Workshop
Neuromatch NeuroAI Course 2027
About
A two-week immersive course on the common principles of natural and artificial intelligence. The core challenge of intelligence is generalisation, and neuroscience, cognitive science and AI are all searching for principles that support it. Major system features that affect generalisation include task structure (multitasking, multiple inputs with the same output and vice versa), microcircuitry (nonlinearities, canonical motifs and their operations, sparsity), macrocircuitry or architecture (modules for memory, information segregation, weight sharing by input symmetry or common development), learning rules (synaptic plasticity, modulation) and the data stream (curriculum). The course presents current understanding of how these issues arise in natural and artificial intelligence, comparing how these features affect representations, computations and learning, through case studies and coding exercises; students learn shared concepts across neuroscience, cognitive science and machine learning, ways of comparing artificial and biological neural networks, and multiple scales of computation and study. The course book covers foundations, architecture, learning and open mysteries. It targets an advanced audience who have taken the Neuromatch Deep Learning and Computational Neuroscience courses or equivalents, with intermediate Python, PyTorch for projects, linear algebra, probability, statistics and multivariable calculus. The course is synchronous and virtual, run in time slots for students around the world, at a full-time effort of 8 hours per day, 5 days per week, with code taught in Python through Google Colab or Kaggle. Students work in pods of about 15 with a dedicated teaching assistant and complete a collaborative research project with a project teaching assistant, presenting it at the end of the course. Applications open in February 2027.
Topics
Details
Organisations
Neuromatch
Host