Julia Gusak
@juliagusak502
Followers
129
Following
190
Media
2
Statuses
26
Research Scientist @inria, former @Skoltech
Joined October 2019
RT @inria_grenoble: [EVENT 📣]. The end 😥.The Grenoble #ArtificialIntelligence for Physical Sciences workshop is now over !. Thanks to all s….
0
3
0
RT @JulyanArbel: Learning a lot about AI for physical sciences at the GAP workshop in Grenoble today.✅PINNs with @EBezenac.✅Neural ODEs & D….
0
3
0
#WANT @NeurIPSConf workshop is happening now! Join for talks and panel on efficient training at room 243-245 or at Next poster session is at 5-5.30pm (GMT+6) offline and in Gather Town. See you there!.
0
0
1
RT @Inria_Bordeaux: Dans le cadre de la conférence #NeurIPS2023, participez à une collecte de statistiques sur la façon dont les #réseaux….
0
5
0
Come to #WANT @NeurIPSConf on Dec'16 to discuss tips and tricks for efficient training! Share your thoughts and insights by filling out the poll ! Stats will be posted during the workshop! #AI #HPC #EfficientTraining.
docs.google.com
Please fill in the poll below to help us understand how the deep learning community trains neural networks, what the computational needs are, and which tricks to improve efficiency are used or...
0
2
6
RT @JeanKossaifi: 📢 Join us at the #AI #WANTScale workshop at #NeurIPS 2023 to learn all about training neural architectures at scale, and….
0
3
0
RT @Inria_Bordeaux: Dans le cadre de #NeurIPS2023, l'équipe Topal, Scool, @nvidiaAI et l'@ufrj organisent un atelier sur l'entraînement eff….
0
4
0
Delighted to be a part of an amazing organizing team from @nvidia @inria and @ufrj for #WANT @NeurIPSConf! Exciting times ahead with @JeanKossaifi @AnimaAnandkumar, OlivierBeaumont, AlenaShilova, Cristiana Barbosa Bentes! 🚀🤝 #AI #HPC #EfficientTraining #NeurIPS2023.
0
0
1
Excited to announce our Workshop on Advancing Neural Network Training, WANT @NeurIPSConf!🚀 Save GPU hours, keep accuracy! Join HPC & AI experts on Dec 16, submit papers by Sep 29. #AI #WANT #HPC #NeurIPS2023 #EfficientTraining Details:
1
13
14
Thrilled to have two papers presented at @icmlconf! Dive in and explore our research on memory-efficient training of neural networks! #ICML2023 #EfficientML
0
1
10
Check out our new papers on memory footprint reduction during training!.
We present the new approximate backward functions which reduce the memory for activations. Interesting math behind! .Paper about activation functions:.About linear layers: .Сode:.
0
0
8
Participate in NeurIPS 2021 Shifts Challenge on uncertainty and robustness to distributional shift! Test how robust is your model!.
1/7 We are happy to launch the Shifts Challenge on uncertainty and robustness to distributional shift @NeurIPSConf! The challenge is jointly organized by Yandex.Research together with the OATML group (@OATML_Oxford) the Cambridge University Speech Group.
0
0
3
Can't wait to learn an interpolation based trick to speed up NeuralODEs? Come to our #NeurIPS2020 poster at 9:00-11:00 PDT, Dec. 8th (Poster Session 2)!
#NeurIPS2020 . Interpolation Technique to Speed Up Gradients Propagation in Neural ODEs. Poster (Session 2): Tue, Dec 8th, 2020 @ 09:00 – 11:00 PST. Code: Arxiv:
0
0
5
Thanks a lot to Mathieu Salzmann, Gabriela Csurka, Tatiana Tomassi, and Timothy M. Hospedales (@tmh31) for clear, informative lectures and engaging Q&A session! . Tutorial ECCV page:
1
0
2
Glad to announce our paper "Stable Low-rank Tensor Decomposition for Compression of Convolutional Neural Network" has been accepted to #ECCV2020! Please, join our poster zoom sessions today (August 25) at 6.00-8.00 am (UTC + 1) and 14.00 - 16.00 (UTC + 1)
arxiv.org
Most state of the art deep neural networks are overparameterized and exhibit a high computational cost. A straightforward approach to this problem is to replace convolutional kernels with its...
2
7
34
RT @JeanKossaifi: Excited to be presenting my work on Factorized Higher-Order CNNs, with an Application to Spatio-Temporal Emotion Estimati….
0
12
0
Check out our new paper).
Active subspaces and neural networks: Simple idea but potentially very powerful.
0
0
4
RT @facebookai: OctConv can replace a standard convolution in neural networks without requiring any other network architecture adjustments.….
0
62
0