New inductive biases in deep learning
Architectures inspired by cortical columns, modularity, and routing for matrices, tensors, graphs, and relational data.
AI Future Projects
Selected projects in reasoning, memory, and scalable intelligence.
A cross-section of ongoing directions in AI Future, spanning inductive bias, memory, compositionality, relational reasoning, multi-agent priors, and structural generalization.
Related pages
Memory architectures
Explicit memory systems for robust generalization, temporal integration, and flexible long-range interaction.
Compositional reasoning
Visual and language systems that uncover and manipulate compositional structure in multimodal settings.
Relational and causal reasoning
Learning systems that model graphical structure, relational dynamics, and causal patterns in complex data.
Indirection and abstraction
Mechanisms for analogy, abstract reference, symbolic manipulation, and stronger cross-domain generalization.
Multi-agent and theory-of-mind systems
Architectures for agents that model others, coordinate, and reason socially under uncertainty.