aihwkit.inference.compensation.drift module

Global drift compensation for inference.

class aihwkit.inference.compensation.drift.GlobalDriftCompensation[source]

Global drift compensation.

Uses a constant factor for compensating the drift.

get_readout_tensor(in_size)[source]

Return the read-out tensor.

Uses the set of one-hot vectors (eye).

readout(out_tensor)[source]

Read outs the mean abs.

class aihwkit.inference.compensation.drift.GlobalDriftCompensationWithExactReference[source]

Global drift compensation using an exact (ideal) reference readout.

Uses a constant factor for compensating the drift.

init_baseline(tile)[source]

Initialize the base line for applying the compensation.

Uses a all one tensor for read_out.

Parameters:: tile (InferenceTileWithPeriphery) – forward output of the read out vector to compensate
Returns:: reference tensor readout
Return type:: Tuple[Tensor, Tensor]

class aihwkit.inference.compensation.drift.PerColumnDriftCompensation[source]

Per column drift compensation. Uses a vector for compensating the drift.

get_readout_tensor(in_size)[source]

Return the read-out tensor. Uses the set of one-hot vectors (eye).

readout(out_tensor)[source]

Read outs the per-column mean abs.