CorentK

👽 ALiEN Postdoc @ COLT UPF | Investigating unnatural language processing.
~ PhD on DL applied to V&L, studying bias vs. reasoning in VQA.

[CorentK @CorentK]

🎉Life is so good now that I am a doctor! But I already feel nostalgic for these incredible three years w/ my fantastic advisers @antigregory @moezbac @chriswolfvision 💖

Thanks to the PhD committee @thomenicolas1 @david_picard @CordeliaSchmid @DamienTeney @zeynepakata 🙏

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Congrats to a successful PhD defense Dr. @CorentK! An excellent presentation after 3 years of hard and great work.

Thanks to the PhD committee @david_picard @thomenicolas1 @CordeliaSchmid @DamienTeney @zeynepakata.

Happy supervisors: @moezbac @antigregory and yours, truly!

[CorentK @CorentK]

👨‍🎓 I am pleased to announce that I will defend my thesis entitled “Bias and reasoning in Visual Question Answering”, supervised by @chriswolfvision @moezbac and @antigregory, next Thursday (09/12) at 10am, Paris time.
Send me a pm if you wish to attend 😊

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Our only submission to #NeurIPS2021 has been accepted. Work by @CorentK, yours truly, @antigregory, @moezbac, M. Nadri.

We transfer reasoning patterns from Oracle VQA models to noisy models, regularizing with an auxiliary loss for program prediction.

https://t.co/dBquSdkLOJ

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Congrats to @CorentK, he is outstanding reviewer at @ICCV_2021! Spending time on high-quality work pays off.

[Théo Jaunet @jaunet_theo]

Excited to see our paper accepted to #ieeevis 2021!

“VisQA: X-raying Vision and Language Reasoning in Transformers”, a deep dive in attention maps to grasp bias in transformers.

Demo: https://t.co/QORx1PyBuO
Paper: https://t.co/GIb7SBmRzj

Now time to write my doctoral thesis😱

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Can oracles (which take GT visual input) learn better reasoning patterns in VQA? Can we visualize them? Are they linked to language/task functions? Can they be transferred to deployable models?

Answers will be given Tuesday by @CorentK
#CVPR2021

12am CET = 6am EST = 3am PT

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Roses are red, Violets are blue ... but should VQA expect them to?
#CVPR2021 poster presented tomorrow Tuesday by @CorentK
4am CET = 10pm EST = 7pm PT

We need to get up at 4am, so you better be there to joint us 😏

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

New paper: we show that transfer from Oracle input to real improves VQA reasoning when a strong link is established through program supervision: empirical results + analysis of sample complexity.

Work with @CorentK @antigregory @moezbac and M. Nadri.
https://t.co/dBquSdCndj

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

We introduce a new benchmark to evaluate VQA in out-of-distribution settings by reorganizing the GQA dataset, taylored for each sample (question group), targeting research in bias reduction in VQA.
Work by @CorentK, @antigregory, @moezbac and yours, truly.
https://t.co/Fji3P7jJgd

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

Second paper: we introduce a new semantic loss for VQA adding structure to the VQA answer space estimated from redundancy in annotations, questioning the classification approach to VQA.
Work by @CorentK, @antigregory, @moezbac and yours, truly.
https://t.co/hvPFvhdtlm

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

New paper from @CorentK, co-supervised with @moezbac @antigregory. We show that object-word alignment does not emerge naturally in BERT like training of transformers on vision-language task and propose weak supervision for it, improving performance:
https://t.co/rB2lI9NrZU