The Modern Mathematics of Deep Learning

Surprisal loss

Penguin learning project

The project uses 342 complete Palmer penguin records: 240 training-pool and 102 held-out rows. The selected training rows determine feature scaling and model parameters. Source CSV

This project trains weights with softmax, unlike the homepage lab, which uses nearest species averages. Same penguin measurements; different classification methods. Predictions can differ.

Classification predicts three species; Ridge predicts body mass in grams using the same observations. These are separate models. Training predictions feed the objective; 102 separate test predictions never update the weights.

Individual losses · Give one observed example a loss score.

Project predictions and known answers produce losses. Classification uses cross-entropy and α/2 times squared non-bias weights; Ridge uses normalized-mass MSE and α times all squared coefficients, including the intercept.

TaskTraining rowsTraining stepsPenalty αHeld-out row ID

Loss

Input: Training: 12 Prediction
Output: [≈0.005482, ≈1.012, ≈0.1422] … (12 entries) · Mean training loss: ≈0.228 (cross-entropy)

Project operation

The project calculation uses these records; the source formula and example follow below. Code edits run locally; project settings travel between pages.

Prediction → Loss → Mean training loss

Project input

Showing first three entries; the operation receives the complete array. Rounded preview; the operation uses full precision.

input.task · Project settings or constant: classification
input.predictions · Prediction: [Species: Adélie · Probabilities: [≈0.9945, ≈0.005412, ≈0.00005477], Species: Adélie · Probabilities: [≈0.6338, ≈0.3635, ≈0.002677], Species: Gentoo · Probabilities: [≈0.1023, ≈0.03028, ≈0.8674]] … (12 entries)
input.targets · Observations: [0, 1, 2] … (12 entries)

JavaScript

Result

[≈0.005482, ≈1.012, ≈0.1422] … (12 entries)

Used in these source sections: 1.2 Learning from examples

The surprisal of an observed value z is −log(f(z)), where f(z) is its positive density.

This loss appears in maximum likelihood estimation.

Source formula example

{\mathcal{L}}(f,z)=-\log(f(z))

illustration · Density values need not be probabilities at individual points; zero density gives infinite loss and is not used in this finite computation.

Symbols

$f$: A probability density.
$z$: An observed value.
$ℒ$: The surprisal loss.

Source formula example

This source example uses its own inputs and assumptions. Editing it does not change the project.

Edit a value or the code, then run the example. This is a finite demonstration, not a proof of the paper’s theorem.

This example’s data flow

Initial values in code

density = 0.5

Calculated here

Result expression: { density, loss: -Math.log(density) }

Returns

Example result

density: 0.5
loss: ≈0.6931

JavaScript

Read numbers from input and return the result. JavaScript basics.

Result

Enable JavaScript to run this example.

Technical statement

The numerical values are invented. Density values need not be probabilities at individual points; zero density gives infinite loss and is not used in this finite computation.

Source: Inline notation