Perceiver: General Perception with Iterative Attention (Google DeepMind Research Paper Explained)

#perceiver #deepmind #transformer

Inspired by the fact that biological creatures attend to multiple modalities at the same time, DeepMind releases its new Perceiver model. Based on the Transformer architecture, the Perceiver makes no assumptions on the modality of the input data and also solves the long-standing quadratic bottleneck problem. This is achieved by having a latent low-dimensional Transformer, where the input data is fed multiple times via cross-attention. The Perceiver's weights can also be shared across layers, making it very similar to an RNN. Perceivers achieve competitive performance on ImageNet and state-of-the-art on other modalities, all while making no architectural adjustments to input data.

OUTLINE: 0:00 - Intro & Overview 2:20 - Built-In assumptions of Computer Vision Models 5:10 - The Quadratic Bottleneck of Transformers 8:00 - Cross-Attention in Transformers 10:45 - The Perceiver Model Architecture & Learned Queries 20:05 - Positional Encodings via Fourier Features 23:25 - Experimental Results & Attention Maps 29:05 - Comments & Conclusion

Paper: https://arxiv.org/abs/2103.03206

My Video on Transformers (Attention is All You Need): https://youtu.be/iDulhoQ2pro

Abstract: Biological systems understand the world by simultaneously processing high-dimensional inputs from modalities as diverse as vision, audition, touch, proprioception, etc. The perception models used in deep learning on the other hand are designed for individual modalities, often relying on domain-specific assumptions such as the local grid structures exploited by virtually all existing vision models. These priors introduce helpful inductive biases, but also lock models to individual modalities. In this paper we introduce the Perceiver - a model that builds upon Transformers and hence makes few architectural assumptions about the relationship between its inputs, but that also scales to hundreds of thousands of inputs, like ConvNets. The model leverages an asymmetric attention mechanism to iteratively distill inputs into a tight latent bottleneck, allowing it to scale to handle very large inputs. We show that this architecture performs competitively or beyond strong, specialized models on classification tasks across various modalities: images, point clouds, audio, video and video+audio. The Perceiver obtains performance comparable to ResNet-50 on ImageNet without convolutions and by directly attending to 50,000 pixels. It also surpasses state-of-the-art results for all modalities in AudioSet.

Authors: Andrew Jaegle, Felix Gimeno, Andrew Brock, Andrew Zisserman, Oriol Vinyals, Joao Carreira

Links: TabNine Code Completion (Referral): http://bit.ly/tabnine-yannick YouTube: https://www.youtube.com/c/yannickilcher Twitter: https://twitter.com/ykilcher Discord: ... https://www.youtube.com/watch?v=P_xeshTnPZg