AI Algorithms & Learning Methods Navigator

Which representation, learning method, or model mechanism should I investigate?

Choose the branch that matches your decision, then choose the relevant article. The destinations are alternatives and references, not a mandatory reading sequence.

Start: Choose Your Decision

• What information must the model represent? → Represent information
• What supervision and feedback are available? → Choose a learning regime
• Which structure must the model process? → Compare architectures
• Is the model failing to learn, or failing to generalize? → Train and regularize
• Does an action change the next state, and is feedback available? → Learn decisions through interaction
• Do you need probability, similarity, causality, or a mathematical limitation? → Investigate mathematical and statistical foundations
• Are you retrieving documents, entities, or similar observations? → Retrieve and compare knowledge

If you are unsure, begin with the first branch and state the outcome you need before selecting a method or tool.

Represent information

Question: What information must the model represent?

Words and documents: compare count-based, vector and topic representations. Graphs: inspect relationship embeddings. Many variables: examine dimensional reduction.

Article destinations:

• Continuous Bag-of-Words (CBoW)
• Deep Features
• Doc2Vec
• Excel
• Global Vectors for Word Representation (GloVe)
• Google Facets
• Isomap
• Latent
• Latent Dirichlet Allocation (LDA)
• Local Linear Embedding (LLE)
• Node2Vec
• Probabilistic Latent Semantic Analysis (PLSA)
• Recursive Feature Elimination (RFE)
• Singular Value Decomposition (SVD)
• Skip-Gram
• T-Distributed Stochastic Neighbor Embedding (t-SNE)
• Term Frequency–Inverse Document Frequency (TF-IDF)
• Term Frequency–Inverse Document Frequency (TF-IDF)
• Voice Vectors

Checkpoint: Explain what information the representation preserves and what it discards.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Choose a learning regime

Question: What supervision and feedback are available?

Labels: start with supervised learning. Few labels: examine semi-supervised, self-supervised and few-shot methods. Changing tasks: investigate transfer, online or continual learning.

Article destinations:

• (Stacked) Denoising Autoencoder (DAE)
• Biclustering
• Decentralized: Federated & Distributed
• Density-Based Spatial Clustering of Applications with Noise (DBSCAN)
• Ensemble Learning
• Expectation–Maximization (EM) Clustering using Gaussian Mixture Models (GMM)
• Few Shot Learning
• Fuzzy C-Means (FCM)
• Hierarchical — broad or mixed title; inspect the article context before proceeding.
• Hierarchical Cluster Analysis (HCA)
• Hierarchical Clustering; Agglomerative (HAC) & Divisive (HDC)
• K-Means
• K-Modes
• Mean-Shift Clustering
• Meta-Learning
• Multi-Task Learning (MTL)
• Neural Structured Learning (NSL)
• OPTICS: Ordering Points To Identify the Clustering Structure
• Online Learning
• Self-Organizing
• Semi-Supervised
• Sparse Autoencoder (SAE)

Checkpoint: Describe the feedback signal and a test that detects failed generalization.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Compare architectures

Question: Which structure must the model process?

Sequences need temporal or contextual processing; images need spatial processing; graphs need relational processing. Compare the mechanism before choosing a named architecture.

Article destinations: Narrow the choice: Which group matches your input or development environment?

• Other predictive and computing structures →
• Sequences, language and memory →
• Images and generative structures →

Other predictive and computing structures

• (Artificial) Immune System
• Apriori, Frequent Pattern (FP) Growth, Association Rules/Analysis
• Boosted Random Forest
• Boosting
• Conditional Adversarial Architecture (CAA)
• Decision Forest Regression
• Deep Belief Network (DBN)
• ELMo
• Elastic Net Regression
• Energy-based Model (EBN)
• Evolutionary Computation / Genetic Algorithms
• Fast Forest Quantile Regression
• Feed Forward Neural Network (FF or FFNN)
• General Regression Neural Network (GRNN)
• Hopfield Network (HN)
• Hypernetworks
• K-Nearest Neighbors (KNN)
• Lasso Regression
• LightGBM
• Markov Model (Chain, Discrete Time, Continuous Time, Hidden)
• Multiclassifiers; Ensembles and Hybrids; Bagging, Boosting, and Stacking
• NeuroEvolution of Augmenting Topologies (NEAT)
• Ordinal Regression
• Perceptron (P)
• Perspective
• Poisson Regression
• Reservoir Computing (RC) Architecture
• Restricted Boltzmann Machine (RBM)
• Ridge Regression
• SMART - Multi-Task Deep Neural Networks (MT-DNN)
• Spatial-Temporal Dynamic Network (STDN)
• Support Vector Regression (SVR)
• Symbolic Regression
• Topology and Weight Evolving Artificial Neural Network (TWEANN)
• Tree-based...
• ULMFiT
• Variational Autoencoder (VAE)
• XGBoost; eXtreme Gradient Boosted trees

Sequences, language and memory

• (Tree) Recursive Neural (Tensor) Network (RNTN)
• Average-Stochastic Gradient Descent (SGD) Weight-Dropped LSTM (AWD-LSTM)
• Bidirectional Encoder Representations from Transformers (BERT)
• Bidirectional Long Short-Term Memory (BI-LSTM)
• Bidirectional Long Short-Term Memory (BI-LSTM) with Attention Mechanism
• Gated Recurrent Unit (GRU)
• Hierarchical Temporal Memory (HTM)
• Long Short-Term Memory (LSTM)
• Manhattan LSTM (MaLSTM)
• Neural Turing Machine
• Sequence to Sequence (Seq2Seq)
• StructBERT
• TaBERT
• Transformer-XL

Images and generative structures

• Capsule Networks (CapNets)
• Context-Conditional Generative Adversarial Network (CC-GAN)
• Convolution vs. Cross-Correlation (Autocorrelation)
• Deconvolutional Neural Network (DN) / Inverse Graphics Network (IGN)
• Generative Facial Prior-Generative Adversarial Network (GFP-GAN)
• Generative Query Network (GQN)
• Kohonen Network (KN)/Self Organizing Maps (SOM)
• Pooling / Sub-sampling: Max, Mean
• ResNet-50
• Semi-Supervised Learning with Generative Adversarial Network (SSL-GAN)

Checkpoint: Identify the input structure, state or memory, and output of the architecture.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Train and regularize

Question: Is the model failing to learn, or failing to generalize?

Training failure: investigate initialization, gradients and objective functions. Good training results with poor held-out results: investigate regularization, stopping and data separation.

Article destinations:

• Cross-Entropy Loss
• Dropout
• Early Stopping
• L1 and L2 Regularization
• Loss
• Optimizer
• Parameter Initialization

Checkpoint: Compare the change with an unchanged baseline and held-out evaluation.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Learn decisions through interaction

Question: Does an action change the next state, and is feedback available?

Sequential feedback suggests reinforcement-learning methods; fixed objectives and constraints may instead require optimization. Examine exploration, observability and policy updates.

Article destinations:

• Advanced Actor Critic (A2C)
• Apprenticeship Learning - Inverse Reinforcement Learning (IRL)
• Asynchronous Advantage Actor Critic (A3C)
• Distributed Deep Reinforcement Learning (DDRL)
• Exploration
• Generative Tensorial Reinforcement Learning (GENTRL)
• Google DeepMind AlphaGo Zero
• Google DeepMind AlphaStar
• Hierarchical Reinforcement Learning (HRL)
• Inverse Reinforcement Learning (IRL)
• Lifelong Latent Actor-Critic (LILAC)
• Minigo
• Partial Observability
• Policy vs Plan
• Proximal Policy Optimization (PPO)
• Reinforcement Learning - Games, Self-driving Vehicles, Drones, Robotics, Management, Finance
• State-Action-Reward-State-Action (SARSA)
• Temporal Difference (TD) Learning
• Trust Region Policy Optimization (TRPO)

Checkpoint: Specify the state, action, reward, observations and stopping conditions.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Investigate mathematical and statistical foundations

Question: Do you need probability, similarity, causality, or a mathematical limitation?

Choose the relevant concept rather than treating mathematical terms as interchangeable models. Distinguish prediction, association and causal claims.

Article destinations:

• Activation Functions
• Autocorrelation
• Blockchain
• Dot Product
• Finding Paul Revere
• Forward-Forward
• Gödel’s Incompleteness Theorems
• Kernel Trick
• Manifold Hypothesis
• Math for Intelligence
• Metric Learning and Kernel Learning
• Monte Carlo
• Stochastic — broad or mixed title; inspect the article context before proceeding.

Checkpoint: State the assumptions required by the mathematical argument.

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Retrieve and compare knowledge

Question: Are you retrieving documents, entities, or similar observations?

Compare lexical, semantic and relational retrieval. Evaluate retrieval separately from the answer generated from retrieved material.

Article destinations:

• Approximate Nearest Neighbor (ANN)
• Document Similarity
• Hierarchical Navigable Small World (HNSW)
• Inverted File Indexes (IVF)
• Papers Search

Checkpoint: Can the system retrieve a known relevant item and reject a misleading one?

• Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
• Not yet: Revisit the relevant explanation or choose a simpler baseline.
• Choose another decision →

Continue Through PRIMO.ai

• How AI Works
• AI Solver
• Data & Knowledge Navigator

ConceptChains · Primo.ai Pages