What should you check in AI exercises?
English translation
AI Applications · Week 1
This week's goal · Explain the data fed to the model and the results obtained, and check whether those results are correct.
Understanding that remains useful when tools change
Implementation Frameworks and libraries keep changing. More useful than memorizing function names is understanding the problem, the inputs, the operations that transform them, and the resulting outputs. Learn concepts and design reasons from lecture videos, then check your understanding by implementing them with current tools in the exercises.
If familiar code does not run unchanged in a new environment, first describe its role. Check official documentation for equivalent functionality and input/output conditions. Test a few data samples against expected results before applying it throughout the code. This course also develops the ability to adapt to new tools using core concepts and verify results.
What problem will you solve, and what will you record?
Examples for distinguishing AI Inputs and Outputs include predicting electricity usage or classifying device status as normal or abnormal. Define the inputs and results to inspect, then keep code, outputs, and interpretations together. For example, “Computed the mean of 6 inputs and found 3 values at or above the criterion” is easier to revisit than “A plot appeared.”
1. Express the problem as inputs and outputs
Image classification takes an image and produces a category. Character prediction takes earlier characters and produces the next one. First describe the problem to solve in one sentence, then distinguish the data received by the model from its outputs.
Express the problem as inputs and outputs
Inputs and targets differ. During training, targets measure prediction error; in actual use, do not include targets in model inputs.
2. Keep execution results as evidence
A Notebook retains explanations, code, and outputs together. A good exercise record shows what was run, under which conditions, and what results were obtained. Alongside an Accuracy value, record the data used and what was evaluated.
Keep execution results as evidence
Fitting the training data well does not establish that a model also performs well on new data. This is why we will distinguish training from evaluation.
3. A weekly learning routine
Learn concepts through videos, use self-study material to clarify confusing points, then participate in exercises. Repeat prediction → execution → observation → explanation. Listen to the opening class instructions, then work with help from teaching assistants.
A weekly learning routine
When an error occurs, first inspect the error message, cell, and input shape rather than the entire screen. Explaining what you checked helps you ask a specific question.
Check your understanding
- Can you describe the inputs and results to inspect for a problem in your field of interest?
- Do you know the order in which to use videos, HTML, and practice files?
- Can you explain why an exercise record should include results and interpretation alongside code?