Work Done
- Loaded in CIFAR-10 data
- ran it through MNIST + the first sae
- cached all the activations
- found max activating images
Confusions
- struggling to understand does the dataset havfe a constat image that it pulls from or is it randomized every time?
- need to validate that the images are not random or that I can control them
- how do I actually visualize the clean image from CIFAR-10?
Next Steps
- Visualize the images [x]
- Visualize the images with ImageNet or another image set that has way more diverse images
- Repeat the process for the meta-sae
- utilize VLM to do labeling (maybe CLIP?)
- go really meta and have a loop that continously constructs meta-saes - trying to find if there is a point at which you cannot get more fine-grained
Cool Stuff
- Hugo Fry's experiments with Vision SAEs
- Hugo Fry et al. work with medical saes : Ranjani could be interested in this