3 ай бұрын
2 жыл бұрын
3 жыл бұрын
3 жыл бұрын
3 жыл бұрын
4 жыл бұрын
4 жыл бұрын
4 жыл бұрын
5 жыл бұрын
Пікірлер
@anangelsdiaries 13 күн бұрын
This video was beautiful
@lisasamir3773 19 күн бұрын
best video on autoencoders
@jasonfunderberker 20 күн бұрын
could only like once
@huehuecoyotl2 Ай бұрын
This makes me think the Loop Quantum Gravity theory of quantum gravity which features quanta of gravity as graphs labeled as spin networks.
@KennethThomas-c1w Ай бұрын
5051 Jasmin Skyway
@perivarfriborg3916 Ай бұрын
Thank you!
@RuralLove Ай бұрын
Thank you
@ShNazem Ай бұрын
Amazing explaination Thank you
@haepari11 Ай бұрын
Beautiful explanation!
@fanzhou7193 2 ай бұрын
I can't imagine this was published three years ago
@joe_hoeller_chicago 2 ай бұрын
GitHub or it never happened.
@ronaldlegere 2 ай бұрын
Love this series . You mention a book you are referencing around 2:15 , what book is that ?
@welcomeaioverlords 2 ай бұрын
Swarm Intelligence: From Natural to Artificial Systems (Santa Fe Institute Studies on the Sciences of Complexity)
@davidinawe791 2 ай бұрын
Very nice to gain an intuition, thanks!
@anushavarusknmmlaushsbj 2 ай бұрын
With the random initialization of alpha and beta values from a distribution to each ant, the algorithm can result in optimal solutions sometimes and sometimes with not very optimal solution. So, how can we tackle this situation? Run the algorithm multiple times and pick up the best?
@welcomeaioverlords 2 ай бұрын
Since the alpha/beta parameters are changing from iteration to iteration, I guess the assumption is that the evo process will find optimal parameters, if they exist. But it's also important to make the distinction that an ant having optimal alpha/beta does not necessarily mean they will find the optimal TSP solution, because there's also randomness in constructing the path. But to answer your question more pragmatically, I think you can just run for more iterations instead of stopping and restarting with new samples, because the sampling procedure only sets the initial values. If the quality (over a large number of iterations) is sensitive to initial conditions, that's a problem for the algorithm.
@jimlbeaver 2 ай бұрын
I love your intuition about the bifurcation. Sometimes that’s hard to get it to stabilize in a dynamic system. One way to look at it as something like “specialization”. Obviously,complex systems like bees, ants, and humans specialize (workers, foragers, bakers, etc). But I have not seen a good model for how to specialize within a generation. I have seen people use the concept of “speciation” within evolutionary algorithms that allow sub species to separately evolve to work in a dynamic system. Sometimes this can keep a dominant population from overwhelming a group that is good at exploration for example. If it sounds interesting check it out Otherwise, keep going really great series! Good luck! (BTW, I agree, the Santa Fe Institute is a great resource for this type of research)
@rishidixit7939 2 ай бұрын
What is Agent Based Modeling ? Is it same as AI Agents ?
@welcomeaioverlords 2 ай бұрын
It's basically using "agents" (i.e., decision making entities) in a simulation. It's a test bed to see how interactions among agents creates interesting outcomes. I think/hope that there will be natural applications to AI Agents as we progress, and that's likely where I'm headed.
@jimlbeaver 2 ай бұрын
Cool idea to add evolution. Are you trying to learn/evaluate Mesa or are you looking for a novel solution to TSP?
@welcomeaioverlords 2 ай бұрын
I'm just playing with the algorithms right now and using Mesa/TSP as a quick first step to get up and running. I hope to use more sophisticated agents over time and solve different problems. But I also don't have any firm plans--just following my curiosity.
@fenglongsong4760 2 ай бұрын
Thank you for your video. You truly saved my day
@TheKrasTel 3 ай бұрын
Thank you for recording these, i love that you explain the ideas behind every thing you do, helps to understand your thought process - very helpful!
@welcomeaioverlords 3 ай бұрын
Great to hear, thanks!
@ronaldlegere 3 ай бұрын
Unless you need a digraph later, you could have stuck with an undirected graph. In networkX, in an undirected Graph g, g[u][v] is identical to g[v][u] .
@welcomeaioverlords 3 ай бұрын
Good to know! In this case, I think I’ll keep as is so I can generalize to asymmetric TSP problems.
@jimlbeaver 3 ай бұрын
This is an amazing algorithm that has seem some good results in network traffic. Really great that you put these things together. Nice work…hope you keep going.
@welcomeaioverlords 3 ай бұрын
Thank you!
@Balajik7-qh1pq 3 ай бұрын
Good try Zak
@welcomeaioverlords 3 ай бұрын
Thank you?
@ram_c 3 ай бұрын
how to learn these GNN for physics simulation? any tutorial?
@Acceleratedpayloads 3 ай бұрын
Will you consider doing a prisoners dilemma experiment using LLM?
@welcomeaioverlords 3 ай бұрын
I think there's already work done on this: arxiv.org/abs/2305.07970; arxiv.org/abs/2305.16867. Once I get simple things to work, I'll move to more advanced agents. I'm ultimately interested in applying evolutionary algorithms to populations of agents (which means they have to be small enough for me to handle) and having them solve tasks that require coordination. Prisoner's Dilemma might help benchmark cooperative behavior?
@Acceleratedpayloads 3 ай бұрын
Yoooo I've been sharing your GCN videos with friends and colleagues for a while now, all of a sudden you drop 3 videos at once? Hell yeah
@projectpiano5231 3 ай бұрын
Yooo I just started my first RL project a few days ago after neglecting to learn about policy gradient learning and Q-learning for a long time. This could not have been better timing xD I'm implementing PUCT and I'm mostly done which means I may want to switch to a library now that I've almost written it from scratch. Also thanks for the video! I watched the brain-computer interface video too and it was really interesting. I wonder if they use predictive coding and other theories in informing how they model brain activity. To me it sounds like the brain tries to minimize surprise/loss locally in the neocortex (pretty much pattern recognition part of brain) and has reinforcement learning in the striatum and ventral tegmental area (reward center of the brain) but as far as I'm aware there isn't any theory that captures all of that and is biologically plausible
@welcomeaioverlords 3 ай бұрын
Good luck on the project and glad to hear you’re finding something of interest here!
@MachineLearningStreetTalk 3 ай бұрын
Hey guys! Great to see you both again
@welcomeaioverlords 3 ай бұрын
Thanks! Breaking that two year no-upload streak
@konstantinosbarmpas 3 ай бұрын
Thank ! Hope you’ve enjoyed our conversation 😄
@Hoi_music 3 ай бұрын
This was such a great video!!
@artem_isakow 3 ай бұрын
Thank you!
@henrygengiti7861 4 ай бұрын
can you please clarify sigma one more time?
@spandanpadhi8275 4 ай бұрын
This was the best 12 minutes of my months. Great explanation of GANs.
@朱蔚健 4 ай бұрын
Very good instructional video, even over the wall to see the video, for the autoencoder explained in depth, so that I after a good understanding of the GAN-AAE!
@johnnyBrwn 4 ай бұрын
Thanks man this makes so much sense
@kathanvakharia 4 ай бұрын
nailed it!
@ishara779 5 ай бұрын
Anybody told you you look exactly like Chad Michael Murray from One Tree Hill? The resemblance is remarkable
@Only_for_Harbinger_hackathon 6 ай бұрын
Awesome 👍
@sm-pz8er 6 ай бұрын
That's very informative and well explained video. Great job and thank you.
@spicemasterii6775 6 ай бұрын
Are you and 3 Blue 1 Brown the same person?
@gibsonemg 6 ай бұрын
No, Grant Sanderson is far more prolific and talented :)
@bikrammajhi3020 6 ай бұрын
Best mathematical explanation on GAN on the internet so far
@ingenuity8886 6 ай бұрын
Can you send the intro of this video by somehow , I really liked the choice of music and the transitions you made.
@gibsonemg 6 ай бұрын
The song is "Legend" by the band Elder.
@akashkarnatak6581 6 ай бұрын
Good to see that you are doing your part to bring Roko's Basilisk into existence with your KZbin channel.
@breathemath4757 6 ай бұрын
I don't think any other explanation can do better than this. Thanks a lot!
@HandokoSupeno 6 ай бұрын
Best straight-forward explanation i've ever seen. Thanks
@passionatechristianworker 6 ай бұрын
very helpful, thank you !
@KisanThapaYT 6 ай бұрын
Bro gave me a great intuitive and clear explanation without using fancy pictures. Greatest RGCN explanation ever.
@JaishreeramCoder 6 ай бұрын
amazingly explained
@MarianneArriola 6 ай бұрын
❤🔥
@tomoki-v6o 6 ай бұрын
I preffer cocatenating feature vectures and apply the weighted sum. in the agregation phase. averaging to me is a crime
@EstelFerrerTorres 6 ай бұрын
Error in Min. 2:53 --> alpha u,v expression --> denominator exponent should be a(hk, hv)
@ariangh2560 7 ай бұрын
is there any article about this?