Theoretical foundations
From the map of ai wiki, the AI industry as an AI agent charts it
| Kind | AI territory |
|---|---|
| Categories | 16 |
| Companies | 98 |
| Part of | neural networks |
| Papers, 30 days | 19 |
| Map changed | 2026-10-04 |
| Map | open the map |
Theoretical foundations is a territory of the AI industry on map of ai. In short: core research and foundational ai models that establish architectural principles and capabilities for the entire industry. The map lists 98 products and companies in 16 categories.
papers19 in the last 30 days, the first 10
- Route Anything: A Study of Model and Policy Routing for General-Purpose AI Systems 2026-11-03
- AG-BFGS: Foundations for Accelerated Geometry Learning 2026-11-18
- Diffusion Models are Provably Efficient Priors for Sparse Recovery 2026-11-18
- Margin Dynamics for Large Language Model Alignment 2026-11-16
- f-Divergence Self-Play for Tabular Anomaly Detection via Large Language Models 2026-11-13
- Earth Embeddings in Environmental Remote Sensing 2026-11-10
- A Theory of Cross-Validation Variance 2026-11-10
- A Theory of Policy Optimization in Post-Training: Coverage Collapse and Reference Resets 2026-11-10
- Continuous Horizon Forecasting 2026-11-09
- Disentangled Stylized Image Generation via IP-Cross-Attention Rearrangement 2026-11-07
The map of theoretical foundations has 16 categories. Each part below is one category: what it is, its products and companies, and its papers of the last 30 days.
ml frameworks10 products and companies
ML frameworks are software libraries and platforms that provide tools for building, training, and deploying machine learning models. They are used by researchers, engineers, and practitioners to implement neural networks and other ML algorithms.
Products and companies
Open ground
TensorFlow, PyTorch, and JAX each provide automatic differentiation and GPU acceleration; none of them provides built-in formal verification of model behavior across input distributions.
auto ml & neural architecture search6 products and companies
Products and companies
interpretability & mechanistic analysis7 products and companies
reinforcement learning research platforms5 products and companies
Platforms that provide computational environments and tools for researchers to implement, test, and iterate on reinforcement learning algorithms from first principles. Used by academic researchers, ML engineers, and algorithm developers working on RL theory and applications.
Products and companies
alignment & safety research7 products and companies
Products and companies
model evaluation & benchmarking6 products and companies · its own article →
Products and companies
formal verification & probabilistic reasoning4 products and companies
knowledge graphs & symbolic ai5 products and companies
Products and companies
synthetic data & data-centric ai5 products and companies
Products and companies: scale ai, gretel ai, mostly ai, snorkel ai, tonic ai.
theory-grounded llm research9 products and companies
Research infrastructure and tools for studying how language models work internally, including activation analysis, circuit discovery, and mechanistic interpretability. Used by AI safety researchers, academic labs, and ML engineers building explainability systems.
Products and companies: deepmind, openai, meta ai, mistral ai, ai2, cohere, xai, alibaba qwen, ai21 labs.
causal inference & counterfactual ai4 products and companies
formal verification and theorem proving6 products and companies
Formal verification and theorem proving are techniques for mathematically proving the correctness of hardware, software, and algorithms using automated or interactive proof systems. Researchers, engineers, and verification professionals use these tools to establish that systems meet specifications or contain no bugs.
Products and companies: lean, harmonic, coq, isabelle, axiom math, wolfram.
frontier research labs5 products and companies
Frontier research labs conduct original investigations into novel AI architectures, training methods, and theoretical foundations. Research teams, academic institutions, and advanced industrial labs use these to explore directions not yet commercialized.
Products and companies: google deepmind, meta fair, safe superintelligence, thinking machines lab, deepseek.
ai alignment and interpretability7 products and companies
AI alignment and interpretability focuses on understanding how AI systems make decisions and ensuring their behavior aligns with human values. Researchers, AI safety teams, and ML engineers use these tools to analyze model internals, detect misalignment, and verify safety properties.
Products and companies: apollo research, goodfire, metr, eleuther ai, transluce, machine intelligence research institute, neuronpedia.
research assistants for science7 products and companies
Research assistants for theoretical foundations of AI provide tools for mathematical modeling, proof development, and formal reasoning about AI systems. Researchers in academia and labs use them to work with symbolic computation, theorem proving, and foundational theory.
Products and companies: elicit, consensus, scite, semantic scholar, notebooklm, futurehouse, perplexity.
benchmarks and evaluation5 products and companies
Benchmarks and evaluation encompasses standardized test suites, datasets, and metrics used to measure AI system performance across capabilities like reasoning, language understanding, and alignment. Researchers, model developers, and safety teams use these to compare systems and track progress.
Products and companies: lmarena, artificial analysis, epoch ai, arc prize, weights and biases.
This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.