The 37th Algorithmic Learning Theory conference (ALT 2026) will be held in Toronto, Canada on February 23-26, 2026. The conference is dedicated to all theoretical and algorithmic aspects of machine learning. We invite submissions with contributions to new or existing learning problems including, but not limited to, the following list of topics.

    Design and analysis of learning algorithms.
    Classical foundations of learning theory: statistical, computational, algorithmic, and information-theoretic.
    Online learning and game theory.
    Optimization: convex, non-convex, new and old algorithms, their implicit biases, overparameterization, and so on.
    Different paradigms of learning: supervised, unsupervised, semi-supervised, active learning, reinforcement learning, and so on.
    All aspects of reinforcement learning: classical control-theoretic perspectives, modern uses such as LLM post-training, new algorithms, etc.
    Large language models, transformers, and all associated questions.
    Theoretical perspectives on trustworthy AI safety and AI safety: privacy, adaptive data analysis, fairness, alignment, and so on.
    Robustness: both classical perspectives (e.g. training data corruption), and modern perspectives (e.g. adversarial examples and LLM jailbreaks).
    Theoretical perspectives on deep learning: approximation, generalization, and optimization aspects of classical architectures such as shallow feedforward networks and simple RNNs, and modern architectures such as transformers.
    Core statistics topics: asymptotics, high-dimensional statistics, non-parametrics, causality, and so on.
    Learning with algebraic or combinatorial structure.
    Bayesian methods.
    Kernel methods. 
    Interpretability and explainability.
    Learning with algorithmic constraints: distributed learning, communication and memory efficient learning, federated learning, streaming algorithms, and so on.
    Different learning modalities: time series, sequence-to-sequence mappings, graph data, and so on.
    Mathematical analysis of sampling methods, including diffusion models and other practical methods.

Despite the theoretical focus of the conference, authors are welcome to support their analysis with relevant empirical results. Accepted papers will be presented at the conference as a full-length talk, and published electronically in the Proceedings of Machine Learning Research (PMLR); see details below and in the eventual submission instructions.