> ML_LITERATURE_ATLAS_v1.0
Research Literature Atlas
253 qualified literature records from foundational statistical learning to frontier reasoning LLMs: verified DOIs, arXiv IDs, and original bilingual syntheses.
Generative Adversarial Nets (GAN)
Historical NeurIPS paper introducing Generative Adversarial Networks, triggering the generative deep learning era by formulating synthesis as an adversarial duel.
Deep Unsupervised Learning using Nonequilibrium Thermodynamics
Visionary ICML paper introducing diffusion probabilistic models, defining data generation as reversing a thermodynamic diffusion process.
Denoising Diffusion Probabilistic Models (DDPM)
Landmark NeurIPS paper showing that denoising diffusion probabilistic models achieve high image synthesis fidelity matching and exceeding GANs.
Score-Based Generative Modeling through Stochastic Differential Equations
Theoretical tour de force unifying generative diffusion through stochastic differential equations, introducing Probability Flow ODEs and predictor-corrector samplers.
High-Resolution Image Synthesis with Latent Diffusion Models (Stable Diffusion)
Monumental CVPR paper presenting Latent Diffusion Models, slashing computational costs by training diffusion in compressed latent spaces while achieving state-of-the-art text-to-image synthesis.
Scalable Diffusion Models with Transformers (DiT)
Seminal ICCV paper introducing Diffusion Transformers (DiT), establishing that Vision Transformers scale predictable generative fidelity with compute.
Q-learning
Classic paper rigorously proving that off-policy temporal difference updates converge to the optimal action-value function in Markovian environments.
Human-level control through deep reinforcement learning (Nature DQN)
Seminal DeepMind Nature paper demonstrating that deep reinforcement learning can master complex video games with human-comparable skill directly from pixels.
Proximal Policy Optimization Algorithms (PPO)
Highly influential paper introducing PPO, replacing fragile policy gradient methods with a robust, easy-to-implement clipped surrogate loss.
Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
Widely adopted ICML paper introducing Soft Actor-Critic (SAC), achieving unmatched sample efficiency and stability in continuous action robotics.
Mastering the game of Go with deep neural networks and tree search (AlphaGo)
Historic Nature cover paper describing AlphaGo, the first computer program to defeat a human world champion in the 2,500-year-old game of Go.
Direct Preference Optimization: Your Language Model Is Secretly a Reward Model (DPO)
Breakthrough NeurIPS paper showing that RLHF can be solved via direct cross-entropy optimization without reinforcement learning or separate reward models.
Semi-Supervised Classification with Graph Convolutional Networks (GCN)
Foundational ICLR paper introducing Graph Convolutional Networks, bridging spectral graph theory and neural networks to launch modern graph ML.
Graph Attention Networks (GAT)
Influential ICLR paper introducing Graph Attention Networks, replacing uniform isotropic message passing with learned anisotropic self-attention.
Inductive Representation Learning on Large Graphs (GraphSAGE)
Seminal NeurIPS paper enabling inductive graph neural networks on massive streaming graphs by replacing full-graph operations with neighborhood sampling.
How Powerful are Graph Neural Networks? (Graph Isomorphism Network - GIN)
ICLR Oral paper theoretically bounding standard GNN expressiveness and introducing GIN, which matches the 1-Weisfeiler-Lehman graph isomorphism test.
Time Series Analysis: Forecasting and Control
Monumental 500-page textbook establishing the Box-Jenkins methodology and ARIMA modeling, which defined the field of time-series analysis for half a century.
Forecasting at Scale (Prophet)
Highly practical paper introducing Meta Prophet, enabling non-expert analysts to generate high-quality operational forecasts at scale.