ISSLoss — Irrelevance Squared Sum Loss¶
Measures irrelevance (irrelevance squared sum) of the prediction residual relative to the gnostic concept center — how far predictions deviate from the learned manifold.
Overview¶
The conceptual form is L = mean(hj²) with hj = sinh(2θ) and θ = (y_pred - y_true) / S. This numerically stable surrogate avoids singularities while emphasizing large directional residuals. The centered coordinate is θ = (y_pred - y_true) / S.
Value range: [0, ∞).
Key Features¶
- Surrogate for
mean(hj²) - Custom autograd kernel (
ISSLossFunction) - Strong gradient for off-center predictions
Parameters¶
| Parameter | Type | Default | Description |
|---|---|---|---|
S |
float | str |
1 |
Scale parameter; "auto" for batch-local scale |
name |
str | None |
None |
Optional display name |
verbose |
bool |
False |
Verbose logging |
Attributes¶
S: Scale parameterS_local: Resolved local scale
Methods¶
forward(y_pred, y_true)¶
Return batch inverse-style residual surrogate (mean(sinh²(2θ))).
Example Usage¶
from machinegnostics.magnet import ISSLoss, Tensor
import numpy as np
y_pred = Tensor(np.zeros((2, 1)), requires_grad=True)
y_true = Tensor(np.zeros((2, 1)))
print(float(ISSLoss()(y_pred, y_true))) # 0.0
Author: Nirmal Parmar