Giancarlo Kerg

PhD student @MILAMontreal with Yoshua Bengio and Guillaume Lajoie.

[Giancarlo Kerg @GcKerg]

@GuiroySimon @CoLLAs_Conf Interesting work!

[Giancarlo Kerg @GcKerg]

3/ We propose a simple architecture merely based on similarity scores, named Compositional Relational Network (CoRelNet), not only generalizing to unseen objects but also to unseen relations on such tasks.

[Giancarlo Kerg @GcKerg]

1/ New preprint on key inductive biases for OoD generalisation on purely relational tasks, w/ @sarthmit, @David Rolnick, Yoshua Bengio, @tyrell_turing and @g_lajoie_ at @Mila_Quebec. https://t.co/bBR9VMsK8v

[Giancarlo Kerg @GcKerg]

2/ We formalize the advantages afforded by ‘partitioned’ representations of relations and sensory details

[Giancarlo Kerg @GcKerg]

@ilyaraz2 Lifting weights

[Giancarlo Kerg @GcKerg]

@ilyaraz2 What led to the epiphany moment ?

[Giancarlo Kerg @GcKerg]

@karimjerbineuro This is hilarious 😂

[Karim Jerbi @karimjerbineuro]

#2020

[Giancarlo Kerg @GcKerg]

@ilyaraz2 Interesting observation 😃

[Giancarlo Kerg @GcKerg]

@ilyaraz2 Great post - love it. Happy New Year!

[Hidenori Tanaka @Hidenori8Tanaka]

Every symmetry of a network has a corresponding conserved quantity through training under gradient flow (Noether's theorem for neural networks!)

For translation, scale, and rescale symmetry the flow is constrained to a hyperplane, sphere, and hyperbola respectively
4/8

[Giancarlo Kerg @GcKerg]

@ilyaraz2 Have you tried walking desk if the goal is to burn calories ? Walking helps to keep hunger down.

[Giancarlo Kerg @GcKerg]

@tsonj Congrats !!

[Giancarlo Kerg @GcKerg]

@ilyaraz2 Wow! Awesome progress! Respect! I lost about 20kg since Jan and I found it quite challenging already.

[Google Canada @googlecanada]

Canada is home to a collaborative network of some of the most creative and brilliant thinkers in AI today. Google Canada recognizes the importance of investing in curiosity-driven research and we're thrilled to reconfirm our support of @Mila_Québec 👉 https://t.co/dZQb1mRN0n

[Giancarlo Kerg @GcKerg]

@efweinberger @ryz_nz I just got an email from NeurIPS saying that the deadline was indeed Oct 21. I am very confused now.

[Victor Geadah @vgeadah]

Glad to share my first preprint! We introduce a biologically-inspired parametric activation, and investigate neural adaptation in RNNs.

https://t.co/3LQY9Z6qBU

Work done @MILAMontreal, with Giancarlo Kerg, Stefan Horoi and Guy Wolf, and under the mentorship of @g_lajoie_.

[Giancarlo Kerg @GcKerg]

Excited to announce our latest work (along with B.Kanuparthi, A. Goyal, K. Goyette, Y. Bengio and G. Lajoie): Untangling tradeoffs between recurrence and self-attention in neural networks
https://t.co/9pDM29WiNK
@anirudhg9119 @g_lajoie_