2022-1 Deep Learning and Applications
In this lecture, we will be learning about two different topics in deep learning: self-supervised learning (SSL) and generative models.
Syllabus
- Historical Review (
AlexNet,DQN,Attention,Adam,GAN,ResNet,Transformer,Pretrained Model,SSL) - Good Old Fashioned SSL (
Jigsaw,BiGAN,RotNet,Auto-Encoding Transform,DeepCluster,Single Image SSL) - Convnet-based SSL (
DrLIM,Contrastive Predictive Coding,SimCLR,MoCo,BYOL,SimCLRv2,SwAV,Barlow Twins) - Transformer-based SSL (
Transformer,ViT,Swin Transformer,DINO,EsViT) - Language-domain SSL (
GPT,GPT-2,BERT,RoBERTa,ALBERT,GPT-3) - Generative Model 1 (
NADE,PixelRNN,PixelCNN) - Generative Model 2 (
VAE,WAE,GAN,PlanarFlow) - Generative Model 3 (
DDPM) - Generative Model 4 (
DDIM) - Generative Model 5 (
InfoGAN,VQ-VAE,VQ-VAE2) - Generative Model 6 (
ADM,CFG,GLIDE,DALL-E2)
Paper Lists
Jigsaw: "Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzles," 2017BiGAN: "ADVERSARIAL FEATURE LEARNING," 2017RotNet: "UNSUPERVISED REPRESENTATION LEARNING BY PREDICTING IMAGE ROTATIONS," 2018Auto-Encoding Transform: "AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations rather than Data," 2019DeepCluster: "Deep Clustering for Unsupervised Learning of Visual Features," 2019Single Image SSL: "A CRITICAL ANALYSIS OF SELF-SUPERVISION, WHAT WE CAN LEARN FROM A SINGLE IMAGE," 2020DrLIM: "Dimensionality Reduction by Learning an Invariant Mapping," 2006Contrastive Predictive Coding: "Representation Learning with Contrastive Predictive Coding," 2019SimCLR: "A Simple Framework for Contrastive Learning of Visual Representations," 2020MoCo: "Momentum Contrast for Unsupervised Visual Representation Learning," 2020BYOL: "Bootstrap Your Own Latent A New Approach to Self-Supervised Learning," 2020SimCLRv2: "Big Self-Supervised Models are Strong Semi-Supervised Learners," 2020SwAV: "Unsupervised Learning of Visual Features by Contrasting Cluster Assignments," 2021Barlow Twins: "Barlow Twins: Self-Supervised Learning via Redundancy Reduction," 2021Transformer: "Attention is All You Need," 2017ViT: "AN IMAGE IS WORTH 16 X 16 WORDS :TRANSFORMERS FOR IMAGE RECOGNITION AT SCALE," 2021Swin Transformer: "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows," 2021DINO: "Emerging Properties in Self-Supervised Vision Transformers," 2021EsViT: "Efficient Self-supervised Vision Transformers for Representation Learning," 2021GPT: "Improving Language Understanding by Generative Pre-Training," 2018GPT-2: "Language Models are Unsupervised Multitask Learners," 2018BERT: "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding," 2019RoBERTa: "RoBERTa: A Robustly Optimized BERT Pretraining Approach," 2019ALBERT: "ALBERT: A LITE BERT FOR SELF-SUPERVISED LEARNING OF LANGUAGE REPRESENTATIONS," 2020GPT-3: "Language Models are Few-Shot Learners," 2020NADE: "Neural Autoregressive Distribution Estimation." 2016PixelRNN: "Pixel Recurrent Neural Networks," 2016PixelCNN: "Conditional Image Generation with PixelCNN Decoders," 2016VAE: "Auto-Encoding Variational Bayes," 2013WAE: "Wasserstein Auto-Encoders," 2017GAN: "Generative Adversarial Networks," 2014PlanarFlow: "Variational Inference with Normalizing Flows," 2016DDPM: "Denoising Diffusion Probabilistic Models," 2020InfoGAN: "InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets," 2016VQ-VAE: "Neural Discrete Representation Learning," 2018VQ-VAE2: "Generating Diverse High-Fidelity Images with VQ-VAE-2," 2019DDIM: "DENOISING DIFFUSION IMPLICIT MODELS," 2020IDDPM: "Improved Denoising Diffusion Probabilistic Models," 2021ADM: "Diffusion Models Beat GANs on Image Synthesis," 2021CFG: "Classifier-Free Diffusion Guidance," 2021BART: "ImageBART: Bidirectional Context with Multinomial Diffusion for Autoregressive Image Synthesis," 2021DiffusionGAN: "TACKLING THE GENERATIVE LEARNING TRILEMMA WITH DENOISING DIFFUSION GANS," 2021GLIDE: "GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models," 2022DALL-E2: "Hierarchical Text-Conditional Image Generation with CLIP Latents," 2022