# Andrew Ng - Reading Research Papers (Highlights) ![rw-book-cover|256](https://i.ytimg.com/vi/733m6qBH-jI/hqdefault.jpg) ## Metadata **Cover**:: https://i.ytimg.com/vi/733m6qBH-jI/hqdefault.jpg **Source**:: #from/readwise **Zettel**:: #zettel/fleeting **Status**:: #x **Authors**:: [[Andrew Ng]] **Full Title**:: Reading Research Papers **Category**:: #podcasts #readwise/podcasts **Category Icon**:: 🎙 **URL**:: [www.youtube.com](https://www.youtube.com/watch?v=733m6qBH-jI) **Host**:: [[www.youtube.com]] **Highlighted**:: [[2021-02-28]] **Created**:: [[2022-09-26]] ## Highlights - **Reading Research Papers** - [03:16](https://youtu.be/733m6qBH-jI?t=196) Compile list of papers (including articles) - [03:57](https://youtu.be/733m6qBH-jI?t=237) Skip around list - [04:41](https://youtu.be/733m6qBH-jI?t=281) First read 10% of each paper, then choose the ones worth the time to delve into. - [05:31](https://youtu.be/733m6qBH-jI?t=331) 5 ~ 20 papers are enough to get a good understanding of the topic. 50 ~ 100 to mastery. - [06:29](https://youtu.be/733m6qBH-jI?t=389) How to Read One Paper. - [07:07](https://youtu.be/733m6qBH-jI?t=427) The bad way is reading from the first word to the last word. - [08:17](https://youtu.be/733m6qBH-jI?t=497) Take multiple passes through the paper. - [08:27](https://youtu.be/733m6qBH-jI?t=507) Step 1: read the title, the abstract, the figures. - [09:11](https://youtu.be/733m6qBH-jI?t=551) Step 2: The intro, conclusions, all figures and skim the rest. - [10:26](https://youtu.be/733m6qBH-jI?t=626) Skim related work. - [11:06](https://youtu.be/733m6qBH-jI?t=666) Step 3: Read the paper but skip the math - [11:23](https://youtu.be/733m6qBH-jI?t=683) Step 4: Read whole things **but skip parts that don't make sense**. - [13:31](https://youtu.be/733m6qBH-jI?t=811) What did authors try to accomplish? - [13:58](https://youtu.be/733m6qBH-jI?t=838) What were the key elements of the approach? - [14:16](https://youtu.be/733m6qBH-jI?t=856) What can you use yourself? - [14:33](https://youtu.be/733m6qBH-jI?t=873) What other references do you want to follow? - [24:18](https://youtu.be/733m6qBH-jI?t=1458) Math - [24:34](https://youtu.be/733m6qBH-jI?t=1474) Re-derive it from scratch - [26:33](https://youtu.be/733m6qBH-jI?t=1593) Code: Run open-source code; Implement from scratch. - [28:03](https://youtu.be/733m6qBH-jI?t=1683) Learn something and space them out can help retention. - Compile list of papers (including articles) ([Time 0:03:16](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#196)) - Skip around list ([Time 0:03:57](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#237)) - First read 10% of each paper, then choose the ones worth the time to delve into. ([Time 0:04:41](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#281)) - 5 ~ 20 papers are enough to get a good understanding of the topic. 50 ~ 100 to mastery. ([Time 0:05:31](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#331)) ### How to Read One Paper. - The bad way is reading from the first word to the last word. ([Time 0:07:07](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#427)) - Take multiple passes through the paper. ([Time 0:08:17](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#497)) - Step 1: read the title, the abstract, the figures. ([Time 0:08:27](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#507)) - Step 2: The intro, conclusions, all figures and skim the rest. ([Time 0:09:11](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#551)) - Skim related work. ([Time 0:10:26](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#626)) - Step 3: Read the paper but skip the math ([Time 0:11:06](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#666)) - Step 4: Read whole things **but skip parts that don't make sense**. ([Time 0:11:23](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#683)) - What did authors try to accomplish? ([Time 0:13:31](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#811)) - What were the key elements of the approach? ([Time 0:13:58](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#838)) - What can you use yourself? ([Time 0:14:16](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#856)) - What other references do you want to follow? ([Time 0:14:33](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#873)) - Math ([Time 0:24:18](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#1458)) - Re-derive it from scratch ([Time 0:24:34](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#1474)) - Code: Run open-source code; Implement from scratch. ([Time 0:26:33](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#1593)) - Learn something and space them out can help retention. ([Time 0:28:03](https://reclipped.com/a/HP-jtHnZEeuTH0cOBrsLOw#1683))