semihard
- adjective
- /ˈsɛmiˌhɑrd/
- Specialized
- During training, the model sampled semihard negatives to help the embedding separate similar classes.
Examples
-
Researchers found that mining semihard examples improved convergence without introducing noisy false negatives in batches.
-
The triplet loss benefits when each batch contains at least one semihard pair alongside easy and hard pairs.
-
The semihard examples greatly improved the model's accuracy.
-
We need to include more semihard examples in our training set.
-
During training, the model sampled semihard negatives, making it easier to distinguish between positive and negative examples.
-
Researchers found that using semihard examples during model training improved accuracy without adding unnecessary noise.
-
Including semihard examples in each training batch significantly enhanced the model's ability to differentiate classes.
Antonyms
How Distinguishable
Surface Forms
Morphology
Etymology
The word semihard in machine learning uses the same parts semi- 'partly' and hard, and it describes an example that is not easy to separate from a correct one but not impossible either. So a semihard example is moderately difficult and helps the model learn to tell similar cases apart.