179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsHard· ANCE· Hard Negative Mining· Training Dynamics
Why does ANCE rebuild the ANN index during training ins...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe effectiveness of hard negative mining depends on the current state of the model parameters (θt). A negative sample that is hard for a current checkpoint will eventually become easy as the model pushes the margin up during training. If you use a statically mined set, the margin effectively decays, leading to a significant loss in gradient weight. ANCE periodically refreshes the index to maintain an optimal sampling distribution, acknowledging that this process is compute-intensive and must be performed asynchronously.