Research to Code - Machine Learning tutorial

[@wolfisraging]

You literally are the first person talking about BFGS on KZbin, that's what I love about you.

[@MusixPro4u]

The "Your code can't fail unit tests - if you don't make unit tests" meme cracks me up every time. So good.

[@nikanj]

I'm usually against piracy but when it comes to academic papers just do it. Academic journals are an archaic institution who hold the world's knowledge to random. They don't fund the research or pay the authors - institutions, corporations and governments do. They don't review the papers, fellow researchers do. So it boils down to this: -Researchers do research and write papers rather than making much more money in industry because they enjoy the pursuit of knowledge and want the world to be a better place. -Institutions and governments fund research with the tax payer's money for the same reason, because knowledge makes the world a better place. -Peer's review papers to validate the integrity of papers. Yet with all the money and effort invested by these parties it is the journals who for some reason get to reap all the profits for what is effectively a spell-check and upload. This knowledge should be free to the world and it's about time we broke free of this nonsensical and outdated system.

[@merth17]

greatest comment on youtube

[@NB19273]

Very intuitive and concise explanation of neural style transfer! These types of videos although slightly harder to digest are much more beneficial for people who are serious about progressing in ML as they teach core competencies of the field . great job as always!

[@smoofwah3552]

World: Hey Siraj ! Teach me more about that Machine Learning.

[@MusixPro4u]

That wave background is so satisfying.

[@2002budokan]

The essence of this (long video with a lot of noisy details) is at 4:46: "A paper is a succession of equations". This sentence implies that you should try to understand how the paper define and compose functions for its purpose, and convert this composite to code. Thank you Siraj Raval, you select useful topics, however try to remove noisy details. Shorter is better.

[@atrumluminarium]

For the "if there's a paywall" part, just email the authors and ask them if they are willing to send you a PDF of the paper (or if you are enrolled in a University, their network usually has access to most paywalled access so download it from there). Most likely the author will be more than glad to send you a copy if he/she has time because they get absolutely nothing from the paywall money anyway.

[@adityapatil325]

atrumluminarium Why not just get it from sci-hub?

[@atrumluminarium]

Aditya Patil Because I have never heard of sci-hub lol. I'll check ot out

[@karthikrajkumars]

Thanks for making the world a better place for AI enthusiast. Keep the awesomeness going!

[@Christian-mn8dh]

Gotta love you Siraj, I was just thinking about how to convert my research to code.

[@masterjiggle5728]

Wow what an explanation for neural algorithm for style extractor and raw image

[@user-qw4jf1rw8l]

To be honest, you never let me down Siraj~~~~!

[@BosakMaw]

Nice pokemon reference at the end

[@MusixPro4u]

I think you struck a happy medium in talking speed in this video!

[@akramsystems]

I swear I literally asked for this and it came you are amazing!!!

[@CristianVasquez]

I would add 1. Search for the slides used in the conference, sometimes they are on the Web. 2. Manage the papers you gather and your annotations with a tool such as Mendeley.

[@jtcothran]

Loving the videos Siraj, thanks for the helpful explanations and coding intros to the latest AI developments and fun attitude in general. One thing you mentioned in an earlier video was some type of 'knowledge tree or graph' for your videos or discussion - that is something that I would find useful as usually when I'm watching a video like this I am back-referencing everything else I've learned to see how what is being discussed is unique in concept or practice and how it might connect to other areas, basically building that mental map of things we all carry around, misconceptions and all.

[@ironmonline]

Great videos! I learn a lot from them, you are like the Salmaan Khan of machine learning :)

[@adityaraj-jr8ph]

My bro doing it for the squad . I am currently on a mission to get a research position on research lab . Once a reach it . I will record a rap for my brother on soundcloud

[@KunwarPratapSingh41951]

There is not much thrill in implementing research paper without learning how it is actually working, why it is working. I would suggest to go for every basics of maths and everything that a research paper uses to really understand it completely. It will be easy if u r phd student.

[@empiricistsacademy7181]

If there is a pay wall for a research paper, just email the author(s) asking them for the paper or latest manuscript.

[@BhanuTejapolukonda]

Thats awesome I was waiting for this kinda video

[@torgammelgard4467]

Cool, inspirational and educational! :)

[@MuhammadAwais-vi8de]

simply awesome

[@AbhishekKumar-mq1tt]

Thank u for this awesome video

[@useryash09]

Good one. I was looking for something like that. Thanks😎

[@subhajitdas2784]

You sir are great. Thank you once again.

[@jameslatief1]

Hello Siraj, it's World.

[@AlokTripathi]

One of the great video, thank you so much!!

[@sladkacokoladka]

Any advice how a non-CS person can land a ML/data scientist job, when requirements usually include BA/MS degree in CS or similar? Being enthusiastic about the field and having own projects on GitHub don't cut it... Love your videos BTW

[@beskamir5977]

I would suggest being knowledgeable/experienced in the field of the problem you are trying to solve with ML. Say the problem you want to help with involves self flying planes but you don't have a degree in computer science. Well maybe you can still get onboard that project by being a capable pilot or experienced with how planes work or geography, or atmosphere stuff, or whatever. Then (probably easier if the team working on the problem isn't very large) you could probably get to a position where you can do some ML stuff which could help buff up your portfolio for future jobs.

[@sladkacokoladka]

Beskamir I get what you're saying and that is at least one way of MAYBE doing some ML stuff if you get a job as pilot/... and the job then evolves to ML - by chance. Lets call that plan Z. Putting luck aside, I am wondering how to get a position specifically for ML/data scientist. I have the knowledge but no degree, so employers filter out people like me in the first stage of recruitment (Siraj even made a video about CV filtering a while ago ... I'm getting filtered by ML applying for ML position. Talk about irony 🤓)

[@adityapatil325]

sladkacokoladka Try to get an internship, possibly an unpaid one

[@humano3753]

Dude, thats great as fuck It would be awesome if you make a video just talking about what are researches struggling with

[@JeJeNtwrk]

this is so so cool....... great work...

[@evolutionisnear]

Projects in the cloud; My projects are of clouds. Rorschach-training, to re atrain my braining trails I have so I re-up to the connects that can be framing.

[@AkashMishra23]

Insight Generation out of large unstructured datasets get me going

[@1minus2]

looking good siraj

[@shahikkhan]

Love u Siraj Keep it Up ✌

[@hardikaggarwal7426]

Great work bro

[@aradarbel4579]

Hello Siraj! Nice video! can you create a video about the softmax activation function and its loss function? this would be great!

[@larryteslaspacexboringlawr739]

thank you for tutorial

[@arshadshaikh4676]

Bro , make new videos soon, we need you.

[@MLwithAlva]

Do you have any tips for how to prepare for conferences? Going to the IJCAI conference on friday and would love some advice on which workshops/talks I should go to 😌

[@Roy51794]

Have fun, I envy you!

[@manashejmadi]

search for meetups nearby on meetup.com

[@vertonical]

Are you presenting a paper @ ICJAI? I would love it if you gave us all a link after to read it

[@MLwithAlva]

vertonical no, I was just there as a student volunteer ☺

[@vertonical]

Alva Liu Any tips on how to become a volunteer at AI conferences? I'm just a CS ugrad sophomore starting out in AI research. I haven't got any papers published yet though I hope to soon. What qualities do they look for in volunteers? Do academics like GPA matter?

[@ukimalla]

You inspire me everyday! I recently graduated with a degree in CS, and I'm deciding on whether I should go to grad school or start looking for jobs. Do you have any suggestions? I'm decent at math, understand the basic concepts of most ML models, got an A in my Deep Learning last semester. But my GPA isn't that amazing. I have an average of 3.61, and don't have a lot of projects to show for in github. I have worked on a hand full of ML projects, and a few DL projects on my own. Do you think I'll be able to get a job around the valley if I spend a few months preparing for it? I'm not sure if I should waste money for another 3 years of college only to learn what I could've learned on my own. You probably don't have enough time to reply, but keep up the good work! You're on fire these days!

[@beskamir5977]

>Not amazing gpa >gpa is a 3.61 thanks for making me feel bad :(

[@Ankitakin]

Loved the video 😍

[@zuenko007]

# Architectures * AlexNet: papers.nips.cc/paper/4824-imagenet-classifica.. * ZFNet: arxiv.org/abs/1311.2901 * VGG16: arxiv.org/abs/1505.06798 * ResNet: arxiv.org/abs/1704.06904 * GoogLeNet: arxiv.org/abs/1409.4842 * Inception: arxiv.org/abs/1512.00567 * Xception: arxiv.org/abs/1610.02357 * MobileNet: arxiv.org/abs/1704.04861 # Semantic Segmentation * FCN: arxiv.org/abs/1411.4038 * SegNet: arxiv.org/abs/1511.00561 * UNet: arxiv.org/abs/1505.04597 * PSPNet: arxiv.org/abs/1612.01105 * DeepLab: arxiv.org/abs/1606.00915 * ICNet: arxiv.org/abs/1704.08545 * ENet: arxiv.org/abs/1606.02147 # Generative adversarial networks * GAN: arxiv.org/abs/1406.2661 * DCGAN: arxiv.org/abs/1511.06434 * WGAN: arxiv.org/abs/1701.07875 * Pix2Pix: arxiv.org/abs/1611.07004 * CycleGAN: arxiv.org/abs/1703.10593 # Object detection * RCNN: arxiv.org/abs/1311.2524 * Fast-RCNN: arxiv.org/abs/1504.08083 * Faster-RCNN: arxiv.org/abs/1506.01497 * SSD: arxiv.org/abs/1512.02325 * YOLO: arxiv.org/abs/1506.02640 * YOLO9000: arxiv.org/abs/1612.08242

[@wafaawardah3264]

Nice video. Q: How do you motivate yourself on low days?

[@RAJATTHEPAGAL]

Wafaa Wardah i think he doesn't have lows , there are no minimas in hyperplane only saddle points :-p ......

[@SirajRaval]

thank you! yes i have lows, i just always try to have something to look forward to

[@paulgarcia2887]

Alcohol

[@theJACKATIC]

cocaine

[@AKASH-zt8oi]

can you make a video on how to prepare for GSoC ?

[@vivek6715]

Awesome Video! But just one question, you suggested to look into industry focused papers, how do you find them? Or how do you differentiate between academic and industry paper without reading all the content? I'm very much interested in applications than machine learning math.....

[@jackflynn3097]

Jez , reading and implementing research papers are truly headache.

[@nimeshpatil164]

whats total_variation_loss in code ? paper doesnt have anything like that

[@sandip7496]

Hi , Should I learn R ? 😎

[@Lycheeee11]

How can you quickly find papers with an industry focus?

[@bilaldendani2259]

Hey Siraj , thank you so much ? What are last trends about speech recognition ?

[@ykochubeev]

Siraj, could you please share info about tools you use for creating your great explanation videos? Is it adobe after effects & some animation plugins?

[@prateeksingh1284]

How machine learning models are created ? What is the difference between a Data Scientist and Machine Learning Engineer ? And what does an M.L. engineer actually does ?

[@StormyDoru]

@siraj make a video on ML in robotics!!

[@teimoorbahrami8634]

hey how can i implement the "Detecting and interpreting myocardial infarction using fully convolutional neural networks " in python

[@takeshifukudda6727]

Is there any good form or chat group where you can ask questions about those "Math equations" , in case some of them are difficult to read..?

[@27.Counting]

Hello Siraj , I request you to suggest me a channel for Data Analysis , with good quality as yours for AI and ML people .

[@spacenodus7959]

Cool cool cool

[@AI_In_The_wild]

I couldn't find the git chive tool you mentioned can any one please direct me to it

[@yusufkolawole8853]

pls I am having a headache right now, why are all these guys having thumbs down... haterx

[@siddhantnanda3692]

Can I get the links of the tools , to check whether the research paper has code linked or not?

[@SanjayGupta-xx8ow]

Where to get videos for ML ?

[@SingularityLabsAI]

What did you say the name of the tool which links a git repo to a paper is?

[@suryameda6215]

How can we make a Q and A answering chatbot with deep learning?

[@freediugh416]

hey siraj you should make a video on machine learning

[@techgurukul4545]

The website name mentioned for research papers is not audible Sir please mention few websites name for easily getting research paper

[@qalqi]

Here is a gist of links mentioned qrd.at/d6H

[@xPROxSNIPExMW2xPOWER]

Li vitanyi Kolomogrov Complexity paper is my bane. That is after the church Turing thesis obvi

[@Hishamhh93]

Early..! 😍

[@nissenebel7380]

Does somebody have a recommendation for learning NLP, like good research papers or a useful book? Have a nice day!

[@vipulpetkar]

holy crow!!!! TTTHHHHHHHAAAAAAANNNNNNNKKKKKKKKSSSSSS A BUNCH................. FOR 60 FPS

[@rajatkumar-ks1cu]

Hello sir . How many hours in a day do u work and sleep ??

[@nsakib99]

What is the tool called again ? Gitkive ? I cant hear it well.

[@SirajRaval]

Gitxiv

[@genkidama7385]

most papers can be found in pdf format on google. type this "filetype:pdf xxx_title_xxx". research projects are funded with public money and shouldnt be locked away from people, same for the code they write, should be forced open source by some authority. Intellectual property and public funding are not compatible.

[@bafa000]

ahahah 6 years ago ?!?!?!?! fck.

[@AndrewErwin73]

"machine learning" seems to be nothing more than the digitalization of classic statistical analysis. Honestly seems like something we should have been doing for a long time.

[@antonylawler3423]

Who thinks that a well written GitHub page is better than a paper these days ?

[@Mr.Sanguine]

Hi Siraj

[@kacperw3366]

is there someone who is looking for learning buddy (to start study Machine Learning)?

[@spayseghost]

how to pirate research papers? links plz :)

[@value1042]

Your from India?

[@criticalnodecapital]

Fuck , where the hell is this lesson. Can you do some ML concepts.

[@codethings271]

I dont think we are learning anything new now. Who else agree?

[@batcavemasters]

No entiendo nada :(

[@SivaPrakashJayaraman]

Am I the only person who do not understand any of his videos... :(

[@parsarahimi335]

These recent videos were too repetitive. Get a brand new topic/tutorial next time

[@SirajRaval]

sure, any suggestions?

[@parsarahimi335]

Recently this game came out, The Crew 2, and I was thinking even though the map is very large how can you actually expand this? Well you don't stand a chance doing it manually but AI does... So you can name your next title "How Google can generate detailed 3D models from street view images" (which I was thinking you can just implement this into games/simulators) Since you may not want to mention Google, you can name the title "How AI can architect faster and better than us” (obviously from previous sample data which we have taught it before hand)

[@atrumluminarium]

These videos are to offer some detail to beginners. Siraj has done a lot of content dedicated to the intermediate/advanced viewers who are more than capable of researching material on their own. Beginners need to be guided a bit more

[@FuZZbaLLbee]

maybe do a video on curiosity learning kzbin.info/www/bejne/b4SVmGqfnt11b8U

[@NickKartha]

HyperGAN, CBLSTM and PixelCNN, maybe.