Learning with examples first is always better than starting with math, Now when I read the math behind RBM it makes more sense since I have something to relate too! Thank you for this wonderful presentation :)
I just have some doubt about some possible scenarios that you explained at the beginning of the video. For example, in your explanations, you explained that the score for the scenario BD is equal to -1. But, then in the table, we see that the score for this scenario is not -1, instead, it is -2. And this inconsistency happens for some other scenarios as well. Thank you very much for your clear explanations.
This clip clarified a lot about RBMs for me. Perfect combination of simple examples and the math behind it which makes it easy to understand. I don't get why literature on this or similar topics aimed to teach has to only focus on the math part/formulas which is much more time consuming to understand and in the end doesn't provide as good intuition as your approach of teaching a subject. Thanks for the video!
Thank you so much for your kind contribution!! Definitely, I’ll continue making videos, if you have suggestions for topics, please let me know. Cheers!
Imagine paying thousands to university to teach me RBM and then finding a RU-vid video that explains it 100 times better then the professor in half the time. Couldn't be me
HAHAHA, coming in here from the video about Generative Adversarial Network and realize need to understand this concept in order to understand GAN, the recc Algo really guessed my thoughts right...
Thank you for this! The only problem is I don't know how you can get Descartes and Euler showing up at the same time if you use your sampling algorithm
Thanks, great question! The probability of that would be really low, since the weight for any of those configurations with Descartes and Euler would have a very low weight (as some negative weights are forced to appear). But if you run it many times, you may run into that configuration.
Very good video! The sound quality should be improved! the dog and cat belongs to the hidden layer! The persons mentioned have no knowledge of which pet is in the house. It is not clear to me what is the input and output of this neuronal net.
Great video! A question: when you pick a sample to increase scores, you increase all the nodes (and edges) that consist this sample. However this not only increase the probability with this particular sample but also other combinations. This seems contradict to the Gibbs sampling method. Is there something I misunderstood, or this is correct but we just tolerate this side effect as it still does the job?
Great question! You understood it correctly. As you improve the scores for a sample, it may affect the scores for other samples (this problem happens in most other ML algorithms too). The hope is that if you do this for all the samples, it starts capturing the form of the data.
Sorry, but I still didn't get this point. In your particular example, you only can see the visible layer. So, when Beto shows up, you'll end up increasing BD OR BE, cause you don't see the hidden layer (cat or dog). Besides, when other samples comes (Alisha with Cameron) it doesn't give you any compensation that increase BE and decrease BD. So, in the end, you should increase all combination related to samples (and not only those truly related with visible and hidden layers). But this is a toy example. Maybe, in real data, this relations are more complex and then such compensation truly happen. Or maybe I'm still misunderstanding something.
Hi and thanks a lot for this video! It is an excellent source to get into RBMs. I have a question regarding the energy function, though. I always see this particualr form of the energy and never a derivation or an argument for it. Is it the only possible form? Or can one come up with any sort of functions, as long as the sum over all possible configurations doesn't lead to zero ( so that we can still divide by the partition function )?
Great question! I don't know this part very well, but if the particles have spins that align or not align, there are different energies. Look for Ising Model, and that describes it better.
Hi Lalit! sorry for the super late reply. Here is my favorite quantum computing course: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-VPsl_5RQe1A.html And this one for quantum machine learning: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-QtWCmO_KIlg.html Enjoy!
Not knowing the animals exist, then Occam's razor applies: Aisha and Cameron are having an affair. Since on of them is married, they lie about knowing each other. =)