Uploaded September 2021 | Updated September 2026, 1 week ago
π¨βπ©βπ§βπ¦ JOIN OUR DISCORD COMMUNITY:
Discord βΊ discord.gg/peBrCpheKE
π’ SUBSCRIBE TO MY MONTHLY AI NEWSLETTER:
Substack βΊ aiepiphany.substack.com
β€οΈ Become The AI Epiphany Patreon β€οΈ βΊ patreon.com/theaiepiphany
In this video I cover:
* Perceiver (Perceiver: General Perception with Iterative Attention)
* Perceiver IO (Perceiver IO: A General Architecture for Structured Inputs & Outputs)
The goal was to create a modality-agnostic, general perception architecture that could work on images, videos, audio, text, etc. alike.
The main idea is to use the cross-attention module as a bottleneck layer that will map the input modality data into the latent space - this way we avoid the quadratic curse of transformers. After that powerful latent transformers are used to refine the representation - rinse and repeat.
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
β Perceiver: arxiv.org/abs/2103.03206
β Perceiver IO: arxiv.org/abs/2107.14795
β Code: github.com/deepmind/deepmind-research/tree/master/perceiver
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
βοΈ Timetable:
00:00 Intro
02:00 Perceiver architecture explained
05:40 Comparison with Facebook DETR model
07:05 Comparison to RNNs
08:35 Algorithmic complexity of Perceiver
10:35 Positional encodings and permutation equivariance
12:00 Results - ImageNet
14:35 Pixel permutation robustness
17:40 Attention visualized
20:20 Results - AudioSet
23:30 Results - Point Cloud
25:00 Perceiver IO
26:15 Decoder explained in depth (main contribution)
28:45 GLUE results (BERT baseline)
29:50 Outro
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π° BECOME A PATREON OF THE AI EPIPHANY β€οΈ
If these videos, GitHub projects, and blogs help you,
consider helping me out by supporting me on Patreon!
The AI Epiphany βΊ patreon.com/theaiepiphany
One-time donation:
paypal.com/paypalme/theaiepiphany
Much love! β€οΈ
Huge thank you to these AI Epiphany patreons:
Eli Mahler
Petar VeliΔkoviΔ
BartΕomiej Danek
Zvonimir Sabljic
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π‘ The AI Epiphany is a channel dedicated to simplifying the field of AI using creative visualizations and in general, a stronger focus on geometrical and visual intuition, rather than the algebraic and numerical "intuition".
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π CONNECT WITH ME ON SOCIAL
LinkedIn βΊ linkedin.com/in/aleksagordic
Twitter βΊ twitter.com/gordic_aleksa
Instagram βΊ instagram.com/aiepiphany
Facebook βΊ facebook.com/aiepiphany
π¨βπ©βπ§βπ¦ JOIN OUR DISCORD COMMUNITY:
Discord βΊ discord.gg/peBrCpheKE
π’ SUBSCRIBE TO MY MONTHLY AI NEWSLETTER:
Substack βΊ aiepiphany.substack.com
π» FOLLOW ME ON GITHUB FOR ML PROJECTS:
GitHub βΊ github.com/gordicaleksa
π FOLLOW ME ON MEDIUM:
Medium βΊ gordicaleksa.medium.com
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
#perceiver #perceiverio #deepmind
π¨βπ©βπ§βπ¦ JOIN OUR DISCORD COMMUNITY:
Discord βΊ discord.gg/peBrCpheKE
π’ SUBSCRIBE TO MY MONTHLY AI NEWSLETTER:
Substack βΊ aiepiphany.substack.com
β€οΈ Become The AI Epiphany Patreon β€οΈ βΊ patreon.com/theaiepiphany
In this video I cover:
* Perceiver (Perceiver: General Perception with Iterative Attention)
* Perceiver IO (Perceiver IO: A General Architecture for Structured Inputs & Outputs)
The goal was to create a modality-agnostic, general perception architecture that could work on images, videos, audio, text, etc. alike.
The main idea is to use the cross-attention module as a bottleneck layer that will map the input modality data into the latent space - this way we avoid the quadratic curse of transformers. After that powerful latent transformers are used to refine the representation - rinse and repeat.
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
β Perceiver: arxiv.org/abs/2103.03206
β Perceiver IO: arxiv.org/abs/2107.14795
β Code: github.com/deepmind/deepmind-research/tree/master/perceiver
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
βοΈ Timetable:
00:00 Intro
02:00 Perceiver architecture explained
05:40 Comparison with Facebook DETR model
07:05 Comparison to RNNs
08:35 Algorithmic complexity of Perceiver
10:35 Positional encodings and permutation equivariance
12:00 Results - ImageNet
14:35 Pixel permutation robustness
17:40 Attention visualized
20:20 Results - AudioSet
23:30 Results - Point Cloud
25:00 Perceiver IO
26:15 Decoder explained in depth (main contribution)
28:45 GLUE results (BERT baseline)
29:50 Outro
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π° BECOME A PATREON OF THE AI EPIPHANY β€οΈ
If these videos, GitHub projects, and blogs help you,
consider helping me out by supporting me on Patreon!
The AI Epiphany βΊ patreon.com/theaiepiphany
One-time donation:
paypal.com/paypalme/theaiepiphany
Much love! β€οΈ
Huge thank you to these AI Epiphany patreons:
Eli Mahler
Petar VeliΔkoviΔ
BartΕomiej Danek
Zvonimir Sabljic
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π‘ The AI Epiphany is a channel dedicated to simplifying the field of AI using creative visualizations and in general, a stronger focus on geometrical and visual intuition, rather than the algebraic and numerical "intuition".
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
π CONNECT WITH ME ON SOCIAL
LinkedIn βΊ linkedin.com/in/aleksagordic
Twitter βΊ twitter.com/gordic_aleksa
Instagram βΊ instagram.com/aiepiphany
Facebook βΊ facebook.com/aiepiphany
π¨βπ©βπ§βπ¦ JOIN OUR DISCORD COMMUNITY:
Discord βΊ discord.gg/peBrCpheKE
π’ SUBSCRIBE TO MY MONTHLY AI NEWSLETTER:
Substack βΊ aiepiphany.substack.com
π» FOLLOW ME ON GITHUB FOR ML PROJECTS:
GitHub βΊ github.com/gordicaleksa
π FOLLOW ME ON MEDIUM:
Medium βΊ gordicaleksa.medium.com
β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬β¬
#perceiver #perceiverio #deepmind
