A paper atlas for classic AI research
Read the papers that shaped modern AI.
AI Blazers is a free, open reading atlas for classic AI papers. Each paper is explained through background, core ideas, terminology, formulas, experiments, historical context, and influence.
Reading routes
Start from a path, not a random list.
Papers are organized as learning routes, so students can follow the evolution of ideas across time.
Road to Large Language Models
From Seq2Seq and Attention to Transformer, BERT, GPT, scaling laws, instruction tuning, and RLHF.
Seq2SeqAttentionTransformerBERTGPT-1Scaling Laws +4
From Language Models to Agents
How language models connect with retrieval, reasoning, acting, decision making, and embodied environments.
Retrieval-Augmented GenerationReActDecision TransformerGeneralist AgentRT-2
Efficient Adaptation
Parameter-efficient tuning and compression methods for adapting large models with fewer trainable parameters.
DistillationAdapterPrompt TuningLoRA
From CNNs to Vision Transformers
The visual recognition path from ImageNet and AlexNet to ResNet, detection, segmentation, and ViT.
Dataset / BenchmarkAlexNetObject DetectionSegmentationResNetVision Transformer
Self-Supervised Representation Learning
Learning useful representations without dense human labels, from CPC and SimCLR to DINO and MAE.
CPCSimCLRDINOMAE
Generative Models
The evolution of modern generation: VAE, GAN, diffusion models, and latent diffusion.
VAEGANDiffusionLatent Diffusion
Vision-Language Models
Connecting images and language through contrastive learning, large-scale alignment, and multimodal generation.
CLIPALIGNDALL·EFlamingo
Reinforcement Learning and World Models
From classic Q-learning to deep RL, latent world models, and sequence modeling for decision making.
Q-learningDQNWorld ModelsDreamerDecision Transformer
People & Labs
Follow research trails.
Some ideas are best understood as a sequence of works from a researcher, lab, or team.
Kaiming He Trail
From initialization and ResNet to Mask R-CNN, MoCo, and MAE.
CVBackbonesDetectionSSL
Fei-Fei Li Trail
From ImageNet to visual understanding, scene graphs, and embodied intelligence.
DatasetsVision-LanguageScene GraphsEmbodied AI
DeepSeek Trail
Open-source LLMs, code models, MoE, mathematical reasoning, RL reasoning, and formal proof.
LLMMoEReasoningFormal Math
ByteDance Seed Trail
Speech, reasoning LLMs, VLMs, video generation, world models, and diffusion language models.
SpeechVLMVideoWorld Models
| Year | Topic | Subtopic | Paper | Status |
|---|---|---|---|---|
| 2014 | NLP | Seq2Seq | Sequence to Sequence Learning with Neural Networks | Read now |
| 2014 | NLP | Attention | Neural Machine Translation by Jointly Learning to Align and Translate | Read now |
| 2017 | NLP | Transformer | Attention Is All You Need | Read now |
| 2018 | NLP | BERT | BERT: Pre-training of Deep Bidirectional Transformers | Read now |
| 2018 | NLP | GPT-1 | Improving Language Understanding by Generative Pre-Training | Read now |
| 2020 | LLM | Scaling Laws | Scaling Laws for Neural Language Models | Coming soon |
| 2020 | LLM | GPT-3 | Language Models are Few-Shot Learners | Coming soon |
| 2021 | LLM | Instruction Tuning | Finetuned Language Models Are Zero-Shot Learners | Coming soon |
| 2022 | LLM | RLHF | Training Language Models to Follow Instructions with Human Feedback | Coming soon |
| 2025 | LLM | RL for NLU | Improving the Language Understanding Capabilities of Large Language Models Using Reinforcement Learning | Read now |
| 2022 | Agent | ReAct | ReAct: Synergizing Reasoning and Acting in Language Models | Coming soon |
| 2020 | RAG | Retrieval-Augmented Generation | Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | Coming soon |
| 2019 | Efficiency | Adapter | Parameter-Efficient Transfer Learning for NLP | Coming soon |
| 2021 | Efficiency | Prompt Tuning | The Power of Scale for Parameter-Efficient Prompt Tuning | Coming soon |
| 2021 | Efficiency | LoRA | LoRA: Low-Rank Adaptation of Large Language Models | Coming soon |
| 2012 | CV | AlexNet | ImageNet Classification with Deep CNNs | Coming soon |
| 2015 | CV | ResNet | Deep Residual Learning for Image Recognition | Coming soon |
| 2014 | CV | Object Detection | R-CNN | Coming soon |
| 2015 | CV | Segmentation | Fully Convolutional Networks | Coming soon |
| 2020 | CV | Vision Transformer | An Image is Worth 16x16 Words | Coming soon |
| 2018 | SSL | CPC | Representation Learning with Contrastive Predictive Coding | Coming soon |
| 2020 | SSL | SimCLR | A Simple Framework for Contrastive Learning | Coming soon |
| 2021 | SSL | DINO | Emerging Properties in Self-Supervised ViT | Coming soon |
| 2021 | SSL | MAE | Masked Autoencoders Are Scalable Vision Learners | Coming soon |
| 2013 | Gen | VAE | Auto-Encoding Variational Bayes | Coming soon |
| 2014 | Gen | GAN | Generative Adversarial Nets | Read now |
| 2020 | Gen | Diffusion | Denoising Diffusion Probabilistic Models | Coming soon |
| 2022 | Gen | Latent Diffusion | High-Resolution Image Synthesis with LDM | Coming soon |
| 2021 | Multimodal | CLIP | Learning Transferable Visual Models From Natural Language Supervision | Coming soon |
| 2021 | Multimodal | ALIGN | Scaling Up Vision-Language Representation Learning | Coming soon |
| 2021 | Multimodal | DALL·E | Zero-Shot Text-to-Image Generation | Coming soon |
| 2022 | Multimodal | Flamingo | Flamingo: a Visual Language Model | Coming soon |
| 1989 | RL | Q-learning | Learning from Delayed Rewards | Coming soon |
| 2013 | RL | DQN | Playing Atari with Deep Reinforcement Learning | Coming soon |
| 2016 | GNN | GCN | Semi-Supervised Classification with GCN | Coming soon |
| 2018 | GNN | GAT | Graph Attention Networks | Coming soon |
| 2015 | Compression | Distillation | Distilling the Knowledge in a Neural Network | Coming soon |
| 2018 | World Model | World Models | World Models | Coming soon |
| 2020 | World Model | Dreamer | Dreamer: Reinforcement Learning with Latent World Models | Coming soon |
| 2021 | Decision | Decision Transformer | Decision Transformer | Coming soon |
| 2022 | Agent | Generalist Agent | A Generalist Agent | Coming soon |
| 2023 | Embodied AI | RT-2 | Vision-Language-Action Models | Coming soon |
| 2017 | Continual Learning | EWC | Overcoming Catastrophic Forgetting | Coming soon |
| 2017 | Continual Learning | GEM | Gradient Episodic Memory | Coming soon |
| 2020 | Continual Learning | Survey | Continual Learning: A Comparative Study | Coming soon |
| 2015 | CV | Initialization | Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification | Coming soon |
| 2016 | CV | CNN / Backbone | Identity Mappings in Deep Residual Networks | Coming soon |
| 2015 | CV | Object Detection | Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks | Coming soon |
| 2017 | CV | Object Detection | Feature Pyramid Networks for Object Detection | Coming soon |
| 2017 | CV | Instance Segmentation | Mask R-CNN | Coming soon |
| 2018 | CV | Normalization | Group Normalization | Coming soon |
| 2020 | CV | Self-Supervised Learning | Momentum Contrast for Unsupervised Visual Representation Learning | Coming soon |
| 2009 | CV | Dataset / Benchmark | ImageNet: A Large-Scale Hierarchical Image Database | Coming soon |
| 2015 | CV | Image Captioning | Deep Visual-Semantic Alignments for Generating Image Descriptions | Coming soon |
| 2017 | CV + NLP | Dense Visual Annotation | Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations | Coming soon |
| 2017 | CV | Scene Graph | Scene Graph Generation by Iterative Message Passing | Coming soon |
| 2018 | CV | Image Generation | Image Generation from Scene Graphs | Coming soon |
| 2022 | Embodied AI | Scene Understanding | Embodied Semantic Scene Graph Generation | Coming soon |
| 2024 | LLM | Base Model | DeepSeek LLM: Scaling Open-Source Language Models with Longtermism | Coming soon |
| 2024 | Code LLM | Code Pretraining | DeepSeek-Coder: When the Large Language Model Meets Programming | Coming soon |
| 2024 | LLM Architecture | MoE | DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models | Coming soon |
| 2024 | LLM Architecture | Efficient MoE / MLA | DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model | Coming soon |
| 2024 | Math LLM | Mathematical Reasoning | DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models | Coming soon |
| 2024 | LLM | Large MoE / Efficient Training | DeepSeek-V3 Technical Report | Coming soon |
| 2025 | Reasoning LLM | RL / Reasoning | DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | Read now |
| 2025 | Formal Math | Lean / Theorem Proving | DeepSeek-Prover-V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition | Coming soon |
| 2024 | Speech AI | TTS | Seed-TTS | Coming soon |
| 2025 | Reasoning LLM | RL / Thinking Model | Seed-Thinking-v1.5 | Coming soon |
| 2025 | Multimodal LLM | VLM | Seed1.5-VL Technical Report | Coming soon |
| 2025 | Video / World Model | Unlabeled Video Learning | VideoWorld | Coming soon |
| 2025 | Video Generation | General Video Model | Seedance 1.0 | Coming soon |
| 2025 | Video Generation | Human Animation | OmniHuman-1 | Coming soon |
| 2025 | Code LLM | Code Model | Seed-Coder | Coming soon |
| 2025 | LLM Architecture | Diffusion LM | Seed Diffusion Preview | Coming soon |
| 2025 | Omni-modal Training | Any-modality Training | VeOmni | Coming soon |
| 2026 | Video Generation | Audio-Video Model | Seedance 2.0 | Coming soon |
Timeline
A compact history of the atlas.
1989
RL Q-learning
2009
CV Dataset / Benchmark
2012
CV AlexNet
2013
Gen VAE
2013
RL DQN
2014
NLP Seq2Seq
2014
NLP Attention
2014
CV Object Detection
2014
Gen GAN
2015
CV ResNet
2015
CV Segmentation
2015
Compression Distillation
2015
CV Initialization
2015
CV Object Detection
2015
CV Image Captioning
2016
GNN GCN
2016
CV CNN / Backbone
2017
NLP Transformer
2017
Continual Learning EWC
2017
Continual Learning GEM
2017
CV Object Detection
2017
CV Instance Segmentation
2017
CV + NLP Dense Visual Annotation
2017
CV Scene Graph
2018
NLP BERT
2018
NLP GPT-1
2018
SSL CPC
2018
GNN GAT
2018
World Model World Models
2018
CV Normalization
2018
CV Image Generation
2019
Efficiency Adapter
2020
LLM Scaling Laws
2020
LLM GPT-3
2020
RAG Retrieval-Augmented Generation
2020
CV Vision Transformer
2020
SSL SimCLR
2020
Gen Diffusion
2020
World Model Dreamer
2020
Continual Learning Survey
2020
CV Self-Supervised Learning
2021
LLM Instruction Tuning
2021
Efficiency Prompt Tuning
2021
Efficiency LoRA
2021
SSL DINO
2021
SSL MAE
2021
Multimodal CLIP
2021
Multimodal ALIGN
2021
Multimodal DALL·E
2021
Decision Decision Transformer
2022
LLM RLHF
2022
Agent ReAct
2022
Gen Latent Diffusion
2022
Multimodal Flamingo
2022
Agent Generalist Agent
2022
Embodied AI Scene Understanding
2023
Embodied AI RT-2
2024
LLM Base Model
2024
Code LLM Code Pretraining
2024
LLM Architecture MoE
2024
LLM Architecture Efficient MoE / MLA
2024
Math LLM Mathematical Reasoning
2024
LLM Large MoE / Efficient Training
2024
Speech AI TTS
2025
LLM RL for NLU
2025
Reasoning LLM RL / Reasoning
2025
Formal Math Lean / Theorem Proving
2025
Reasoning LLM RL / Thinking Model
2025
Multimodal LLM VLM
2025
Video / World Model Unlabeled Video Learning
2025
Video Generation General Video Model
2025
Video Generation Human Animation
2025
Code LLM Code Model
2025
LLM Architecture Diffusion LM
2025
Omni-modal Training Any-modality Training
2026
Video Generation Audio-Video Model