## [Embedded Machine Learning for Dummies](https://knowen-production.s3.amazonaws.com/uploads/attachment/file/5252/embedded-machine-learning-design.pdf) fromARM

podcast [OctoML: Automated Deep Learning Engineering with Jason Knight and Luis Ceze](https://podcasts.apple.com/gb/podcast/software-engineering-daily/id1019576853?i=1000508309481) Software Engineering Daily

https://www.quora.com/What-does-the-terms-Top-1-and-Top-5-mean-in-the-context-of-Machine-Learning-research-papers-when-report-empirical-results

> The output of the network at test time is a score per class (typically 1000 classes for image net). The top 5 performance checks whether the right class label is among the 5 highest output scores. It's a laxer measure than top 1 (whether the top scoring class is the correct label). You can check the imagenet competition, I think this is explained there in more detail

## Google research 2021

[LEARNED HARDWARE/SOFTWARE CO-DESIGN OF NEURAL ACCELERATORS](https://arxiv.org/pdf/2010.02075.pdf#page10) Zhan Shi, Chirag Sakhuja

## https://tvm.apache.org/
> An End to End Machine Learning Compiler Framework for CPUs, GPUs and accelerators






## Cloudflare

[Cloudflare's Intelligent Design](https://diff.substack.com/p/cloudflares-intelligent-design) Byrne Hobart

https://stratechery.com/2021/cloudflare-on-the-edge/

https://blog.cloudflare.com/cloudflare-workers-serverless-week/