SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.06917 · ICLR · 2021

Is Disentanglement all you need? Comparing Concept-based & Disentanglement Approaches

Pietro Liò, Dmitry Kazhdan, Adrian Weller, Mateja Jamnik, Botty Dimanov, Helena Terre

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 19 functions out of this paper's own repositories and ran 10 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
dmitrykazhdan/concept-based-xai canonical 10 of 19
FunctionStatusWhere it lives
built_task_fn Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/latentFactorData.py
code served (permissive licence) · get_code("f859bf3dd114e7d7")
cardinality_encoding Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/dSprites.py
code served (permissive licence) · get_code("b19d09ea8a8e0374")
compute_activation_per_layer Ran dmitrykazhdan/concept-based-xai/concepts_xai/methods/CME/ItCModel.py
code served (permissive licence) · get_code("b3d3a0d10f9d6c5f")
flatten_activations Ran dmitrykazhdan/concept-based-xai/concepts_xai/methods/CME/ItCModel.py
code served (permissive licence) · get_code("d9c290aa08862671")
get_latent_bases Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/dataset_utils.py
code served (permissive licence) · get_code("0b50f14e6d4b1ad7")
get_task_data Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/latentFactorData.py
code served (permissive licence) · get_code("1fa117d288391f24")
latent_to_index Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/dataset_utils.py
code served (permissive licence) · get_code("681a89569c3f3cb2")
load_dataset_paths Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/load_paths.py
code served (permissive licence) · get_code("ed967be24bf64400")
small_skip_ranges_filter_fn Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/dSprites.py
code served (permissive licence) · get_code("cf5d142e3d671d42")
small_skip_ranges_filter_fn Ran dmitrykazhdan/concept-based-xai/concepts_xai/datasets/shapes3D.py
code served (permissive licence) · get_code("72718db40bbd6986")
get_all_concepts Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/cars3D.py
code served (permissive licence) · get_code("794a85c6f0e9d261")
get_category_full Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/smallNorb.py
code served (permissive licence) · get_code("baf0160539f306cc")
get_elevation_full Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/cars3D.py
code served (permissive licence) · get_code("db1bb50b81c139cb")
get_model Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/utils/model_loader.py
code served (permissive licence) · get_code("58f491becc62bed0")
get_shape_full Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/dSprites.py
code served (permissive licence) · get_code("17598bb41881ef43")
get_shape_full Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/shapes3D.py
code served (permissive licence) · get_code("54b1d202a35ed398")
get_shape_small_skip Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/shapes3D.py
code served (permissive licence) · get_code("1e0aa241637e9968")
load_mesh Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/datasets/cars3D.py
code served (permissive licence) · get_code("bca977322866f5f2")
produce_bottleneck Not yet run dmitrykazhdan/concept-based-xai/concepts_xai/methods/CBM/CBModel.py
code served (permissive licence) · get_code("7bde78aa9a7f31ce")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Concept-based explanations have emerged as a popular way of extracting humaninterpretable representations from deep discriminative models. At the same time, the disentanglement learning literature has focused on extracting similar representations in an unsupervised or weakly-supervised way, using deep generative models. Despite the overlapping goals and potential synergies, to our knowledge, there has not yet been a systematic comparison of the limitations and trade-offs between concept-based explanations and disentanglement approaches. In this paper, we give an overview of these fields, comparing and contrasting their properties and behaviours on a diverse set of tasks, and highlighting their potential strengths and limitations. In particular, we demonstrate that state-of-the-art approaches from both classes can be data inefficient, sensitive to the specific nature of the classification/regression task, or sensitive to the employed concept representation.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2104.06917")
get_code_for_paper("2104.06917")
have("2104.06917")

Connect an agent — have() is free.