> 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.
DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks
Pioneering Amazon paper showing that global neural network models trained across massive collections of time series outperform isolated per-series classical models.
Mamba: Linear-Time Sequence Modeling with Selective State Spaces
Groundbreaking paper introducing Mamba, proving that selective state spaces solve the attention scaling bottleneck and achieve 5x higher inference throughput.
Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
Foundational systems paper from NVIDIA establishing intra-node tensor model parallelism for training billion-parameter transformers.
ZeRO: Memory Optimizations Toward Training Trillion Parameter Models
Landmark Microsoft paper presenting DeepSpeed ZeRO, eliminating memory redundancies to train 100B+ models without complex tensor parallelism.
Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning
OSDI award-winning systems paper automating distributed model parallel execution plans across heterogeneous clusters.
Ray: A Distributed Execution Framework for Emerging AI Applications
Seminal OSDI paper introducing Ray, the universal compute fabric powering modern distributed training, hyperparameter tuning, and model serving.
TVM: An Automated End-to-End Optimizing Compiler for Deep Learning
Landmark OSDI compiler paper introducing Apache TVM, enabling automated hardware compilation and graph optimization for neural networks.
Triton: An Intermediate Language and Compiler for Tiled Neural Network Computations
Revolutionary PLDI compiler paper introducing OpenAI Triton, which powers FlashAttention, PyTorch 2.0 Inductor, and modern custom GPU kernels.
PyTorch: An Imperative Style, High-Performance Deep Learning Library
The defining systems paper for PyTorch, the undisputed global standard library powering academic research and modern production deep learning.
TensorFlow: A System for Large-Scale Machine Learning on Heterogeneous Distributed Systems
Seminal OSDI paper introducing TensorFlow, Google flagship framework that drove industrial-scale enterprise deployment of deep neural networks.
Spark: Cluster Computing with Working Sets
Classic HotCloud systems paper introducing Apache Spark, replacing MapReduce disk-bound processing with 100x faster in-memory big data computation.
Spark SQL: Relational Data Processing in Spark
SIGMOD paper presenting Spark SQL and DataFrames, which became the standard data-loading and ETL foundation for machine learning pipelines globally.
Intriguing properties of neural networks
Foundational safety paper uncovering adversarial vulnerabilities in deep neural networks, launching the field of machine learning security.
Explaining and Harnessing Adversarial Examples (FGSM)
Influential ICLR paper showing that adversarial examples stem from linear behavior in high-dimensional spaces, introducing fast adversarial training.
Differential Privacy
Historical theoretical breakthrough providing the rigorous mathematical standard for data privacy in machine learning, statistics, and census databases.
Deep Learning with Differential Privacy (DP-SGD)
Landmark ACM CCS paper proving that deep neural networks can be trained with formal mathematical privacy guarantees against training data extraction.
Equality of Opportunity in Supervised Learning
Seminal NeurIPS paper establishing non-discrimination criteria in supervised learning, moving algorithmic fairness from demographic parity to opportunity equality.
Datasheets for Datasets
Widely adopted standard paper advocating mandatory documentation for datasets to foster accountability and mitigate dataset bias in commercial AI.