View larger version (24K):
[in this window]
[in a new window]
|
FIG. A2. Simulated data with known error floor. Each row corresponds to a different independent noise variance: A: 0.5, B: 2, C: 8. Left column: prediction errors for two-stage method with FA (green) and reduced GPFA (black), along with error floor (orange), at different state dimensionalities. Each green curve corresponds to a different kernel width (numbers of time steps are labeled). Star indicates minimum of black curve. Middle column: denser sampling of kernel widths for p = 3. Minimum of green curved denoted by green dot. Right column: each panel corresponds to an observed dimension. The same 2 observed dimensions are used in A, B, and C. Shown are the activity level of each neuron before noise was added (orange curves), noisy observations (orange dots), leave-neuron-out prediction using best two-stage method (green), and leave-neuron-out prediction using reduced GPFA (black).
|
