Agustinus Kristiadi

Results 37 issues of Agustinus Kristiadi

Hi, I noticed in the new version of the paper, you mentioned 1-(0.1 N) scoring and I'd like to try that. I didn't see any parameter for this in the...

question

**Useful for:** Users who want to implement custom predictive approximations. **Issue:** Currently, the predictive approximation is tightly coupled with the Laplace class. So, if the user wanted to implement a...

question

I'm trying to use ResNet50 from `keras.applications` as a feature extractor for my model. I successfully accomplished that with this: ``` python inputs = Input(shape=(224, 224, 3)) base_model = ResNet50(include_top=False,...

Main features of this pull request: 1. Support only doing Laplace on params that require grad. Use case: PEFT (like LoRA) on top of a frozen foundation model. This is...

enhancement

PR for #138. Very useful for LLMs or diffusion models or any models with many outputs. TODO: Implementation for `KronLLLaplace`. I'd like input from @aleximmer, who's the author of `matrix.py`.

enhancement

Currently, we compute the Jacobians explicitly. We can improve this by using VJPs. Reference for full GGN: https://github.com/f-dangel/curvlinops/blob/5852711aedf2728bc609fabfa95eac00da1beb63/curvlinops/examples/functorch.py#L72-L138 Not a high priority since KFAC is usually used and (diag/full) EF...

enhancement

Based on @aleximmer's paper: https://openreview.net/forum?id=A6EquH0enk Would be very useful e.g. in Bayesian optimization with (heteroscedastic) noisy observation---super relevant in AI4Science.

enhancement

@AlexImmer @runame BackPACK has just been updated to support BCE loss [1]. We should add the support. Not sure if ASDL supports this, though. If it doesn't, how do we...

enhancement

@AlexImmer, @runame, @edaxberger: As you know, I'm currently working on last-layer Laplace for img2img tasks, e.g. autoencoder, image segmentation. We can't use the current implementation in this library mainly due...

Because of the dependency on TensorFlow v1, Python 3.7 is a must. Moreover `SQLAlchemy==1.4.47` is also required (the newest version of SQLAlchemy is v2.x. Probably easiest to provide Conda's `environment.yml`.