How biological synapses self-assemble gradient learning.

Self-assembling motifs (SAM) can form hierarchical networks that approximate stochastic gradient descent, matching backpropagation-level performance in biological systems.

• Why it matters: Understanding how biological circuits are assembled is crucial for explaining brain learning mechanisms, addressing a gap left by existing models that focus only on learning processes.
• What they did: The study introduces SAM, which self-assemble from random connectivity via heterosynaptic plasticity rules, and demonstrates that networks of SAMs can organize into dynamics that emulate gradient descent.
• The result: This work suggests that biological learning systems may emerge from local interactions without being explicitly prescribed, expanding the understanding of how complex neural circuits could form.