Vinith Suriyakumar

How to compute the empirical after kernel?

1

I'm looking to use the library to compute the after kernel for a model trained with the FLAX library? I followed this Colab: https://colab.research.google.com/github/google/neural-tangents/blob/main/notebooks/empirical_ntk_resnet.ipynb. Instead of these lines: ``` #...

Hyperparameters Used to Get Models on OneDrive

1

Hi, I'm looking to reproduce the performance of the celebrity 1,5,10,100 erasure models in the paper and the ones on OneDrive. I was wondering what the hyperparameters used to train...

Computing FID for Celebrity Experiments

1

Hi, I am looking to replicate the results from the celebrity experiments in the paper using this repo. I'm a bit confused as to what dataset is used to compare...

Error with Training

2

Hi, I'm trying to run the training script as described in the README but am encountering this error? `TypeError: Linear.forward() got an unexpected keyword argument 'scale'` Thanks, Vinith

Retain Set Used for UCE Baselines

3

Hi, I was wondering what retain set (prior preservation set) was used for the UCE baselines in each experiment? Thanks! Best, Vinith

Training Error For Erase 100 Art Styles

1

Hi, I believe there is an error in the repo such that it is not possible to replicate the erase 100 art experiments. Currently, in Line 64 of `cfr_lora_training.py` there...

Captions Used for Caches

1

Hi, I was wondering what captions were used to create the celebrity and art prior knowledge caches? I would like to run experiments for SD2.1 but the current caches provided...

Poor Performance when Applying MACE to SD2.1

1

Hi, I tried using the code provided to apply MACE to SD2.1 and the models are simply not working. I was wondering what hyperparameters worked well? Thanks, Vinith