Barn Owls Know When to Wait
3 months ago
- #neuroscience
- #STDP
- #machine learning
- Noisy timing disrupts STDP (Spike-Timing-Dependent Plasticity), causing neurons to thrash.
- Uncertainty intervals in spike timing help neurons decide when to learn, freezing in noisy conditions.
- The barn owl analogy illustrates how uncertainty affects decision-making: clean signals lead to action, while noise leads to inaction.
- Traditional STDP fails under noisy conditions, leading to random weight changes ('thrashing').
- iuSTDP (interval-based STDP) introduces uncertainty bounds for spike times, improving stability.
- Two learning strategies in iuSTDP: conservative (learn only when certain) and probabilistic (scale learning by confidence).
- Confidence-based plasticity allows neurons to self-regulate learning based on timing reliability.
- A 'plasticity governor' adjusts learning rates dynamically, reducing updates under uncertainty.
- Simulations show iuSTDP with a governor outperforms vanilla STDP in noisy conditions.
- Future work includes integrating reward signals (e.g., dopamine) to further refine learning.