DeepMind Perceiver and Perceiver IO | Paper Explained

👨‍👩‍👧‍👦 JOIN OUR DISCORD COMMUNITY: Discord ► https://discord.gg/peBrCpheKE 📢 SUBSCRIBE TO MY MONTHLY AI NEWSLETTER: Substack ► https://aiepiphany.substack.com/ ❤️ Become The AI Epiphany Patreon ❤️ ► https://www.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. ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ ✅ Perceiver: https://arxiv.org/abs/2103.03206 ✅ Perceiver IO: https://arxiv.org/abs/2107.14795 ✅ Code: https://github.com/deepmind/deepmind-research/tree/master/perceiver ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ ⌚️ 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 ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 💰 BECOME A PATREON OF THE AI EPIPHANY ❤️ If these videos, GitHub projects, and blogs help you, consider helping me out by supporting me on Patreon! The AI Epiphany ► https://www.patreon.com/theaiepiphany One-time donation: https://www.paypal.com/paypalme/theaiepiphany Much love! ❤️ Huge thank you to these AI Epiphany patreons: Eli Mahler Petar Veličković Bartłomiej Danek Zvonimir Sabljic ▬▬▬▬▬▬