This PR introduces the CGCAL and TCAL models to Topographica, based on the work done in my thesis:
CGCAL: Continuous GCAL. A model that applies all mechanisms on every step using a clocked model driven by the GeneratorSheet. This means Hebbian learning, homeostatic adaptation and a new hysteresis mechanism (relative to GCAL) are updated on every step. Designed to behave as closely to GCAL as possible.
TCAL: Temporally CALibrated GCAL. This model extends CGCAL to allow real temporal calibration by mapping topo.sim.time() to milliseconds. Features plausible PSTH profiles for the LGN and V1 units by tuning CGCAL appropriately (no new mechanisms). TCAL can develop good quality orientation maps as long as snapshot Hebbian learning is re-enabled at the sheet level (for possible reasons discussed in my thesis).
Currently I've only added CGCAL in this PR - I'll be adding TCAL very soon.
This PR introduces the CGCAL and TCAL models to Topographica, based on the work done in my thesis:
GeneratorSheet. This means Hebbian learning, homeostatic adaptation and a new hysteresis mechanism (relative to GCAL) are updated on every step. Designed to behave as closely to GCAL as possible.topo.sim.time()to milliseconds. Features plausible PSTH profiles for the LGN and V1 units by tuning CGCAL appropriately (no new mechanisms). TCAL can develop good quality orientation maps as long as snapshot Hebbian learning is re-enabled at the sheet level (for possible reasons discussed in my thesis).Currently I've only added CGCAL in this PR - I'll be adding TCAL very soon.