Merge pull request #58 from AlexImmer/subnetlaplace

Subnetwork Laplace

README.md

@@ -4,17 +4,17 @@
 
 [![Main](https://travis-ci.com/AlexImmer/Laplace.svg?token=rpuRxEjQS6cCZi7ptL9y&branch=main)](https://travis-ci.com/AlexImmer/Laplace)
 
-The laplace package facilitates the application of Laplace approximations for entire neural networks or just their last layer.
+The laplace package facilitates the application of Laplace approximations for entire neural networks, subnetworks of neural networks, or just their last layer.
 The package enables posterior approximations, marginal-likelihood estimation, and various posterior predictive computations.
 The library documentation is available at [https://aleximmer.github.io/Laplace](https://aleximmer.github.io/Laplace).
 
 There is also a corresponding paper, [*Laplace Redux — Effortless Bayesian Deep Learning*](https://arxiv.org/abs/2106.14806), which introduces the library, provides an introduction to the Laplace approximation, reviews its use in deep learning, and empirically demonstrates its versatility and competitiveness. Please consider referring to the paper when using our library:
 ```bibtex
-@article{daxberger2021laplace,
-  title={Laplace Redux--Effortless Bayesian Deep Learning},
-  author={Daxberger, Erik and Kristiadi, Agustinus and Immer, Alexander
-          and Eschenhagen, Runa and Bauer, Matthias and Hennig, Philipp},
-  journal={arXiv preprint arXiv:2106.14806},
+@inproceedings{laplace2021,
+  title={Laplace Redux--Effortless {B}ayesian Deep Learning},
+  author={Erik Daxberger and Agustinus Kristiadi and Alexander Immer 
+          and Runa Eschenhagen and Matthias Bauer and Philipp Hennig},
+  booktitle={{N}eur{IPS}},
   year={2021}
 }
 ```
@@ -39,18 +39,24 @@ pytest tests/
 ## Structure
 The laplace package consists of two main components:
 
-1. The subclasses of [`laplace.BaseLaplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/baselaplace.py) that implement different sparsity structures: different subsets of weights (`'all'` and `'last_layer'`) and different structures of the Hessian approximation (`'full'`, `'kron'`, and `'diag'`). This results in six currently available options: `laplace.FullLaplace`, `laplace.KronLaplace`, `laplace.DiagLaplace`, and the corresponding last-layer variations `laplace.FullLLLaplace`, `laplace.KronLLLaplace`,  and `laplace.DiagLLLaplace`, which are all subclasses of [`laplace.LLLaplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/lllaplace.py). All of these can be conveniently accessed via the [`laplace.Laplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/laplace.py) function.
+1. The subclasses of [`laplace.BaseLaplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/baselaplace.py) that implement different sparsity structures: different subsets of weights (`'all'`, `'subnetwork'` and `'last_layer'`) and different structures of the Hessian approximation (`'full'`, `'kron'`, `'lowrank'` and `'diag'`). This results in _eight_ currently available options: `laplace.FullLaplace`, `laplace.KronLaplace`, `laplace.DiagLaplace`, the corresponding last-layer variations `laplace.FullLLLaplace`, `laplace.KronLLLaplace`,  and `laplace.DiagLLLaplace` (which are all subclasses of [`laplace.LLLaplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/lllaplace.py)), [`laplace.SubnetLaplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/subnetlaplace.py) (which only supports a `'full'` Hessian approximation) and `laplace.LowRankLaplace` (which only supports inference over `'all'` weights). All of these can be conveniently accessed via the [`laplace.Laplace`](https://github.com/AlexImmer/Laplace/blob/main/laplace/laplace.py) function.
 2. The backends in [`laplace.curvature`](https://github.com/AlexImmer/Laplace/blob/main/laplace/curvature/) which provide access to Hessian approximations of
 the corresponding sparsity structures, for example, the diagonal GGN.
 
 Additionally, the package provides utilities for
-decomposing a neural network into feature extractor and last layer for `LLLaplace` subclasses ([`laplace.feature_extractor`](https://github.com/AlexImmer/Laplace/blob/main/laplace/feature_extractor.py))
+decomposing a neural network into feature extractor and last layer for `LLLaplace` subclasses ([`laplace.utils.feature_extractor`](https://github.com/AlexImmer/Laplace/blob/main/laplace/utils/feature_extractor.py))
 and
-effectively dealing with Kronecker factors ([`laplace.matrix`](https://github.com/AlexImmer/Laplace/blob/main/laplace/matrix.py)).
+effectively dealing with Kronecker factors ([`laplace.utils.matrix`](https://github.com/AlexImmer/Laplace/blob/main/laplace/utils/matrix.py)).
+
+Finally, the package implements several options to select/specify a subnetwork for `SubnetLaplace` (as subclasses of [`laplace.utils.subnetmask.SubnetMask`](https://github.com/AlexImmer/Laplace/blob/main/laplace/utils/subnetmask.py)).
+Automatic subnetwork selection strategies include: uniformly at random (`laplace.utils.subnetmask.RandomSubnetMask`), by largest parameter magnitudes (`LargestMagnitudeSubnetMask`), and by largest marginal parameter variances (`LargestVarianceDiagLaplaceSubnetMask` and `LargestVarianceSWAGSubnetMask`).
+In addition to that, subnetworks can also be specified manually, by listing the names of either the model parameters (`ParamNameSubnetMask`) or modules (`ModuleNameSubnetMask`) to perform Laplace inference over.
 
 ## Extendability
 To extend the laplace package, new `BaseLaplace` subclasses can be designed, for example,
-a block-diagonal structure or subset-of-weights Laplace.
+Laplace with a block-diagonal Hessian structure.
+One can also implement custom subnetwork selection strategies as new subclasses of `SubnetMask`.
+
 Alternatively, extending or integrating backends (subclasses of [`curvature.curvature`](https://github.com/AlexImmer/Laplace/blob/main/laplace/curvature/curvature.py)) allows to provide different Hessian
 approximations to the Laplace approximations.
 For example, currently the [`curvature.BackPackInterface`](https://github.com/AlexImmer/Laplace/blob/main/laplace/curvature/backpack.py) based on [BackPACK](https://github.com/f-dangel/backpack/) and [`curvature.AsdlInterface`](https://github.com/AlexImmer/Laplace/blob/main/laplace/curvature/asdl.py) based on [ASDL](https://github.com/kazukiosawa/asdfghjkl) are available.
@@ -60,18 +66,19 @@ for a regression (MSELoss) loss function.
 
 ## Example usage
 
-### *Post-hoc* prior precision tuning of last-layer LA
+### *Post-hoc* prior precision tuning of diagonal LA
 
 In the following example, a pre-trained model is loaded,
-then the Laplace approximation is fit to the training data,
+then the Laplace approximation is fit to the training data
+(using a diagonal Hessian approximation over all parameters),
 and the prior precision is optimized with cross-validation `'CV'`.
 After that, the resulting LA is used for prediction with
 the `'probit'` predictive for classification.
 
 ```python
 from laplace import Laplace
 
-# pre-trained model
+# Pre-trained model
 model = load_map_model()  
 
 # User-specified LA flavor
@@ -87,7 +94,7 @@ pred = la(x, link_approx='probit')
 
 ### Differentiating the log marginal likelihood w.r.t. hyperparameters
 
-The marginal likelihood can be used for model selection and is differentiable
+The marginal likelihood can be used for model selection [10] and is differentiable
 for continuous hyperparameters like the prior precision or observation noise.
 Here, we fit the library default, KFAC last-layer LA and differentiate
 the log marginal likelihood.
@@ -107,6 +114,45 @@ ml = la.log_marginal_likelihood(prior_prec, obs_noise)
 ml.backward()
 ```
 
+### Applying the LA over only a subset of the model parameters
+
+This example shows how to fit the Laplace approximation over only
+a subnetwork within a neural network (while keeping all other parameters
+fixed at their MAP estimates), as proposed in [11]. It also exemplifies
+different ways to specify the subnetwork to perform inference over.
+
+```python
+from laplace import Laplace
+
+# Pre-trained model
+model = load_model()
+
+# Examples of different ways to specify the subnetwork
+# via indices of the vectorized model parameters
+#
+# Example 1: select the 128 parameters with the largest magnitude
+from laplace.utils import LargestMagnitudeSubnetMask
+subnetwork_mask = LargestMagnitudeSubnetMask(model, n_params_subnet=128)
+subnetwork_indices = subnetwork_mask.select()
+
+# Example 2: specify the layers that define the subnetwork
+from laplace.utils import ModuleNameSubnetMask
+subnetwork_mask = ModuleNameSubnetMask(model, module_names=['layer.1', 'layer.3'])
+subnetwork_mask.select()
+subnetwork_indices = subnetwork_mask.indices
+
+# Example 3: manually define the subnetwork via custom subnetwork indices
+import torch
+subnetwork_indices = torch.tensor([0, 4, 11, 42, 123, 2021])
+
+# Define and fit subnetwork LA using the specified subnetwork indices
+la = Laplace(model, 'classification',
+             subset_of_weights='subnetwork',
+             hessian_structure='full',
+             subnetwork_indices=subnetwork_indices)
+la.fit(train_loader)
+```
+
 ## Documentation
 
 The documentation is available [here](https://aleximmer.github.io/Laplace) or can be generated and/or viewed locally:
@@ -122,7 +168,7 @@ pdoc --http 0.0.0.0:8080 laplace --template-dir template
 
 ## References
 
-This package relies on various improvements to the Laplace approximation for neural networks, which was originally due to MacKay [1].
+This package relies on various improvements to the Laplace approximation for neural networks, which was originally due to MacKay [1]. Please consider citing the respective papers if you use any of their proposed methods via our laplace library.
 
 - [1] MacKay, DJC. [*A Practical Bayesian Framework for Backpropagation Networks*](https://authors.library.caltech.edu/13793/). Neural Computation 1992.
 - [2] Gibbs, M. N. [*Bayesian Gaussian Processes for Regression and Classification*](https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.147.1130&rep=rep1&type=pdf). PhD Thesis 1997.
@@ -132,4 +178,6 @@ This package relies on various improvements to the Laplace approximation for neu
 - [6] Khan, M. E., Immer, A., Abedi, E., Korzepa, M. [*Approximate Inference Turns Deep Networks into Gaussian Processes*](https://arxiv.org/abs/1906.01930). NeurIPS 2019.
 - [7] Kristiadi, A., Hein, M., Hennig, P. [*Being Bayesian, Even Just a Bit, Fixes Overconfidence in ReLU Networks*](https://arxiv.org/abs/2002.10118). ICML 2020.
 - [8] Immer, A., Korzepa, M., Bauer, M. [*Improving predictions of Bayesian neural nets via local linearization*](https://arxiv.org/abs/2008.08400). AISTATS 2021.
-- [9] Immer, A., Bauer, M., Fortuin, V., Rätsch, G., Khan, EM. [*Scalable Marginal Likelihood Estimation for Model Selection in Deep Learning*](https://arxiv.org/abs/2104.04975). ICML 2021.
+- [9] Sharma, A., Azizan, N., Pavone, M. [*Sketching Curvature for Efficient Out-of-Distribution Detection for Deep Neural Networks*](https://arxiv.org/abs/2102.12567). UAI 2021.
+- [10] Immer, A., Bauer, M., Fortuin, V., Rätsch, G., Khan, EM. [*Scalable Marginal Likelihood Estimation for Model Selection in Deep Learning*](https://arxiv.org/abs/2104.04975). ICML 2021.
+- [11] Daxberger, E., Nalisnick, E., Allingham, JU., Antorán, J., Hernández-Lobato, JM. [*Bayesian Deep Learning via Subnetwork Inference*](https://arxiv.org/abs/2010.14689). ICML 2021.