✳flâneur — a map of the web's best reading

Label Smoothing Explained using Microsoft Excel

amaarora.github.io · 2,271 words · saved by 1 readers

In this blogpost, together, we: - Read and understand about Label Smoothing from Rethinking the Inception Architecture for Computer Vision research paper - Look at why we need Label Smoothing? - Re-implement Label Smoothing in Microsoft Excel step by step - Compare the results from our MS Excel implementation with Fastai/PyTorch versions of Label Smoothing Why are we using Microsoft Excel? It’s a valid question you might ask and I wasn’t a big fan of MS Excel either until I saw this video by Jeremy Howard about Cross Entropy Loss. In the video Jeremy explains Cross Entropy Loss using Microsoft Excel. It clicked and I understood it very well even with the fancy math in the cross entropy loss formula. And that is my hope here too! In this blogpost I hope that together we can see past the math and get the intuition for Label Smoothing and then later be able to implement it in a language/framework of our choice. So, let’s get started! Let’s consider we are faced with a multi-class image cl

Label Smoothing Explained using Microsoft Excel 1 Introduction In this blogpost, together, we: - Read and understand about Label Smoothing from Rethinking the Inception Architecture for Computer Vision research paper - Look at why we need Label Smoothing? - Re-implement Label Smoothing in Microsoft Excel step by step - Compare the results from our MS Excel implementation with Fastai / PyTorch versions of Label Smoothing Why are we using Microsoft Excel? It’s a valid question you might ask and I wasn’t a big fan of MS Excel either until I saw this video by Jeremy Howard about Cross Entropy Loss

Explore this link on the map →

saved by

related reading