Видео

why would the attention adjacency matrix be symmetrical? If the weight vector is learnable, then it does matter which order the two input vectors are concatenated. It doesn't seem like there would be any reason to enforce symmetry.

it was the best explanation that gave me hope for the understanding these mechanisms. Everything was so good explained and depicted, thank you!

Thanks for this amazing presentation! This topic which is not trivial is also not easy to show in pictures and you succeeded perfectly. Great help!

Ist almost as if its just a normal neural network but projected onto a graph

Well-organized video and description, abundant references. I love this series. Cheer up!

I really appreciate the information that you shared in this video/playlist. Do you have an example of where you used used the Heterogeneous graph data to create a GNN or GCN?

mathematician reeee-man

What is this for? I am intrigued ( im a musician)

I have trouble finetuning the model?? has anyone managed to recreate the results from the paper?

Bro u killed it. Best explanation. Trust me I have watched all tutorials but all other explanations were shitty. Please create one video on quantization.

AMAZING!

This might be the best and simple explanation of GAT one can ever find! Thanks man

Why use dropout with GeLU? Didn’t the GeLU paper specifically say one motivation for GeLU was to replace ReLU+dropout with a single GeLU layer?

its just awesomee... the way u do research before making a video is reallyyyy fascinating. can you tell how u collect papers which are relevant, it will help a lot while doing projects on our own

you make the best explanation videos everrrr! is there one on how to train custom models using LORA?

Did not get :( moving to other videos

Banger vid from big dog as always🐶

Collaborative Filtering is such an underrated computation method, imho Many real life problems can be re-interpteted as a recommendation problem. It's just a matter of perspective, but the results can be huge!

Ah, tough to understand, guess will have to read more on this to fully understand

It's a good, concise walk-through with good code implementation examples. However I'd recommend avoiding some ambiguous variable names in code like betas, Block, etc.

The best explanation i found in internet

A wonderful and succinct explanation with crisp visualisations about both the attention mechanism and the graph neural network. The way the learnable parameters are highlighted along with the intuition (such as a weighted adjacency matrix) and the corresponding matrix operations is very well done.

Hello great video. I was just wondering about something you mentioned in the video. At time 14:19 you say that graph VAE's might not be the best architecture for graph generation, and I was wondering what other models do you recommend that might be better than GVAE's. I am asking this because I am working on the exact thing you are working on just for another disease, so I was wondering if you have any recommendations to make the process a lot simpler.

perfect video

Fantastic explanation. You made it look easy!

bro is educated!

This is a perfect explanation for LoRA I found so far !!!

What's the similarity between NTXentLoss (or InfoNCE) and SimCLR loss?

Can I predict next 12 timesteps but more than one target feature with this model?

Very good explanation

This video is pure gold. As someone coming from linguistics and trying to use MDS in research, I really appreciate it!

1:00

Highly appreciate the effort. I really like how you started with the theoretical part and then worked on a real dataset. Thanks!

The best GNN explanation video also answered many questions of mine

Bravo!

I think the lora is scaled by the square root of the rank, not the rank.

Would this also work on large graphs?

Very Nice Explanations Sir. GNN clearly explained very well for beginners like me. Thank You so much

Sehr gut erklärt und umgesetzt 👏

For some reason I have only 15 million parameters, instead of 17 million. I double checked everything. What could be the reason? Could it simple be because of the latest versions that I use?

Ahh thats nice one I found! Can you explain how GNN can be linked to topological data analysis?

stellargraph is another good library too

After watching many LoRA videos, this one finally makes me satisfied. I have a question: I see in the fine tuning code, they talk about merging lora adapters. What is that? Is this h + = x @ (W_A @ W_B) * alpha ? Can you mix and match adapters to improve the evaluation score?

thanks

Isn't the embedding layer redundant? I mean we have then the projection matrices meaning that embedding + projection is a composition of two linear layers.

(rome is bigger than nyc)

Keep making content like this, I am sure you will get a very good recognition in the future. Thanks for such amazing content.

best

A great video as usual. Very detailed and comprehensive 😊. Only one thing left me confused: Why isn’t it a problem to making use of Euclidian distance in t-SNE and Umap? You could have assumed they skip it completely.

thank u a lot for all ur videos :D can u do one about graphsage ?