Reported in a candidate post, May 2025
Regularization, MLE/MAP, cross entropy and weight initialization
Filed under Meesho
The question
Questions on regularization and why L1 gives sparsity, optimal batch size, why we follow the negative gradient, MLE versus MAP, cross entropy and its link to KL divergence, equal weight initialization, dropout at training time, and multi-GPU training, PEFT and QLoRA.
No write-up yet. Related questions below have one.
Related Meesho questions
Collected from public candidate posts and coding forums. Updated Nov 2026.