> ML_ALGORITHMS_ATLAS_v1.0
Algorithms & Method Families
100 qualified algorithmic method families across 9 disciplines: Classical Supervised, Unsupervised, Time Series, Anomaly Detection, Deep Learning, Recommenders, Causal Inference, Reinforcement Learning, and Specialized Methods.
Score-Based Generative Modeling through Stochastic Differential Equations
Predictor-corrector sampling combines numerical SDE integration with Langevin MCMC error correction.
MaskGIT (Masked Generative Image Transformer)
Non-autoregressive decoding progressively unmasks highest-confidence image tokens in constant iterations.
Consistency Models
Achieves high image synthesis quality in 1 or 2 forward passes without multi-step numerical integration.
Energy-Based Models (EBM)
Energy function directly acts as an unnormalized negative log-probability and out-of-distribution scoring metric.
Graph Convolutional Networks (GCN)
Node embeddings are explicit linear combinations of normalized immediate neighbor features.
Graph Attention Networks (GAT & GATv2)
Learned edge attention weights directly quantify which topological connections dominate node predictions.
GraphSAGE (Sample and Aggregate)
Generalizes inductively to completely unseen nodes and separate new graphs without retraining.
Graph Isomorphism Network (GIN)
Theoretically proven to match the maximum distinguishing power of the classical 1-WL graph isomorphism test.
Node2Vec (Biased Random Walks)
Parameters p (return) and q (in-out) provide continuous interpolation between micro-network community and macro-structural role similarity.
TransE (Translational Embeddings for Multi-Relational Data)
Geometric vector translation h + r = t enables transparent algebraic reasoning over facts.
RotatE (Knowledge Graph Embedding by Relational Rotation)
Rotational angles in the complex plane directly capture algebraic relational patterns without parameter bloat.
Relational Graph Convolutional Networks (R-GCN)
Basis sharing reveals structural similarities across different multi-relational edge types.
Temporal Graph Networks (TGN)
Maintains an up-to-date internal state memory vector for every node updated continuously upon each event.
Genetic Algorithms (GA)
Population fitness trajectories and allele selection frequencies can be visually tracked across generations.
Covariance Matrix Adaptation Evolution Strategy (CMA-ES)
Adapted covariance matrix directly recovers the inverse Hessian of the local objective landscape.
Differential Evolution (DE)
Direct vector arithmetic mutations make step sizes self-adapting to the spread of active solutions.
Particle Swarm Optimization (PSO)
Particle trajectory kinematics can be visualized as physical trajectories in feature space.