at #IROS 2022 in Kyoto. probabilisticrobotics.github.io Jointly organized with @FengJianxiang @JongseokLee11 @matthiashumt @YizheWu5 Rudolph Triebel

English

Janis Postels retweetledi

Markus Nagel@mnagel87·21 Tem

If your are at #ICML2022, join us Thursday at 11:00 in the Ballroom 1&2 for our oral presentation on "Overcoming Oscillations in Quantization-Aware Training" and later at poster #227. Paper: arxiv.org/abs/2203.11086 @mfournarakis, @yell1337, @TiRune.

English

Janis Postels retweetledi

Augustine Mavor-Parker@MavorParker·20 Tem

Would you rather play games or watch random images? In our #ICML2022 work, we give RL agents this choice and turns out many curiosity algos would prefer to watch random images. We propose an acetylcholine-inspired solution to help curiosity to avoid these "noisy TVs"

English

Janis Postels retweetledi

Auke Wiggers@aukejw·20 Tem

The MobileCodec paper is now online: arxiv.org/abs/2207.08338 Check it out if you want to know how we were able to deploy a neural video codec to a mobile phone and decode 720p video in real time 📲

English

Janis Postels retweetledi

Ferjad Naeem@ferjadnaeem·19 Tem

I am excited to share that our work "3D Compositional Zero-shot Learning with DeCompositional Consensus" has been accepted at #ECCV2022. Joint work with @evinpinar @xyongqin, Luc Van Gool and @fedassa between ETH Zurich, TUM, and Google. arxiv.org/abs/2111.14673 [1/7]

English

Janis Postels@JanisPostels·17 Tem

@abursuc Congrats! Let me know when the preprint is available. Looking forward to that work and seeing you at ECCV!

English

Andrei Bursuc@abursuc·17 Tem

@JanisPostels Congrats Janis et al. for this work. I remember seeing it a while go. This work and MIR were among our inspirations to look into DUMs and we will present our work at next ECCV (preprint not yet online).

English

Janis Postels@JanisPostels·16 Tem

English

Janis Postels@JanisPostels·16 Tem

We shed light onto a recent trend in uncertainty quantification which treats the weights of DNNs deterministically. Turns despite good OOD detection these methods do not necessarily yield well calibrated uncertainty.

English

Janis Postels@JanisPostels·20 Haz

Great work guys!

Mattia Segu@MattiaSegu

@CVPR time has finally come! Check out our paper "SHIFT: A Synthetic Driving Dataset for Continuous Multi-Task Domain Adaptation" and drop by our poster session on Friday 24 (afternoon) at 257B to learn more! Project Page: vis.xyz/shift/ Paper: go.yf.io/shift-paper

English

Janis Postels retweetledi

Andrei Bursuc@abursuc·5 Haz

Cool new AD dataset in town 👉 SHIFT: A Synthetic Driving Dataset for Continuous Multi-Task Domain Adaptation by @JanisPostels @MattiaSegu @DrFisherYu @fedassa et al. vis.xyz/shift/ 1/

English

Janis Postels@JanisPostels·3 Ara

Single NFs struggle to model geometries with different geometric properties than the base distribution (e.g. transforming a Gaussian into a ring-shaped distribution) We show that mixtures of NFs are strong inductive bias to overcome this problem in a parameter-efficient way.

English

Janis Postels@JanisPostels·3 Ara

Glad to share that we will present our work "Go with the Flows: Mixtures of Normalizing Flows for Point Cloud Generation and Reconstruction" tomorrow at 3DV. Paper: arxiv.org/abs/2106.03135 Code: github.com/janisgp/go_wit… Poster at 11:20am GMT (03.12.)/00:20am GMT (04.12.) >>

English

Janis Postels@JanisPostels·25 Tem

@prunetruong Congrats :-)

English

Prune Truong@prunetruong·24 Tem

It got accepted as an oral at #ICCV2021!! 🎉🎉

Dmytro Mishkin 🇺🇦@ducha_aiki

Warp Consistency for Unsupervised Learning of Dense Correspondences @prunetruong @MDanelljan, @drfisheryu, Luc Van Gool arxiv.org/abs/2104.03308… tl;dr: Cycle-consistency and augmentation supervision come in very different flavors. Repo is there, but w/o code.

English

100

Janis Postels retweetledi

Federico Tombari@fedassa·9 Haz

Our work on Variational Transformer Networks for layout generation got featured on the Google AI blog. The work will be presented at CVPR21 on June 24th and a pre-release is available on arXiv. Blog post: Official Tweet: https:/…lnkd.in/dSuZVFx lnkd.in/dBgXeRp

English

Keşfet

@mentzer_f @MDanelljan @fedassa @lilianweng @wolfram_burgard @contactrika @nikoSuenderhauf @yaringal