NNFM with varying training size

Qing Lan from the National Institutes of Health created a dataset to test neural network performance on the forward problem of diffuse reflectance spectroscopy. The dataset records total normalized mean absolute errors across all source detector distances. For each training data percentage from 5% to 95%, the neural network was trained independently 20 times.

Use Cases

• Benchmarking neural network robustness based on varying training data percentages from 5% to 95%.
• Analyzing error trends in forward problem solutions based on total normalized mean absolute error metrics.
• Studying the relationship between training set size and model performance for diffuse reflectance spectroscopy.

Strengths

• Training data percentage was systematically varied from 5% to 95%, providing a controlled experimental range.
• Each training size condition involved 20 independent training runs, which suggests results are based on repeated trials.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

National Institutes of Health

Collection Method

Results from neural network training experiments on the forward problem of diffuse reflectance spectroscopy.