The mathematical foundations of trustworthy, explainable and hybrid AI. Researchers from logic, optimisation, probability theory and proof theory convene in Dubai to set the formal basis for next-generation intelligent systems.
Building strong, dependable artificial intelligence requires breakthrough mathematical solutions — particularly in the training, verification and interpretability of intelligent systems. MathAI is the venue where those solutions are presented, debated and formalised.
The conference unites leading experts in mathematics and theoretical programming around three guiding directions: Trustworthy AI, Explainable AI, and Hybrid AI. We invite breakthrough work grounded in mathematical logic, theory of algorithms, probability theory and statistics, optimisation and control, proof theory and deductive verification, differential equations, and game theory — applied to large language models, diffusion models, variational autoencoders and Bayesian neural networks.
MathAI 2027 marks the conference's first edition in Dubai, hosted under IAIC FZCO. The 2026 edition was held at the Sirius International Mathematics Center in Sochi, Russia.
Formal systems, model theory, computability. The substrate of provable behaviour in AI — from neuro-symbolic reasoning to verified inference.
Complexity, learnability, algorithmic information theory. Foundations for understanding what intelligent systems can — and cannot — efficiently compute.
Generalisation bounds, uncertainty quantification, Bayesian methods. The probabilistic backbone of modern machine learning.
Convex and non-convex optimisation, stochastic methods, control theory. The mathematics that turns architectures into trained models.
Deductive verification, type theory, formally verified neural-symbolic pipelines. AI that we can prove — not just test.
Continuous-time dynamics, neural ODEs, score-based generative models. Where calculus meets deep learning.
Multi-agent equilibria, mechanism design, interpretation of diffusion models, LLMs, VAEs and Bayesian networks through game-theoretic lenses.
Mathematical analysis of large language models, diffusion processes, variational autoencoders and Bayesian neural networks — emergence, alignment, scaling laws.
All deadlines are 23:59 London time, except where noted as Anywhere on Earth (AoE).
MathAI 2027 brings together the same international leadership that built the conference series — researchers from Russia, China, India, Serbia, the UAE and the US who define the mathematical state of the art for AI.
Participation is by invitation. Register through Enigma — the conference page for MathAI 2027 lists tracks, dates, committee, and the registration channel. Accepted authors are automatically enrolled.
Future editions will use Enigma — a blockchain-based registration platform with on-chain attendance certificates and verifiable peer review.
MathAI 2027 is organised by IAIC FZCO — the International AI Council based in Dubai — in partnership with the Sobolev Institute of Mathematics SB RAS, Novosibirsk State University, Innopolis, IAMM, ISP RAS and IIS SB RAS.
Conference operations are powered by the Enigma protocol — open-source blockchain infrastructure for scientific events: on-chain registration, verifiable attendance certificates, transparent peer review.
Learn more: iaic.world · enigma.ist