How to Read Machine Learning Research Papers
Informational
How You Should Read Research Papers According To Andrew Ng (Stanford Deep Learning Lectures)
Instructions on how to approach knowledge acquisition through published enquiry papers by a recognized effigy within the earth of machine learning and education
Learn the nuts of AI and Deep Learning with TensorFlow and Keras in this Live Training Session hosted by Me.
"Wisdom is non a product of schooling but of the lifelong effort to learn it."
— Albert Einstein
Introduction
The ability to empathise information produced by the individuals at the cutting edge of research within Artificial Intelligence and the Motorcar learning domain is a skill that every serious automobile learning practitioner should acquire.
To stay relevant and increase your knowledge, machine learning practitioners demand to have an academic mindset and addiction. AI, ML and DL are evolving at a fast pace, and nosotros accept to equip ourselves with the knowledge to continue upwardly with the field, cognition that is but attainable through research papers.
This article wi l l provide you within instructions on how to get through a research newspaper effectively, and besides provide the following:
- A systematic approach to reading a collection of papers to proceeds cognition within a domain
- How to properly read a inquiry paper
- Useful online resources that can aid you lot in searching for papers and fundamental information
For those who would like to get to the central content inside this article, gyre downward to the section titled "Reading Enquiry Papers".
Get-go, Who is Andrew Ng?
The information I provide in this article was derived from a Stanford lecture taught by Andrew Ng. I've also supplemented the data contained in this article with personal tips and information from resources on the net.
Only first, a brief introduction on Andrew Ng.
Andrew Ng is probably the most known(and watched) motorcar learning teacher on the internet. He is likewise the co-founder of both Deeplearning.ai and Coursera.
Apart from his ongoing piece of work inside online education, he'due south also a professor at Stanford University.
More information on Andrew Ng is only a google search away.
It's natural for a person to selection up skills and habits demonstrated by individuals around their environment; this is why most PhD students will learn the skill of effectively digesting the content of a research paper appropriately. This is somewhat a fact, and Andrew mentions it very early on in the video referenced earlier.
Simply nosotros are not PhD students, well some might exist, only how do we normal individuals gain the required skills to read a enquiry paper and understand its content wholeheartedly.
Reading Research Papers
Specialization within the car learning domain is favourable if you are talented. For instance, having a generalist knowledge on the field of Reckoner vision is laudable, but having specialized noesis and expertise within primal techniques such as Pose interpretation volition exist more appealing to companies and organization looking for practitioners within that domain.
So let's use Pose Estimation as a guide to how we would approach reading research papers related to the bailiwick matter: pose estimation.
1. Assemble collections of resource that focus on the field of study matter. Resource can come up in the grade of inquiry papers, Medium manufactures, blog posts, videos, GitHub repository etc.
A quick google search on the phrase "pose estimation" will provide y'all with summit resources that contain data in regards to the subject matter. At this initial step, the aim is to collate all resources that are relevant, such as YouTube videos, implementation documentations and of grade research papers. Ideally, at this stage, there is no limit to the number of resources yous consider important, simply be sure to create a shortlist of papers, videos and manufactures that are useful.
two. In this adjacent step, yous will bear a deep dive of whatever resource you deem relevant to the subject field matter. It is crucial that at that place's a method to rails the agreement of each shortlisted resources. Andrew Ng, suggests a table of resources plotted against your understanding level that looks like to the tabular array beneath.
It is appropriate to ensure you go through at to the lowest degree 10–twenty% of the content of each paper yous accept added to the list; this will ensure that you have been exposed to enough of the introductory content inside an identified resource and are able to gauge its relevancy accurately.
For the more relevant papers/resource identified, it is expected that y'all progress to a higher level of understanding. Eventually, you will take identified some advisable resources with content that you lot sympathise fully.
You are probably asking yourself, "what number of papers/resource is sufficient".
Well, I don't have the answer, but Andrew does.
According to Andrew, an understanding of v–20 papers volition showcase a basic understanding within the subject thing, possibly plenty understanding to progress to implementation of techniques.
50–100 papers will primarily provide you with a very skillful understanding of the domain.
Later going through the resources and extraction of vital information, your tabular array might await something similar to what's shown below.
3. The 3rd pace is a quick tip from what I've observed works for me when trying to empathise inquiry papers. The third step is to take structured notes that summarises the key discoveries, findings and techniques within a paper, in your ain words.
The following steps will now be focused on how to read a unmarried research newspaper.
Reading A Single Inquiry Paper
Reading for the purpose of understanding is non done through i pass of the contents within the paper. According to Andrew, reading a paper from the first word to the last word in ane sitting might not be the best manner to form an understanding.
Be prepared to go through a newspaper at least 3 times to take a good agreement of its content
4. In your showtime laissez passer, start with reading the following sections within the paper: title, abstruse and figures.
5. The second pass entails you reading the following sections: introduction, conclusion, another pass through figures and scan through the rest of the content.
The introduction and conclusion department of a paper contains clear and curtailed data on the content of the paper and a summary of any findings. The information presented in this section ordinarily dismisses any supplementary data and only key data are included. This is beneficial to y'all as a reader as you get the vital information required to go along to the other sections within the paper.
6. The third pass of the paper involves reading the whole sections within the paper only skipping any complicated maths or technique formulations that might exist conflicting to y'all. During this pass, you can also skip whatsoever terms and terminologies that yous practice not understand or aren't familiar.
seven. Those conducting in-depth research into a domain can accept a few more than passes. These additional passes volition mainly exist focused on an understanding of the maths, techniques and unknown terminologies presented within the paper.
For those who are mostly reading research papers for advisory and technology purposes, and then in-depth research might be very time consuming, especially if y'all take xx more papers to become through.
I went through the process presented in this article with the original paper introducing the LeNet convolutional neural network, and I summarised the key content in notes which I afterwards and then converted to a serial of Medium articles.
Questions To Ask Yourself
Andrew provides a gear up of questions that you should ask yourself as y'all read a newspaper. These questions generally will show you understand the critical information presented in a paper. I use the questions below as beacons to ensure I don't stray from the aim of understanding vital information.
They are equally follow:
- Describe what the authors of the paper aim to accomplish, or perhaps did achieve.
- If a new approach/technique/method was introduced in a paper, what are the key elements of the newly proposed approach?
- What content within the paper is useful to you?
- What other references do you want to follow?
Conclusion
"Larn steadily rather than brusk burst for longevity."
— Andrew Ng
I'thousand nonetheless relatively new to the field of Machine Learning and Computer Vision, there is a lot that I do not know (that'due south an understatement). Still, I believe that if an individual is consistent in their search for cognition, regardless of the domain, they'll be rewarded with an understanding and expertise that surpasses the norm.
From the techniques introduced by Andrew Ng, I'll be reading at least 4 research papers a calendar month, reading to the bespeak of understanding. I'll be honest and say that the LeNet paper took me near a week and a one-half to complete wholeheartedly. But you get better and faster at reading and agreement research papers the more than times you do it.
Andrew states in his video that he carries a batch of research papers around with him, intending to read them. Andrew is a prominent figure inside the field of machine learning, and I believe emulating his habits and learning techniques can be advantageous to your learning journey.
Source: https://towardsdatascience.com/how-you-should-read-research-papers-according-to-andrew-ng-stanford-deep-learning-lectures-98ecbd3ccfb3
0 Response to "How to Read Machine Learning Research Papers"
Post a Comment