hyperparameters
Learning rate, epochs
Hyperparameters are settings in machine learning that govern how a model learns, as opposed to parameters the model learns on its own.
Learning rate, batch size, and network architecture are hyperparameters—chosen before training begins and adjusted through experimentation rather than computed from data: We tuned the hyperparameters to reduce overfitting.
Distinct historical stretches
- dropoutOne who quits early
- momentumProduct of mass, velocity
- regularizationPenalty preventing model complexity
- cross-validationResampling method for model assessment
sweep · configuration · scheduling · cycling · adaptive · auto-tuning · fine-tuning
Excess adaptation to samples
- underfittingModel too simple
- generalizationDeriving principles from examples
- holdoutOne refusing to participate
- stratifiedArranged in stacked layers
convergence · plateau · benchmark · baseline · comparison · reproducibility
Degree of correctness
- precisionAccuracy of measurement
- recallMemory retrieval act
- AUCIntegrated area beneath plotted line
- objectivePurpose or goal
cost · threshold · sensitivity · specificity · perplexity
epochs
Distinct historical stretches
- dropoutOne who quits early
- momentumProduct of mass, velocity
- regularizationPenalty preventing model complexity
cross-validation · sweep · configuration · scheduling · cycling · adaptive · auto-tuning · fine-tuning
overfitting
Excess adaptation to samples
- underfittingModel too simple
- generalizationDeriving principles from examples
- holdoutOne refusing to participate
stratified · convergence · plateau · benchmark · baseline · comparison · reproducibility
accuracy
Degree of correctness
objective · cost · threshold · sensitivity · specificity · perplexity
layers
Stacked levels or coatings
width · pooling · dilation · normalization · encoder-decoder
lambda
Eleventh Greek letter
L1 · ridge · lasso · sparsity · constraint · smoothing · clipping · pruning
hyperparametersparameters
Defined boundaries or constraints
weights · coefficients · embeddings · biases · activations
hyperparameterspredictions
Forecasts of future events
outputs · inferences · training data · raw data · features