Andrei C. Coman retweetledi
⏰ We're ready for Idiap's upcoming Perspectives on #AI event and very much look forward to listening to our guest speakers🔍
Register now
▶️eventbrite.ch/e/ai-and-democ…
Full program
▶️perspectives-on-ai.idiap.ch
English
Andrei C. Coman
92 posts
Andrei C. Coman
@AndreiCComan
Research Assistant @idiap_ch | Ph.D. Candidate @EPFL | Prev. Intern @AmazonScience
Martigny, Switzerland Katılım Ocak 2012
589 Takip Edilen124 Takipçiler
Andrei C. Coman retweetledi
The next few years are going to be critical for setting the norms for safe deployment and societal integration of AI.
We've launched the #SwissAI initiative to meet this challenge.
English
Andrei C. Coman retweetledi
Are Transformers graph-to-graph models?
In our #EMNLP2023 (BigPicture) #NLProc paper, we argue that attention weights and graph structure edges are effectively the same thing.
Link: arxiv.org/abs/2310.17936
Code: github.com/idiap/g2g-tran…
English
Andrei C. Coman retweetledi
As #AI technologies reach everyday new performances, their #energy cost is also increasing significantly.
Our researchers are proposing a novel approach to address this challenge during a period of rising #EnergyCosts.
About more cost effective AI
▶️idiap.ch/en/allnews/tow…
English
Andrei C. Coman retweetledi
5 #postdoc openings @Idiap_ch on
Neuro-symbolic learning & reasoning
Multilingual natural language proc.
Large language models
Generative #AI & large language models
Computational biology
w/ @RLuisier @LonnekePlas @andrenfreitas @DamienTeney
idiap.ch/en/join-us/job…
pls RT
English
Andrei C. Coman retweetledi
New #JobOpening ❗️
Our Computation, Cognition, and Language group is looking for a #postdoc to complement its cross-disciplinary team for a duration of 2 years. This position is under the supervision of Lonneke van der Plas.
Find out more and apply now
▶️careers.werecruit.io/en/idiap/offre…
English
Andrei C. Coman retweetledi
Our work about explaining GNNs via Logic has been accepted at #ICLR2023 🎉
While waiting for the conference, you can check out my talk here youtu.be/7dCrlFDqeB4
arxiv.org/abs/2210.07147
@AntonioLonga94 @pietro_barbiero @pl219_Cambridge @andrea_whatever
YouTube
English
Andrei C. Coman retweetledi
🤗🚨 The demo of SMaLL-100 model is now available on HF space: huggingface.co/spaces/alireza…
#MT #NLProc
@naverlabseurope @Idiap_ch
Alireza Mohammadshahi@alireza_mshi
1/n: Our paper "SMaLL-100: Introducing Shallow Multilingual MT Model for Low-Resource Languages" is accepted to #EMNLP2022 . It is a compact multilingual model covering 10K language pairs. 👉 arxiv.org/abs/2210.11621 @laurent_besacie @JamieBHenderson @naverlabseurope @Idiap_ch
English
Andrei C. Coman retweetledi
I am excited to announce our @iclr_conf 2023 paper:
"A VAE for Transformers with Nonparametric Variational Information Bottleneck"
openreview.net/pdf?id=6QkjC_c…
We propose to model Transformer embeddings as nonparametric mixture distributions using Dirichlet processes. @FabioFehr
English
Andrei C. Coman retweetledi
I am very excited to announce my first paper towards my PhD has been accepted at @iclr_conf 2023 in Kigali, Rwanda!
"A VAE for Transformers with Nonparametric Variational Information Bottleneck"
@JamieBHenderson @Idiap_ch
arxiv.org/abs/2207.13529
openreview.net/forum?id=6QkjC…
English
Andrei C. Coman retweetledi
Attending #EMNLP2022, the first in-person conference as an #Amazonian—looking to meet anyone interested in different flavors of QA, intern positions with my team at Alexa (semi-structured QA) and anything cool you have done so far! @AmazonScience @alexa99 #NLProc
English
Andrei C. Coman retweetledi
I have an opening for a 6-month internship at Idiap in Switzerland. MSc/PhD students/postdocs. Expertise should be a close match with our current work e.g. vision/language, OOD generalization, domain adaptation, causal representation learning... Questions/applications by email!
English
Andrei C. Coman retweetledi
We are #hiring ❗️
We invite applications for 2 #PhD and 2 #Postdoc Positions to join our #RobotLearning & Interaction group.
This position is under the supervision of Dr. Sylvain Calinon.
To apply or find out more
PhD ▶️ careers.werecruit.io/en/idiap/offer…
Postdoc ▶️careers.werecruit.io/en/idiap/offer…
English
Andrei C. Coman retweetledi
Our colleagues Florian Mai , Arnaud Pannatier , Fabio James Fehr , Haolin Chen , François Marelli , François Fleuret and James Henderson are glad to share their new academic writing.
Find out more about their work ⬇️
AK@_akhaliq
HyperMixer: An MLP-based Green AI Alternative to Transformers abs: arxiv.org/abs/2203.03691
Français
Andrei C. Coman retweetledi
TL;DR: we propose an MLP-Mixer whose parameters are modulated by an MLP. It avoids the O(T^2), captures long-range dependencies attention-style, and is easier to meta-optimize!
With @_florianmai, @ArnaudPannatier, @fabiofehr, H. Chen, F. Marelli, and @JamieBHenderson.
AK@_akhaliq
HyperMixer: An MLP-based Green AI Alternative to Transformers abs: arxiv.org/abs/2203.03691
English
Andrei C. Coman retweetledi
@KarimiRabeeh @MetaAI @myazdani999 @LukeZettlemoyer @vesko_st @JamieBHenderson @marzieh_saeidi Congrats Rabeeh!
English
Yay! My internship at @MetaAI got accepted in the ACL Main. I am grateful to my wonderful manager @myazdani999 and my awesome collaborators @LukeZettlemoyer @vesko_st, @jamieBHenderson @marzieh_saeidi, and Lambert Mathias #NLProc #ACL2022 This internship was my dream internship!
English
Andrei C. Coman retweetledi
Can we enforce (knowledge) structure to a text encoder using the Contrastive Learning paradigm?
In our #CoNLL2021 paper we address the research question.
Co-authors: @JamieBHenderson , @AndreiCComan , M.F Moens
📖: arxiv.org/pdf/2109.00840…
💻: github.com/christos42/CLD…
#IE
English
Andrei C. Coman retweetledi
NLP students, look out for openings. We now have multiple core NLP groups, plus several that work on multi-disciplinary problems including NLP, with group heads that are very driven and able. @Idiap_ch is currently growing into one of the strongest #NLProc labs in Europe.
Idiap Research Institute@Idiap_ch
We welcome our first female professor! Head of our #Computation, #Cognition & #Language Group, Lonneke van der Plas was nominated professor by the @UMmalta. She is affiliated to this university since 2014. Congratulations Professor van der Plas! 🙌 More ▶️idiap.ch/en/allnews/idi…
English
Andrei C. Coman retweetledi
Today at #ACL2021NLP - Parameter-efficient Multi-task Fine-tuning for Transformers via Shared Hypernetworks
If at ACL, come talk to us with @seb_ruder @m__dehghani @JamieBHenderson !
Date: Monday, 2 August 2021
Time: 11:20 UTC - 11:30 UTC
Session: #success" target="_blank" rel="nofollow noopener">zoom.us/j/92062341045?…
Rabeeh Karimi@KarimiRabeeh
I am happy to present my work with Google Brain and DeepMind at #ACL2021: arxiv.org/abs/2106.04489 We propose HyperFormer, a parameter-efficient multi-task framework that outperforms single and multi-task T5 on GLUE while adding only 0.29% per task.
English