> 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.
Simulated Annealing (SA)
Single-agent trajectory with explicit temperature decay mirroring physical metallurgical annealing.
Gaussian Process Bayesian Optimization (GP-BO)
GP posterior provides mean prediction and calibrated epistemic uncertainty bounds at all unobserved points.
Tree-structured Parzen Estimators (TPE)
Directly displays which hyperparameter ranges yield top quantile performance via 1D kernel densities.
Hyperband & Successive Halving
Successive halving brackets clearly discard non-promising models early to focus budget on top performers.
NSGA-II (Non-dominated Sorting Genetic Algorithm II)
Directly yields the full Pareto frontier of optimal tradeoff solutions for stakeholder selection.
Ant Colony Optimization (ACO)
Pheromone trail distribution directly highlights high-confidence graph edges and route corridors.
Markov Chain Monte Carlo (MCMC / Metropolis-Hastings)
Generates full empirical posterior probability distributions and credible intervals for every parameter.
Hamiltonian Monte Carlo & No-U-Turn Sampler (HMC / NUTS)
NUTS automatically calibrates trajectory lengths to eliminate arbitrary tuning parameters.
Stochastic Variational Inference (SVI)
Transforms slow integration into fast stochastic optimization, scaling Bayesian inference to massive datasets.
Gaussian Process Regression & Classification (GPR / GPC)
Provides exact, analytical closed-form Gaussian predictive mean and variance at all input locations.
Bayesian Additive Regression Trees (BART)
Generates full posterior distributions for Individual Treatment Effects (ITE) and counterfactuals.
Kalman Filter (Linear, EKF & Unscented UKF)
Kalman Gain explicitly quantifies how measurement surprise balances against prior model uncertainty.
Particle Filtering (Sequential Monte Carlo / SMC)
Particle cloud represents the exact multimodal posterior belief over the agent location in real-time.
Expectation-Maximization (EM)
E-step estimates soft responsibilities while M-step updates parameter estimators in closed analytical form.
Bayesian Structural Time Series (BSTS)
Spike-and-slab prior outputs explicit posterior inclusion probabilities for synthetic control predictors.
Copula Dependence Modeling (Clayton, Gumbel, Frank & Gaussian Copulas)
Decouples marginal feature distributions completely from the underlying dependency structure.
ARIMA & SARIMAX (Box-Jenkins)
Autoregressive (AR) and Moving Average (MA) polynomials directly quantify momentum and shock persistence.