PCL

PCL copied to clipboard

Single GPU

2

I can only use a single GPU. Is it possible to modify your code to run on a single GPU? If so, could you tell me where I should modify...

Hepta-Col

Question about eq(9) in your paper.

12

Hi, thanks for your paper and code. I have a question about eq(9) in your paper, it seems that this eq is p(ci|xi), not p(xi|ci). I think p(xi|ci) includes only...

randydkx

Questioning about the cluster method?

1

I want to know that have you considered other cluster methods because KNN is unsuitable for some high-dimension and discrete data. I believe this framework has the potential to change...

Odysseussambition

question about concentration around a prototype

2

In the paper, you have mentioned "With the proposed φ, the similarity in a loose cluster (larger φ) are down-scaled, pulling embeddings closer to the prototype", but i am wondering...

vmmm123

About Loss of InfoNCE and Cluter_results

2

Hi, 1. I notice that the labels created in InfoNCE loss is always a zero-vector:(https://github.com/salesforce/PCL/blob/964da1fb7c0546e8ce55627fa3c0debde4b7e456/pcl/builder.py#L163) I think this is wrong since otherwise the loss will always be zero. Did I...

lsyysl9711

Why the term about Q(ci) in eqn.(4) in the paper is a constant?

1

I don't understand that you have Q(ci) = p(ci;xi, \theta), then why \sigma Q(ci) log Q(ci) is a constant? Looking forward to your reply.

Tristan-Liu-13

Hi ! Could I ask how do you design the self-adaptive temperature parameter, especially the denominator Z log(Z + α). I don't quite know why there exists the "log(Z +...

Hugo-cell111

Hello, I'm trying to apply this PCL technique to CIFAR-10 since I don't directly have widespread gpu access. After modifying it to run on the cpu and mps for pytorch,...

kroegern1