Two Minute Papers - 2019-01-15
Tensorflow experiment link: https://www.reddit.com/r/MachineLearning/comments/4eila2/tensorflow_playground/d20noqu/ Karpathy’s classifier neural network: https://cs.stanford.edu/people/karpathy/convnetjs/demo/cifar10.html Pick up cool perks on our Patreon page: › https://www.patreon.com/TwoMinutePapers We would like to thank our generous Patreon supporters who make Two Minute Papers possible: 313V, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Brian Gilman, Christian Ahlin, Christoph Jadanowski, Dennis Abts, Eric Haddad, Eric Martel, Evan Breznyik, Geronimo Moralez, Jason Rollins, Javier Bustamante, John De Witt, Kaiesh Vohra, Kjartan Olason, Lorin Atzberger, Marcin Dukaczewski, Marten Rauschenberg, Maurits van Mastrigt, Michael Albrecht, Michael Jensen, Morten Punnerud Engelstad, Nader Shakerin, Owen Campbell-Moore, Owen Skarpness, Raul Araújo da Silva, Richard Reis, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Thomas Krcmar, Torsten Reil, Zach Boldyga, Zach Doty. https://www.patreon.com/TwoMinutePapers Thumbnail background image credit: https://pixabay.com/photo-1807526/ Splash screen/thumbnail design: Felícia Fehér - http://felicia.hu Random walk image source: https://en.wikipedia.org/wiki/Random_walk Tensorflow experiment link: https://www.reddit.com/r/MachineLearning/comments/4eila2/tensorflow_playground/d20noqu/ Karpathy’s classifier neural network: https://cs.stanford.edu/people/karpathy/convnetjs/demo/cifar10.html Károly Zsolnai-Fehér's links: Facebook: https://www.facebook.com/TwoMinutePapers/ Twitter: https://twitter.com/karoly_zsolnai Web: https://cg.tuwien.ac.at/~zsolnai/
You should also remember, that when an AI starts his first baby steps, it's dumber than anyone who has ever lived... but through determination, trail and error, and the inability to give up, it finds its way up and eventually preforms better than any human would ever do learning from its previous mistakes :)
:')
This comment was a deep learn for me :^)
Well that and an inability to forget a single detail about every attempt since the first one, plus an ability to recall all that information at once in order to guide the next decision in an educationally meaningful direction, plus an inability to feel demotivated, or fall into bad habits, or suffer from fatigue, or be misled.
So I mean it's still amazing, but it's not like it was a fair fight for humans. :p
The longest ever Two Minute Papers episode that's not two minutes and it's not about a paper (but many). You know, just the usual. Hope you Fellow Scholars enjoyed it! :)
Thank you very much for your work!
Dear Min - thank you very much for your support! I was trying to look up this place, but was unable to find it.
lol
I'm not directionless, I'm just taking a weird route to my goal.
This video is about zooming out and evaluate.
My take on this, after becoming an amateur on this field:
1) Truth is impossible to grasp because it is an infinitely dimensional reality map. We can make correct assertions but never uncover the full truth, as Plato explained in his Myth of the Cavern. Most of the people do realize when something they think is wrong, but are unable to locate it and restructure their reality map so they try to either simplify their reality or ignore their cognitive dissonances, with harmful results.
2) There is never a general solution for everything, but particular solutions for a particular problems. Also, most of the time a problem has more than one solution, even within the same restrictions. Absoluteness and uniqueness are simplifications fantasized by the human mind to cope with the universe.
3) Politics, economics, evolution, games, life choices... all of them work much like intelligence, where you want to improve your position by discovering the solution space and then making the choice of whether or not you take the risk of changing your position. Either you dare making progress or you play a conservative Minimax strategy. Taoism embraces this idea of balancing these two opposite mindsets.
4) The path to mastery resembles these steps.
-First, you don't know that you really don't know.
-Then, you know that you don't know.
-After that, you start to know.
-Later, you don't know that you know.
-And finally, you know that you know.
It takes one to start exploring to imagine the boundaries of the universe and put its truths into context. There exists the possibility of not realizing the real size of the universe and to be trapped in one's own ignorance as was shown by the Dunning-Kruger (and Reverse Dunning-Kruger) effect. Besides, Socrates long ago also talked about this when he said he didn't know anything. Finally, this means that you have to discover the map of reality twice: first to see everything without context and second to place everything in its correspondent context. Only then you'll be able to see the big picture.
5) In the Age of Internet it is increasingly assumed that you are using all the information contained in the Internet. You have all the theory in it. What you need now is most probably experience. I found this quote which i find relevant to this idea: "Just build something".
6) Do not reinvent the wheel, search for a guide to help you instead, because impulse can only get you so far and it takes a lot of knowledge or luck to really create something new or better. To help you be more prolific, you can "automatize, study, test, refactor and repeat" for all the boring but necessary tasks in your life so you can perform them ever more efficiently and save time and effort. Also, boring tasks are necessary, not only because of everyone's daily needs, but because being bored is good for your creativity. You can make the most of those tasks if your mind is doing one thing and your body is doing another as done in meditation, mindfulness and other therapies that are based on this so, when doing them, place your mind into a passive-creative mode like Archimedes until it is you who says Eureka in the bath. All the energy consumed for concentration is very valuable so you better reduce all interferences when using it, saving during those daily chores.
So to summarize attempt, measure error, correct, attempt again...or attempt, predict error, measure error, correct error prediction, correct attempt, attempt again...meta that cognition?
@stalinvlad In a way, totally yeah. I'd add that we should generalize less when doing predictions and try to add others to our prediction pool.
Great analysis and introduction of deeper context to lessons learned in the video through an epistemological and later practical angle.
I also really liked how you analysed this from another angle, in a more epistemologic and philosophical way.
And I couldn't help but notice how you put 3 different references to classical thinker, I found that really amusing.
Trying to add on what you made, I would include the Dunning-Kruger effect as a main step of the mastering process, "you think you know but you don't know", as the second step.
After all I just want to state that it's a great feeling being able to hear from other people thinking.
Piripiri, very well written. I agree with all of these. I used to dive head first into something without reading anything thinking it was better because it wasn’t influenced by anyone, but then I learned that it is only in knowing what has been done before that you can truly do something new. You need to build on those that came before you. Just as you must learn basic arithmetic before diving into calculus. One builds upon the rest.
This is simply beautiful
And beautifully simple
that where the longest 2 min i have ever experienced. And i enjoyed every second of it. Lllllove it
This video touched me and pulled me from the lack of progress I had. It just showed, objectively, that I'm progressing in my studies, I just have to keep focus and persist
Great video! I love what AI can teach us about learning in ourselves. I.e. the same techniques we can use to make machines learn effectively, can be used when we revise for exams. Revising diagnoses by thinking of concrete “examples”, like specific patients I’ve met with that condition, makes learning them so much easier.
Excellent video! Thank you, these are great lessons :)
Nathan Indeed, if written over the video
Fell in a local minima? Get wasted and try again! (Or change your hidden layers!)
Do you mean change the loss function?
@Remy XU some times the size of your hidden layers dictates wheter you can map a solution. Question your "design" randomize your weights!
@Remy XU He meant change the wasted function
Ah, yeah. Just found out such like the learning rate, optimiser can also affect the "felling in a local minima". @Damian Reloaded
Neuron dropout :)
So when will you be releasing your AI life coach?
Maybe it's a good idea to blur out the personal email addresses towards the end.
Good catch. Thanks!
Their names should probably be blurred too.
They agreed to that.
I loved the message behind this, thank you for your inspiring videos.
Hey man, I really love that you did this. I appreciate you aggregating a bunch of life advice from AI research. It's an interesting parallel between how our minds work and how we achieve our goals and how AI and mathematical experiments achieve the goals of the papers.
Please keep doing long videos sometimes ( or always if possible)..me and my friends learn a lot from your videos in a very short time .
2 Minute Papers : 7 Minute Edition 😅😅
Loved this video as it brought the intelligence component into real-world next steps for me
I enjoyed this 7+ minutes papers
a lot
"Choose your objectives like a drunkard" ;+}
1:50 Sanic
It wasn't exactly clear what you mean by "zoom out and evaluate" what was actually done to achieve this? Or is it supposed to be a generic life lesson?
Yeah, just stop for a moment and look how far you have improved from your past
To me it made me think of when I'm solving a problem, for instance a math problem and that afterwards I come back to see how I could optimize the process (and sub-processes) to derive a solution faster. Whether it's what was implied or not, I find this is a great strategy. It's a bit like tracing back your steps and eliminating the ones that are superfluous. It's about finding the optimal path to arrive at a solution (or, at least, approaching an optimal solution).
Perfect timing to recommend this youtube ! <3
The RW Analogy is massively indpiring and motivating. Thank you so much for your exceptional work !
For the change the objective change the strategy part,
I learned that through No Free Lunch Theorem for Machine Learning and Optimization
Thanks for the video, I found some really interesting ideas for myself. I'm talking about drunk walking. The huge takeaway is even if you are working stupid an inefficient it is still very valuable, and one more important thing from this is working smart is one power higher than working stupid (which is still better by infinity than doing nothing).
So 5 and 6 are really significant lessons.
Thank you for these important life lessons!
On the contrary, that we do not yet have General Intelligence is because current AI requires an objective...
You have just brought AI to... life :)
Good stuff!
Thanks
that number 6 is what we must remember.
AI trains human to operate in super human way.. 🤠
Great video! It's nice to also have your personal touch along with the other great informative videos you make.
I totally relate to all those people who said they were inspirer by your work. And dang am I too. Without you I wouldn't have discovered half of what I know about AI, so much inspiring!
Thank you for your hard work, and keep on going, I don't think I'd be so much into AI if it weren't for you :)
“And I’ll see you next time” very cool channel! I especially liked the one with the hide and go seek bots.
And enjoy some time spent, like a kid, playing & exploring new things :) Maybe the fun from novelty is wiser than it might first seem. As a strategy, novelty seeking can help lead to better models. Generally, explore then exploit.
Except when, you know.. you find a tv in a wal and you get stuck in there, forever. l
@2:22 I love that you choose to use "stupendously" instead of the vastly more popular "tremendously"
My respect for you today is just quadrupled 😀 , thank you for the life lessons that were always staring at me when i am working or studying...... Thanks , and thank you for all of your amazing videos.
Love your videos, dude! It's so exciting to see this sort of thing and imagine what's coming in the future! Keep it up bro!
This short video is much more informative than those so-called idea worth sharing videos from ted talks.
This is an amazing episode! Thank you c:
Absolutely loved this!! Thank you for sharing all of this incredible research and insight :)
At first glance I thought AI had taught itself life lessons
This is so creative and inspiring! Thanks you :)
I can appreciate your video's even more together with my progress and experience with data science. Keep up the good work!
Thank you Karoly! I love the mathematics, and most of the subjects (AI and ML especially) you are sharing an information of scientific progress with us. Unfortunately, my random steps didn't bring be to a point, where I can make enough money for living with my knowledge of related sciences. So I have to sacrifice my time for less innovative stuff and I don't have enough time to progress in these fields. But you're my savior, thanks to you, I can enjoy to have knowledge on what is the most recent achievements in these fields. Thank you a lot!
Amazing!
(Like the rest of his videos!)
Simply Amazing :)
Absolutely amazing as always! For more videos like these. 👏
Love the random walk reference.
This new format is actually awesome, great input from people that I follow. Would be great would be to gather this kind of advices from your fellow colleagues or researchers around the world. I guess the channel could grow even bigger. Kudos!
This is amazing thank you.
Kashmira Zambad - 2019-01-15
Summary in two-minutes:
1] You need an objective
2] A change in the objective changes the strategy required to achieve it – change with efficiency
3] If the objective was wrong, do not worry and aim again. – try to define and achieve the new objective, improve the predictor, you will find why the ideas that did not work don’t work
4] Zoom out and evaluate – Phase1: collect experiences, Phase2: experience we play; relearn, recalibrate and reflect.
5] If you find something that works, hold on to it – explore more, the pain will be worth it, seek the light in similar directions.
6] As long as you keep moving, you will be progressing – Random Walk, A mathematical theorem- After N steps, the expected distance from where we’ve started is proportional to the square root of N. That is progress, never stagnate!
I absolutely loved the video! Continue inspiring...
Tamas Konrad - 2020-02-06
Szep munka, gratulalok! Orulok hogy osszekotod a cumputer mukodeset az agy mukodesevel, valojaban sokban hasonlitanak szerintem is, mivel a kozponti resze az eletunknek, abszolut nem mindegy hogy autopilotaban futnak a programok rajta, vagy egy tudatos vezerlo inditja el, allitja meg es (amire felhivod a figyelmet) "updatealja" ezeket a programokat. Uj tudas nelkul nincs haladas a celjaink fele, mert azok altalaban egy szamunkra meg nem felfedezett terben vannak. Mindig tobbre vagyunk, amihez elengedhetetlen az uj tehnika es tudas kifejlesztese. A tanulas az elso vele szuletett adottsagunk, az elet utjan ez celjat vesztheti az informaciodus tarsadalmunkban. De ha sikerul a "tanulast" folyamatat tudatossa tenni mindenki szamara, akkor sok illuziok es tudatlansag altal okozott betegsegtol lehetne megmenteni az emberiseget a jovoben. Mint pl celtalansag, depressio es sok mas mentalis "program hiba" ami a fejbol indul es vegul a fizikaban is megmutatkozik negativ modon.
Sok sikert tovabbra is az ilyen tipusu videokhoz! Feliratkoztam.
SinusoiDave - 2020-02-08
@Charles Brightman that dismal conclusion belies your sanguine last name 🤣🤣🤣
Charles Brightman - 2020-02-08
@SinusoiDave What part of the above do you believe to be wrong?
AJ12Gamer - 2020-03-26
Tldr: Never give up and keep going.
Chris Durias - 2020-07-29
7] If you practice you will improve.