Pablo

Machine Learning and Signal Processing researcher

[Pablo @pablixnm]

A Linearscope (ICCV19) is a framework that hides the non-linearities in your architecture and let you see the action of your model as a gigantic and adaptive linear system. For every input you get a different linear system. #XAI https://t.co/idLG1rc33C @XAI_Research @MLStreetTalk

[Pablo @pablixnm]

Poster in ICCV19:
https://t.co/DicV3Hsikp

[Pablo @pablixnm]

"A Tour of Convolutional Networks Guided by Linear Interpreters" at #iccv2019 on Oct 31st, 10:30–13:00 Poster 3.1 (Hall B) poster #33. Also at Explainable AI (#XAI) Workshop on Nov 2nd. Learn about #linearscopes!
https://t.co/idLG1rt65C

[Pablo @pablixnm]

A Linearscope (ICCV19) is a framework that hides the non-linearities in your architecture and let you see the action of your model as a gigantic and adaptive linear system. For every input you get a different linear system. #XAI https://t.co/idLG1rc33C @XAI_Research @MLStreetTalk

[Pablo @pablixnm]

The eigenvectors show local features for the largest eigenvalues! (eyes, mouth, etc). This is remarkable since each conv-layer is linear-space-invariant with Fourier-like eigenvectors. Somehow, the composition with ReLUs can change a global basis into a local basis😱.

[Pablo @pablixnm]

Presentation in https://t.co/mfwM4tO9QU #wacv2022 #wacv

[Pablo @pablixnm]

"edge-SR: Super-Resolution For The Masses"
Happy to share that our submission to WACV 2022 got accepted!
My favorite take-away from the paper is that we can understand Self-Attention as a Template-Matching module (Matching-Filter for the DSP crowd).
#wacv @wacv2022

[Pablo @pablixnm]

Source code in https://t.co/01tExrZVZz
#wacv2022 #wacv

[The Nobel Prize @NobelPrize]

Roger Penrose – awarded this year’s #NobelPrize in Physics – was born in 1931 in Colchester, UK.

He is a professor at @UniofOxford, UK.

https://t.co/Jm9iOU99N3

[Pablo @pablixnm]

"A Tour of Convolutional Networks Guided by Linear Interpreters" at #iccv2019 on Oct 31st, 10:30–13:00 Poster 3.1 (Hall B) poster #33. Also at Explainable AI (#XAI) Workshop on Nov 2nd. Learn about #linearscopes!
https://t.co/idLG1rt65C

[Pablo @pablixnm]

"A Tour of Convolutional Networks Guided by Linear Interpreters" is accepted to #iccv2019. Interpretability for classification (e.g. VGG), super-resolution (e.g. EDSR) and I2I (e.g. CycleGAN) using #linearscopes.
https://t.co/idLG1rt65C

[Pablo @pablixnm]

MGBP : Multi-Grid Back-Projection super-resolution, today at #AAAI2019
https://t.co/JzxOaEMqz1