Big Bird: Transformers for Longer Sequences (Paper Explained)

Yannic Kilcher · Beginner ·🧠 Large Language Models ·5y ago
#ai #nlp #attention The quadratic resource requirements of the attention mechanism are the main roadblock in scaling up transformers to long sequences. This paper replaces the full quadratic attention mechanism by a combination of random attention, window attention, and global attention. Not only does this allow the processing of longer sequences, translating to state-of-the-art experimental results, but also the paper shows that BigBird comes with theoretical guarantees of universal approximation and turing completeness. OUTLINE: 0:00 - Intro & Overview 1:50 - Quadratic Memory in Full Attention 4:55 - Architecture Overview 6:35 - Random Attention 10:10 - Window Attention 13:45 - Global Attention 15:40 - Architecture Summary 17:10 - Theoretical Result 22:00 - Experimental Parameters 25:35 - Structured Block Computations 29:30 - Recap 31:50 - Experimental Results 34:05 - Conclusion Paper: https://arxiv.org/abs/2007.14062 My Video on Attention: https://youtu.be/iDulhoQ2pro My Video on BERT: https://youtu.be/-9evrZnBorM My Video on Longformer: https://youtu.be/_8KNb5iqblE ... and its memory requirements: https://youtu.be/gJR28onlqzs Abstract: Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP. Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to 8x of what was previo
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Yannic Kilcher · Yannic Kilcher · 0 of 60

← Previous Next →
1 Imagination-Augmented Agents for Deep Reinforcement Learning
Imagination-Augmented Agents for Deep Reinforcement Learning
Yannic Kilcher
 2 Learning model-based planning from scratch
Learning model-based planning from scratch
Yannic Kilcher
 3 Reinforcement Learning with Unsupervised Auxiliary Tasks
Reinforcement Learning with Unsupervised Auxiliary Tasks
Yannic Kilcher
 4 Attention Is All You Need
Attention Is All You Need
Yannic Kilcher
 5 git for research basics: fundamentals, commits, branches, merging
git for research basics: fundamentals, commits, branches, merging
Yannic Kilcher
 6 Curiosity-driven Exploration by Self-supervised Prediction
Curiosity-driven Exploration by Self-supervised Prediction
Yannic Kilcher
 7 World Models
World Models
Yannic Kilcher
 8 Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
Yannic Kilcher
 9 Stochastic RNNs without Teacher-Forcing
Stochastic RNNs without Teacher-Forcing
Yannic Kilcher
 10 What’s in a name? The need to nip NIPS
What’s in a name? The need to nip NIPS
Yannic Kilcher
 11 BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Yannic Kilcher
 12 Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Yannic Kilcher
 13 GPT-2: Language Models are Unsupervised Multitask Learners
GPT-2: Language Models are Unsupervised Multitask Learners
Yannic Kilcher
 14 Neural Ordinary Differential Equations
Neural Ordinary Differential Equations
Yannic Kilcher
 15 The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
Yannic Kilcher
 16 Discriminating Systems - Gender, Race, and Power in AI
Discriminating Systems - Gender, Race, and Power in AI
Yannic Kilcher
 17 Blockwise Parallel Decoding for Deep Autoregressive Models
Blockwise Parallel Decoding for Deep Autoregressive Models
Yannic Kilcher
 18 S.H.E. - Search. Human. Equalizer.
S.H.E. - Search. Human. Equalizer.
Yannic Kilcher
 19 Reinforcement Learning, Fast and Slow
Reinforcement Learning, Fast and Slow
Yannic Kilcher
 20 Adversarial Examples Are Not Bugs, They Are Features
Adversarial Examples Are Not Bugs, They Are Features
Yannic Kilcher
 21 I'm at ICML19 :)
I'm at ICML19 :)
Yannic Kilcher
 22 Population-Based Search and Open-Ended Algorithms
Population-Based Search and Open-Ended Algorithms
Yannic Kilcher
 23 XLNet: Generalized Autoregressive Pretraining for Language Understanding
XLNet: Generalized Autoregressive Pretraining for Language Understanding
Yannic Kilcher
 24 Conversation about Population-Based Methods (Re-upload)
Conversation about Population-Based Methods (Re-upload)
Yannic Kilcher
 25 Reconciling modern machine learning and the bias-variance trade-off
Reconciling modern machine learning and the bias-variance trade-off
Yannic Kilcher
 26 Learning World Graphs to Accelerate Hierarchical Reinforcement Learning
Learning World Graphs to Accelerate Hierarchical Reinforcement Learning
Yannic Kilcher
 27 Manifold Mixup: Better Representations by Interpolating Hidden States
Manifold Mixup: Better Representations by Interpolating Hidden States
Yannic Kilcher
 28 Processing Megapixel Images with Deep Attention-Sampling Models
Processing Megapixel Images with Deep Attention-Sampling Models
Yannic Kilcher
 29 Gauge Equivariant Convolutional Networks and the Icosahedral CNN
Gauge Equivariant Convolutional Networks and the Icosahedral CNN
Yannic Kilcher
 30 Auditing Radicalization Pathways on YouTube
Auditing Radicalization Pathways on YouTube
Yannic Kilcher
 31 RoBERTa: A Robustly Optimized BERT Pretraining Approach
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Yannic Kilcher
 32 Dynamic Routing Between Capsules
Dynamic Routing Between Capsules
Yannic Kilcher
 33 DEEP LEARNING MEME REVIEW - Episode 1
DEEP LEARNING MEME REVIEW - Episode 1
Yannic Kilcher
 34 Accelerating Deep Learning by Focusing on the Biggest Losers
Accelerating Deep Learning by Focusing on the Biggest Losers
Yannic Kilcher
 35 [News] The Siraj Raval Controversy
[News] The Siraj Raval Controversy
Yannic Kilcher
 36 LeDeepChef 👨‍🍳 Deep Reinforcement Learning Agent for Families of Text-Based Games
LeDeepChef 👨‍🍳 Deep Reinforcement Learning Agent for Families of Text-Based Games
Yannic Kilcher
 37 The Visual Task Adaptation Benchmark
The Visual Task Adaptation Benchmark
Yannic Kilcher
 38 IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
Yannic Kilcher
 39 AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
Yannic Kilcher
 40 SinGAN: Learning a Generative Model from a Single Natural Image
SinGAN: Learning a Generative Model from a Single Natural Image
Yannic Kilcher
 41 A neurally plausible model learns successor representations in partially observable environments
A neurally plausible model learns successor representations in partially observable environments
Yannic Kilcher
 42 MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
Yannic Kilcher
 43 Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
Yannic Kilcher
 44 NeurIPS 19 Poster Session
NeurIPS 19 Poster Session
Yannic Kilcher
 45 Go-Explore: a New Approach for Hard-Exploration Problems
Go-Explore: a New Approach for Hard-Exploration Problems
Yannic Kilcher
 46 Reformer: The Efficient Transformer
Reformer: The Efficient Transformer
Yannic Kilcher
 47 [Interview] Mark Ledwich - Algorithmic Extremism: Examining YouTube's Rabbit Hole of Radicalization
[Interview] Mark Ledwich - Algorithmic Extremism: Examining YouTube's Rabbit Hole of Radicalization
Yannic Kilcher
 48 Turing-NLG, DeepSpeed and the ZeRO optimizer
Turing-NLG, DeepSpeed and the ZeRO optimizer
Yannic Kilcher
 49 Growing Neural Cellular Automata
Growing Neural Cellular Automata
Yannic Kilcher
 50 NeurIPS 2020 Changes to Paper Submission Process
NeurIPS 2020 Changes to Paper Submission Process
Yannic Kilcher
 51 Deep Learning for Symbolic Mathematics
Deep Learning for Symbolic Mathematics
Yannic Kilcher
 52 Online Education - How I Make My Videos
Online Education - How I Make My Videos
Yannic Kilcher
 53 [Rant] coronavirus
[Rant] coronavirus
Yannic Kilcher
 54 Axial Attention & MetNet: A Neural Weather Model for Precipitation Forecasting
Axial Attention & MetNet: A Neural Weather Model for Precipitation Forecasting
Yannic Kilcher
 55 Agent57: Outperforming the Atari Human Benchmark
Agent57: Outperforming the Atari Human Benchmark
Yannic Kilcher
 56 State-of-Art-Reviewing: A Radical Proposal to Improve Scientific Publication
State-of-Art-Reviewing: A Radical Proposal to Improve Scientific Publication
Yannic Kilcher
 57 Dream to Control: Learning Behaviors by Latent Imagination
Dream to Control: Learning Behaviors by Latent Imagination
Yannic Kilcher
 58 POET: Endlessly Generating Increasingly Complex and Diverse Learning Environments and Solutions
POET: Endlessly Generating Increasingly Complex and Diverse Learning Environments and Solutions
Yannic Kilcher
 59 Evaluating NLP Models via Contrast Sets
Evaluating NLP Models via Contrast Sets
Yannic Kilcher
 60 [Drama] Who invented Contrast Sets?
[Drama] Who invented Contrast Sets?
Yannic Kilcher

Related AI Lessons

⚡
SubDownload — plug YouTube into Claude, Cursor, and ChatGPT via MCP
Turn YouTube videos into searchable text with SubDownload, integrating with Claude, Cursor, and ChatGPT via MCP
Dev.to AI
 ⚡
I Built a Neural Memory Layer for a Voice AI Assistant: Embeddings + Vector Search + Activity Neurons
Learn how to build a neural memory layer for a voice AI assistant using embeddings, vector search, and activity neurons to enable persistence in conversations
Dev.to AI
 ⚡
Rate Limiting in LLM Applications: Why You Need It and How to Build It
Implement rate limiting in LLM applications by counting tokens, not requests, to prevent abuse and optimize performance
Dev.to · Pranay Batta
 ⚡
I Tried 10 ChatGPT Resume Prompts. Here's What Actually Got Me Interviews.
Learn how to use ChatGPT prompts to improve your resume and get more interview callbacks
Dev.to AI

Chapters (13)

Intro & Overview
1:50 Quadratic Memory in Full Attention
4:55 Architecture Overview
6:35 Random Attention
10:10 Window Attention
13:45 Global Attention
15:40 Architecture Summary
17:10 Theoretical Result
22:00 Experimental Parameters
25:35 Structured Block Computations
29:30 Recap
31:50 Experimental Results
34:05 Conclusion
Up next
5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems
Dave Ebbelaar (LLM Eng)
Watch →