How would you augment a dataset to simulate one-shot scenarios while ensuring the embeddings remain consistent

Question

With the help of code can you tell me How would you augment a dataset to simulate one-shot scenarios while ensuring the embeddings remain consistent?

Answer 1

To simulate one-shot scenarios while keeping embeddings consistent, augment data using semantically similar paraphrasing or contrastive pairs that preserve embedding space proximity.

Here is the code snippet you can refer to:

In the above code we are using the following key points:

• Uses SynonymAug to retain semantic consistency.

• Applies SentenceTransformer to check embedding similarity.

• Maintains embedding proximity via cosine similarity check.

• Simulates multiple views of one-shot input for robustness.

Hence, this approach enables realistic one-shot data augmentation by generating semantically close variants while ensuring the embeddings stay consistent, ideal for training with limited samples.