Deep Learning Lecture 1: Introduction

[@ProfNandoDF]

After my comments regarding the lack of enough GPUs at Oxford, NVIDIA kindly donated many cards for research in deep learning. We are keeping them nice and warm performing many calculations. Thank you, NVIDIA !!!

[@jcpate1295]

No, actually while CUDA is ahead OpenCL has been catching up. Most of the largest apps support both now.

[@WaitingForStorm]

my computer lacks gpu too

[@ksrajavel]

@@WaitingForStorm :P

[@WhishingRaven]

@@WaitingForStorm :P

[@dc33333]

Excellent work Professor Nando de Freitas!!!!! My favorite youtube series of all time!!! Thank You!!!

[@chspx]

Nando, your lectures are awesome! The best I've seen since I started my studies on this topic. Are helping me a lot. Keep up the good work and thank you!

[@chainesanbuenaventura2874]

Omy! Nando! I found you again! I watched your machine learning lecture series and it inspired me to pursue a research on machine learning. And now I'm also learning about deep learning! Love your lectures!

[@adamwkeane]

Just to say thank you for the lectures and all you are doing to advance this technology.

[@saltcheese]

Thanks a lot for these videos! I especially like the little academic history you sprinkle about in these lectures.

[@thestopper5165]

Great series (the playlist is out of order, but that's so trivial that it's a non-problem). I bet you get lots of chicken-based jokes (or you would in Australia - there's a fast food chain called "Nando's").

[@Brucejpw]

Nandos started in South Africa. Interesting trivia..

[@ivanfinaev7104]

around 9:35 Professor Nando de Freitas talks about the origin of the face detection technology which we treat now for granted. Is it Boosting? Have I heard it correctly? If the answer is YES, what article have been in mind? I suppose, it is "Robust real-time face detection" by P.Viola and M.J.Jones

[@albenali]

The cat experiment at 15:00 was carried out by Hubel and Wiesel, for those who want to look it up online.

[@jonsaposhnik5583]

Well done, thank you for posting

[@3dsideviewface]

Great lectures, but they general all have some background noise, and the voice sounds like far away and sometimes high pitched.. This phenomenon happens a lot in lecture videos online, I wonder if there are any automatic methods for filtering out background noises (while keeping human voice) and normalize human voices on those videos, it would be much more pleasant to listen to.

[@scargomez9437]

Thank you Nando. Why you stopped you videos?

[@lkd982]

at around 31:00 "at the coffeeshop people say the same things all the time"...here we have the beginnings of introspection, wisdom, and science

[@amdaliHADID]

+lawrence mcdonell i couldn't catch the sentence before this one could you write them to me please? " on the use of the same expression ..... occasionally, these is a beautiful difference ... "

[@pi5549]

@Nando, the videos in this playlist are in the wrong order -- you can reorder them by dragging them.

[@martinhoang95ify]

Thank you professor, your lectures are awesome!

[@b1510]

Excellent lecture, easy to follow. I enjoyed that :)

[@smacedo99]

Thank you Nando! Thanks Shawn for a relevant comment.

[@suntank6]

It's not all shape recognition, some raw data does get pushed through since the brain is optimized to always be learning even from current experiences. This is why if you stare deeply at a word long enough it starts to look strange to you because you're seeing through your brains expectations and focusing on the reality of it. The brain is able to push raw data through to the consciousness and then it's up to the consciousness to spur a deeper understanding of that information.

[@TheArech]

Thank you very much for the lecture!

[@stanleychrismutocho8109]

we

[@jeffreycliff922]

access to the source code to do this would be useful

[@SkatingErinsMom]

My Internet is too slow for streaming anything and you've blocked these videos from downloading, so I can't watch them.

[@shoaibahmedsiddiqui2238]

Thanks a lot for sharing these invaluable resources. Can you pl also share the lab tutorials so that we can benefit from them as well.

[@palfers1]

The sound quality is entirely unworthy of Oxford. Why not use some of that intelligence to fix it?

[@adaylateacoinshort]

+Andrew Palfreyman the answer is outsourcing

[@naqqashabbassi684]

Professor mentioned something like, "We get fooled by our own minds". Anybody who is interested to know more about it should definitely read the book "Thinking Fast and Slow". There is whole a field dedicated now on this topic commonly referred as "Behavioral Economics"

[@theodorosgalanos9663]

Or you can read a bit of Marx, or Freud, or even Plato :)

[@luisleal4169]

Hello Nando, i teach ML in a masters in data science program and i really liked your introduction with the videos of the plane(and the car) is it possible that you share them with me?(or provide a way to look for them in case they are public)

[@siyuzhou7512]

Thank you for posting

[@zuodongzhou3334]

Excellent lectures!! any online assignment accessible for this course?Thanks!

[@A-Go-5]

The production quality is extremely poor given the number of people interested and watching from afar.

[@hirdeshkumar7027]

which book are you using?

[@supailinpichai]

Excellent lectures, Thank you

[@sriramhegde758]

+supailin pichai are you girlfriend of sundar pichai?

[@supailinpichai]

+sriram hegde No.

[@sriramhegde758]

supailin pichai oh ok lol

[@lkd982]

around 10:00 "another thing about intelligence: once you've figured it out, it's no longer intelligence" not very profound, but good anyway :)

[@chainesanbuenaventura2874]

I don't get the part where from the raw pixel data, it is fed into the first layer neurons, and new features are developed such as edges, corners, etc. How did this happen? You mentioned that if your cut through the lines you'll form these new features? Can you explain further? Thanks!

[@赵三谦]

Hi, Nando. Have you tried to imply the idea of deep learning based on large scaled pcs using techs such as hadoop + mapreduce?

[@SpongeWarlock]

Thanks for your lecture, I appreciate! :')

[@HiTechTopTIPs]

the audio is not good

[@omar778]

Thank so much professor it's really very interesting lecture

[@skyoceanfun]

thank you for your sharing.

[@lyltencent]

Try to learn deep learning and CNN from your video. Thanks !

[@p.z.8355]

Can someone explain how those images are computed in which f.e. the edge, one neuron encodes, is shown ? In general how can i know what kind of attributes one neuron encodes ?

[@LokeshSharmaa]

Is these series about Deep learning or machine learning? The title says Deep Learning but the link to slides says Machine Learning.

[@w3bst3rWu]

Deep learning is a branch of Machine learning.

[@beaugogh]

why were the two images switched back and forth instead of being shown at the same time, only due to poor presentation style?

[@ProfNandoDF]

The switching has to do with change blindness. This website by one of my heroes explains it really well: www2.psych.ubc.ca/~rensink/flicker/

[@juandenz2008]

Stupid question, but what level is this course aimed at ? I notice it is taught at Oxford, but is it for undergrads or graduate students ? Also what are the assumed prerequisites. Thanks.

[@YalnekH]

+juandenz2008 Postgrad (masters), and prerequisites are "Good background in probability, linear algebra and calculus. Programming experience is essential."

[@ThomasKirwanE8]

The only pre-rec is passion.

[@omkay2569]

It's a third year undergrad course. Prerequisites are Continuous Mathematics, Linear Algebra and "Design and Analysis of Algorithms".

[@dinasaif4082]

I want to run CNN and DBN on ARFF file or I can convert it to excel sheet , it contains binary features (0,1) also two classes binary label(0,1) can you help me how can I make classification for that by CNN or DBN

[@jiansenxmu]

awesome！

[@IanAtkinson555]

Deep Learning Lecture 1: Introduction

[@pablodiaz1811]

Thanks

[@gulpowski]

I think this Hubel & Wiesel approach is going into the wrong direction. As you have shown in the change blindness examples the brain is a difference detector and in visual input differences occur mostly because you move your eyes about 50 times per second. In my opinion a horizontal line is defined by the fact that there are small changes when you perform a horizontal eye movement and large changes when you perform a vertical movement. Lines with similar angles result in similar changes though only few pixel might overlap.

[@lkd982]

"if we don't introspect, then we have no hope of programming a visual system.." Notice the use of the word "introspect", determinists

[@judgeomega]

"Each human optic nerve contains between 770,000 and 1.7 million nerve fibers" -google

[@allanng78]

Hi, Do you know how can I apply deep learning on malware detection?

[@PauloConstantino167]

tu eh brasi'eiro que eu sei

[@timanderson1054]

What you described as the Scottish eyebrow problem for AI images classified in Japan is also the real world Kenya problem, AI images are classified in Kenya (lower pay rates $9 a day exploiting impoverished countries), but the AI is for use in America. This is bad science using the wrong data sets. Kangaroos can leap nine metres in Australia. I would rather have control of my own car than let AI drive for me, your focus of saving one million lives seems like you do not care for humans as intelligent agents. Maybe the one million killed are the stupid ones. Your picture is all together too rosy and robbing people of agency. There are many problems with car AI you should acknowledge, so much vulnerability to unexpected events or a deceptive environment! The Washington University school of law advised that deceiving car AI would not be hacking under current laws. A hackers delight, too easy to exploit with simple means of; sticky tape on STOP signs or white dots on the road and no penalty for any accidents caused. www.hereticpress.com/google/index.html#Kenya

[@timanderson1054]

PS You should do a course on philosophy and study free will, you have nothing to offer real humans except misery and slavery to the machines. I expected better from Oxford, some mention of jobs lost for drivers perhaps and how easily driving AI can also be fooled by simple white dots on the road. What a miserable dystopia you are promoting along with a sound recording unworthy of Oxford university.

[@YossiKahlon]

it is a shame how they treat animals for this. these poor animals need to suffer for this. we are a terrible species

[@timanderson1054]

He doesn't care about human agency or jobs for drivers, animals mean nothing to him at all, a heartless man promoting a dystopia of unemployed humans.

[@0sba]

Jesus christ that cough is annoying

[@divinerevelation9883]

Your accent and that low fi recording = unwatchable. A little ironic you are discussing such high technology but you would have sounded better with 1935 era recording equipment.

[@ludenstempest8244]

+Divine Revelation Who forced you to watch this ?? I mean you should been pleased to have the chance to learn from the internet..What an idiot lol.

[@divinerevelation9883]

HodorDisco The person who decided to record it, recorded it, saw it was shit but still posted it is the idiot. Your logic is shit.

[@ludenstempest8244]

+Divine Revelation You can hear the sound well every pdf is on the net and u can see it so clear. There isnt any kind of problems with this video.Atleast I have logic :D..

[@timanderson1054]

You call anyone making a valid comment an idiot so you are stupid, undiscerning and accept any old crap, you are the idiot. How stupid are you to accept everything as it is, this is about AI so you would have a right to expect quality recording, but the lecture also lacks any human context or concern for humans. What a stuck up name for an egomaniac @@ludenstempest8244

[@Titu-z7u]

why everyone so busy with unnecessary theories..we need a simple step by step process so we can code and implement it

[@theodorosgalanos9663]

Every scientific endeavor at the so-called bleeding edge really needs nothing but these 'unecessary' theories you mention about. That is how scientific progress happens. Operationalizing these theories is important but it comes later and thankfully it is rarely the focus on early stages of development. If not, all we would ever do with ML would be making apps to make money out of people tweets.

[@Titu-z7u]

o bro,at first we better see how it works so we could at least hold our passion for it.and the thing is its not science at all.this is technology.science is studying the nature which it has nothing to do with

[@abijo5052]

This isn't an easy ML tutorial. This is a lecture from Oxford. There are plenty of tutorials but if you don't understand the theory you'll struggle to actually adapt to your own ML problems. Also a lot of oxford students will go onto research, in which case the theory is entirely necessary.