Optimizing Model Latency

Estimating ______ Latency is an important step in optimizing a model.

The goal of ______ is to reduce the computational resources required by a model.

Before optimizing a model, you should first make sure that ______ matters.

Contributing to ______ source projects can help to improve model efficiency.

The chapter also discusses ______ Cycling as a method to reduce energy usage.

To prevent momentary glitches from causing problems, we could take the average of all of our ______ outputs across a period of time.

Real-world ______ is often messy and machine learning models aren’t perfect.

We use this technique to help with wake-word detection in Chapter ______.

This would mean that transient issues are ______, and we only act upon consistent behavior.

Since real-world ______ is often messy and machine learning models aren’t perfect, it’s possible that a temporary glitch might result in an incorrect classification.

We could run our ______ on the current data window every 10 seconds, and take the averages of the last 6 outputs.

To make the most of limited data, a technique called ______ can be used to generate new, artificial datapoints.

Regularization techniques are used to make deep learning ______ less likely to overfit their training data.

One regularization technique involves ______ the connections between neurons during training.

Data augmentation is a way to artificially expand the size of a ______ dataset.

Regularization techniques generally involve ______ the model in some way to prevent it from perfectly memorizing the data.

The best way to beat overfitting, when possible, is to get your hands on a larger and more ______ dataset.