100 thoughts on “MIT 6.S094: Introduction to Deep Learning and Self-Driving Cars

  1. Hi Lex, thank you for sharing your lectures. I'm catching up on AI after a number of years researching biologically realistic models of neural processing. I used MS cog toolkit in the past. May I ask if there is a reason you didn't mention Open AI here?

    (I understand Numenta is falling from favour of late.)

  2. It is congruently amazing to the topic that you are posting this online with the course materials on your website. Not many aspiring engineers and software designers get a chance to go to MIT and by making this open source you are providing a pathway not only for the future of AI and self-driving cars, but self-learning and the education revolution that is bound to happen. Thank you.

  3. so supervised learning is when you you know inputs and you know outputs.and pong is made with unsupervised learning in a way that in 200 000 games you create those inputs and recreate supervised learning greatly explained XD

  4. Lex, for the game of pong, why not to use the "ball" and paddle pixel position as the neural input instead of the whole image 80×80 pixel? The problem aproximation looks more simple considering the ball moves on a linear path.

    The proposed method looks also nice, but not actually so incredible: in other words the neural network is understanding that a reward is given when the paddle position matches the ball position.

    Although, reinforcement learning always looks beautiful! And thanks for the lecture!

  5. Why is the traffic jam such a big deal? Smh 🤦🏽‍♂️ how many of you are enrolled at or teaching at MIT? 🤔 there a lot of better questions to be asking about and putting your energy into. Leave the lecturer alone!

  6. What if human brain is as easily fooled as neural network? But maybe we just don't have tools to optimize image received by our eyes to fool us.

  7. Poor lecture from MIT. I am sorry for the offense. After watching Dr.Gilbert Strang's Lecture, this is not so cool. I am sorry again

  8. Mr. Fridman,
    I have recently installed front and rear dashcams in my car.
    After watching so many road rage incidents on youtube I decided to get cameras.
    I have been driving since the 1960's.
    It is my perception that the average driver has been getting
    dumber,
    more careless,
    more discourteous (it's my lane and I am not going to share it with you).

    It has appeared to me that drivers as a whole do not
    apply any strategy to their driving.
    The main problems that most drivers are guilty of are:
    1. They drive too fast.
    2. They drive too close to the car in front of them.
    3. Young Street Outlaw, wanna be's zigzagging through traffic as is everyone else is standing still. THAT'S INSANE.
    4. Driving IS NOT a competition but functions more efficiently if based on a
    cooperative strategy.

    Think of a cooperative Game Theory instead of competitive chess.

    If we all drive in a more cooperative mentality we all get to our destinations quicker and safer.

    Although the idea of the government mandating any more in our lives,
    I can't wait till care manufactures add three features to ALL CARS:

    1. Anti-talegating devices.
    2. Anti-speeding devices.
    3. Certain collision avoidance systems.

    I can't wait till the day this happens.
    Although I think these technologies are only going to be
    added incrementally.

    What they could do today is add
    forward-looking radar directly linked to the breaking systems.

  9. The dark chocolate is not German translation :), chance is higher to be Swedish (And at the beginning o the video, the car traffic nightmare going around should be somewhere like Paris 🙂 ).

    That was the fun part, the serious part is, wow. had to watch all of them several times. pause, look up several parts, go through the material over and over again. This is heavy stuff, not easy, but oooohhhh so interesting!

  10. 41:20 I immediately thought of the 11% chance of survival little girl in "I, Robot"… sometimes humans have to make calls that involves self sacrifice, and each human will make the call based on their own psychology, I guess it all comes down to learning and randomness anyway, since not everyone will choose to sacrifice themselves to save another.

  11. When he says "…unlike the human brain, where neurons die and are born all the time…" (around 48:43) this is NOT true. Adult brains do create some new neurons but some parts of the brain never undergo neurogenesis. What's really changing are the CONNECTIONS between neurons.

  12. Nice to learn about the concept. Its simply awesome. My question if this concepts comes Live, how many jobs will be lost. Does this world really need Self-Driving Cars. Can we call this act as Intellectuals eating away less Intellectual jobs on the name of advancement?

  13. STEP ahead and boost your career with deep learning tutorial
    Top 3 Udemy courses related to Deeplearning at discount price $10

    1.Data Science, Deep Learning, & Machine Learning with Python
    ——–> http://bit.ly/2yzOSAC

    2. Natural Language Processing with Deep Learning in Python
    ——–> http://bit.ly/2wcxiRt

    3. Deep Learning A-Z™: Hands-On Artificial Neural Networks
    ——–> http://bit.ly/2vtqcaG

    #python #deeplearning #naturallearning #AI #ML

  14. Lex Fridman,
    Thanks for your valuable lectures, during your lecture you are very slow , reading form screen , continues long pauses while speaking , looks like you are confused or not well prepared for lectures.Please don't mind , its for improvement because your are helping world.

  15. Only a idiot would get into a car with no human driver. Period. Listen. Understand. That Terminator is out there. It can't be reasoned with, it can't be bargained with. It doesn't feel pity of remorse or fear and it absolutely will not stop. Ever. Until you are dead.

  16. I would like to point out my disappointment and disbelief at the students in this class, who are so fortunate to be present there unlike me, but are so casually distracted from the course.

  17. Do you have to store every Markov chain from start to finish of each "move?" Wouldn't it be more efficient to store a starting position, i.e. where (a range of pixels) the "ball" was hit by your opponent's "paddle" and a small range of slopes of the trajectory? That way, you could build a complete lookup table. On such a simple game, the neural network would probably be unnecessary after training. For although a complete Markov table would be huge, that's only the case if you grab the whole environment every time. I realize you are looking for a general solution, however.

  18. U.S. Military, watch out for deep learning presentations and such if you do not wear glasses. (antique style glasses are not the best way to go with technology and neural advancements.) BCG's may just confuse you, but that could be the best thing, better thing when dealing with these guys presentations

  19. It's so wonderful to be able to learn from you. Thanks to MIT for extending the opportunity. I'm thankful to be able to see so much information and learn from someone so educated. I'm really optimistic and interested to be part of my future because of technology.

  20. Don't be fooled A.I. will take us over. It will get so bad that GOD gets involved, you will see… Just avoid the implaints unless you have no regard for the health of your soul.

  21. Some sages speak on how we as humans have this body-mind, I call the highest form of tech in the world. We have 90% of inactivated dna… maybe one of these days we’ll be able to “interface” with the internet itself. Dream of mine in 2012. I was in a walking machine. I was connected to it thru my mind. I freaked out because I stepped on a human below me.. I lost focus.. fun dream 🤓

  22. AI Cars and Self-driven cars are all scams to obtain university grants from the government> Evidence: https://www.youtube.com/watch?v=9sgetWQGYxY&lc=z22mdfebvnfrspqfbacdp431lncbovaua3dhpynkby1w03c010c.1531226423315275

  23. Damn, that General Purpose Intelligence shit is good. Magnificent video. Liked and subscribed! The way you explain and describe it, it's so simple but so deep. Love it.

  24. Tokyo and Europe are lacking rules, good roads and street markings? Less requirements? Have you heard of the German Autobahn? Have you driven a car in Tokyo? Dude… You should really get out more…

  25. guys please help me on this – i had decided to choose the deep traffic and deep tesla as a part of my mini project
    well i wanna know whether after watching all the lectures in the playlist will i be in a position to solve the deep traffic project
    i want to know whether these lectures are enough to solve the deep traffic problem since i have just 8 days for my project deadline
    please reply

  26. Hi Lex,

    Thank you so much for the wonderful courses. Really great !!

    Also, for students like us who like to understand deeper on how to build the Deep Learning for self driving cars algorithms, do you also provide the code(Python?) which we can take a look at?

    Thanks again !

  27. The general purpose intelligent pong game playing network does beat the computer but it goes up and down like a drunk guy. That's expensive. How can we modify the training algorithm to also conserve energy?

  28. Take whatever the most powerful GPU can do graphically. Now, imagine that backwards. Matrix transformations on each layer depend on every other layer simultaneously. Apparently continuous values were only suggested this year as opposed to discrete.

  29. 1:02:06 I don't know what language that is, probably one of the Nordics. Definitely not German, which would be something like dunkel Schokolade.

  30. Sir, I am a researcher from Pakistan working on self driving cars using deep learning. Your way of teaching is excellent and very effective.

    I need the presentations slides if you may share with me.

  31. I know this is an old lecture but I will still comment on it. Watching the pingpong game of Andrej Karpathy I must say that the ML racket is moving most of the time in the lower and middle left area of the play field. Once the ball is sent in the upper left corner the human scores. Another thing is that the ML racket moves faster, it has а higher frequency of movement, meaning that it is not predicting balls position but rather focusing on speed of movement. Then again all movements in this simple game are geometrical and easy ti predict. I would say that the same result can be achieved without a nerion network and ML. I do not think this game and the result of it is impressive at all or it can serve as a good example of how good are neuron networks.

  32. 8:47 – I didn't realize Artificial Intelligence had been born yet… Unless, of course, we are talking by concept – conceptually it is very easy and has been very easy…I think Turing would agree.

    But, if the concept is guided by a question asked and answered in a 1996 Time article… yeah, if I think about it, I guess it was pretty difficult. But that answer to the question is a false assumption from the outset. That's why A.I. today isn't A.I. There are intelligent systems that have developed, and I imagine it was difficult getting there. …It was indeed difficult getting to not A.I. – that seems accurate.

    And listening to the interpretation of AI sufficiency of Turing explained by this dude, it's clear A.I. isn't even on the damn radar. …Even where Neural Networks are today, how he so non-elloquently transitions to from Turing, are developed completely missing the point. The state of the art is a failure as far as A.I. is concerned.

    Listen to this dude, you'll never get to Artificial Intelligence. It's upsetting this is where the next generation is being taken to with respect to AI. As someone who has studied AI longer than this dude inspired in HS in 1996, it's amusing…but at the same time a goddamn tragedy…

    My graduate advisor would puke in his soup listening to this… But, that's the World's foremost Epiphenominologist (former; retired), a scholar of A.I. longer than this dude has been alive, and a former graduate assistant to Karl Popper.

    …I suppose a lecturer is just a lecturer, even if it is MIT…

  33. Rogan effect. For ya’ll don’t know….watch Joe. He will teach you much much more than a singular focus genius. Joe will give you jujitsu genius from the street. We are all watching MIT.

  34. Looking at the figure at 27:26 causes me wonder why the green bar is not configured to calculate where/how to hit it based on the location of the white bar and it's probability of successfully hitting the ball back.

  35. I really love how you teach and present these complex topics. You make the ideas very easy to grasp and also it's just relatable

  36. Lex please tell your students to develop anti-hackware for AIs . They are particularly vulnerable to hacking i've read. Self-driving cars could be used a weapons. That needs to be done before they are considered roadworthy. Safety first.

    The below is from a 2017 article from Gizmodo.co

    https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425

  37. Neuron – computational building block for the brain
    Artificial neuron – computational building block for the neural network
    Human brains have 10000 computational power than computer brain

    Doesn't match expected output – adjust the weights

  38. Universality – for any arbitrary function f(x) there exists a neural network that closely approximate it for any input x

    Special purpose intelligence – bedrooms, sq feet, neighborhood -> final price estimates

  39. Do I move up or move down?
    – 200000 simulations
    Supervised learning – need examples to generalize
    Unsupervised learning – deep learning – no major breakthroughs yet

  40. Neural network –
    Requires big data. Inefficient at learning from data

    That game in your mine
    Local packets of high reward. Local optimza

  41. Deep learning – pr term for neural networks. Network layers that have many layers
    More data we give

    Imagenet classification error
    Translation of text in images

  42. Morwavec – encoded in the largely highly evolved sensory of the human brain is a billion years of experience about the nature of the world

    Face detector Incorporated in consumer culture
    Lower level tinkering

Leave a Reply

Your email address will not be published. Required fields are marked *