---
_id: '14190'
abstract:
- lang: eng
  text: "Learning meaningful and compact representations with disentangled semantic\r\naspects
    is considered to be of key importance in representation learning. Since\r\nreal-world
    data is notoriously costly to collect, many recent state-of-the-art\r\ndisentanglement
    models have heavily relied on synthetic toy data-sets. In this\r\npaper, we propose
    a novel data-set which consists of over one million images of\r\nphysical 3D objects
    with seven factors of variation, such as object color,\r\nshape, size and position.
    In order to be able to control all the factors of\r\nvariation precisely, we built
    an experimental platform where the objects are\r\nbeing moved by a robotic arm.
    In addition, we provide two more datasets which\r\nconsist of simulations of the
    experimental setup. These datasets provide for\r\nthe first time the possibility
    to systematically investigate how well different\r\ndisentanglement methods perform
    on real data in comparison to simulation, and\r\nhow simulated data can be leveraged
    to build better representations of the real\r\nworld. We provide a first experimental
    study of these questions and our results\r\nindicate that learned models transfer
    poorly, but that model and hyperparameter\r\nselection is an effective means of
    transferring information to the real world."
article_processing_charge: No
arxiv: 1
author:
- first_name: Muhammad Waleed
  full_name: Gondal, Muhammad Waleed
  last_name: Gondal
- first_name: Manuel
  full_name: Wüthrich, Manuel
  last_name: Wüthrich
- first_name: Đorđe
  full_name: Miladinović, Đorđe
  last_name: Miladinović
- first_name: Francesco
  full_name: Locatello, Francesco
  id: 26cfd52f-2483-11ee-8040-88983bcc06d4
  last_name: Locatello
  orcid: 0000-0002-4850-0683
- first_name: Martin
  full_name: Breidt, Martin
  last_name: Breidt
- first_name: Valentin
  full_name: Volchkov, Valentin
  last_name: Volchkov
- first_name: Joel
  full_name: Akpo, Joel
  last_name: Akpo
- first_name: Olivier
  full_name: Bachem, Olivier
  last_name: Bachem
- first_name: Bernhard
  full_name: Schölkopf, Bernhard
  last_name: Schölkopf
- first_name: Stefan
  full_name: Bauer, Stefan
  last_name: Bauer
citation:
  ama: 'Gondal MW, Wüthrich M, Miladinović Đ, et al. On the transfer of inductive
    bias from simulation to the real world: a new disentanglement dataset. In: <i>Advances
    in Neural Information Processing Systems</i>. Vol 32. ; 2019.'
  apa: 'Gondal, M. W., Wüthrich, M., Miladinović, Đ., Locatello, F., Breidt, M., Volchkov,
    V., … Bauer, S. (2019). On the transfer of inductive bias from simulation to the
    real world: a new disentanglement dataset. In <i>Advances in Neural Information
    Processing Systems</i> (Vol. 32). Vancouver, Canada.'
  chicago: 'Gondal, Muhammad Waleed, Manuel Wüthrich, Đorđe Miladinović, Francesco
    Locatello, Martin Breidt, Valentin Volchkov, Joel Akpo, Olivier Bachem, Bernhard
    Schölkopf, and Stefan Bauer. “On the Transfer of Inductive Bias from Simulation
    to the Real World: A New Disentanglement Dataset.” In <i>Advances in Neural Information
    Processing Systems</i>, Vol. 32, 2019.'
  ieee: 'M. W. Gondal <i>et al.</i>, “On the transfer of inductive bias from simulation
    to the real world: a new disentanglement dataset,” in <i>Advances in Neural Information
    Processing Systems</i>, Vancouver, Canada, 2019, vol. 32.'
  ista: 'Gondal MW, Wüthrich M, Miladinović Đ, Locatello F, Breidt M, Volchkov V,
    Akpo J, Bachem O, Schölkopf B, Bauer S. 2019. On the transfer of inductive bias
    from simulation to the real world: a new disentanglement dataset. Advances in
    Neural Information Processing Systems. NeurIPS: Neural Information Processing
    Systems vol. 32.'
  mla: 'Gondal, Muhammad Waleed, et al. “On the Transfer of Inductive Bias from Simulation
    to the Real World: A New Disentanglement Dataset.” <i>Advances in Neural Information
    Processing Systems</i>, vol. 32, 2019.'
  short: M.W. Gondal, M. Wüthrich, Đ. Miladinović, F. Locatello, M. Breidt, V. Volchkov,
    J. Akpo, O. Bachem, B. Schölkopf, S. Bauer, in:, Advances in Neural Information
    Processing Systems, 2019.
conference:
  end_date: 2019-12-14
  location: Vancouver, Canada
  name: 'NeurIPS: Neural Information Processing Systems'
  start_date: 2019-12-08
date_created: 2023-08-22T14:09:13Z
date_published: 2019-06-07T00:00:00Z
date_updated: 2023-09-13T09:46:38Z
day: '07'
department:
- _id: FrLo
extern: '1'
external_id:
  arxiv:
  - '1906.03292'
intvolume: '        32'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://arxiv.org/abs/1906.03292
month: '06'
oa: 1
oa_version: Preprint
publication: Advances in Neural Information Processing Systems
publication_identifier:
  isbn:
  - '9781713807933'
publication_status: published
quality_controlled: '1'
status: public
title: 'On the transfer of inductive bias from simulation to the real world: a new
  disentanglement dataset'
type: conference
user_id: c635000d-4b10-11ee-a964-aac5a93f6ac1
volume: 32
year: '2019'
...
---
_id: '14191'
abstract:
- lang: eng
  text: A broad class of convex optimization problems can be formulated as a semidefinite
    program (SDP), minimization of a convex function over the positive-semidefinite
    cone subject to some affine constraints. The majority of classical SDP solvers
    are designed for the deterministic setting where problem data is readily available.
    In this setting, generalized conditional gradient methods (aka Frank-Wolfe-type
    methods) provide scalable solutions by leveraging the so-called linear minimization
    oracle instead of the projection onto the semidefinite cone. Most problems in
    machine learning and modern engineering applications, however, contain some degree
    of stochasticity. In this work, we propose the first conditional-gradient-type
    method for solving stochastic optimization problems under affine constraints.
    Our method guarantees O(k−1/3) convergence rate in expectation on the objective
    residual and O(k−5/12) on the feasibility gap.
article_processing_charge: No
arxiv: 1
author:
- first_name: Francesco
  full_name: Locatello, Francesco
  id: 26cfd52f-2483-11ee-8040-88983bcc06d4
  last_name: Locatello
  orcid: 0000-0002-4850-0683
- first_name: Alp
  full_name: Yurtsever, Alp
  last_name: Yurtsever
- first_name: Olivier
  full_name: Fercoq, Olivier
  last_name: Fercoq
- first_name: Volkan
  full_name: Cevher, Volkan
  last_name: Cevher
citation:
  ama: 'Locatello F, Yurtsever A, Fercoq O, Cevher V. Stochastic Frank-Wolfe for composite
    convex minimization. In: <i>Advances in Neural Information Processing Systems</i>.
    Vol 32. ; 2019:14291–14301.'
  apa: Locatello, F., Yurtsever, A., Fercoq, O., &#38; Cevher, V. (2019). Stochastic
    Frank-Wolfe for composite convex minimization. In <i>Advances in Neural Information
    Processing Systems</i> (Vol. 32, pp. 14291–14301). Vancouver, Canada.
  chicago: Locatello, Francesco, Alp Yurtsever, Olivier Fercoq, and Volkan Cevher.
    “Stochastic Frank-Wolfe for Composite Convex Minimization.” In <i>Advances in
    Neural Information Processing Systems</i>, 32:14291–14301, 2019.
  ieee: F. Locatello, A. Yurtsever, O. Fercoq, and V. Cevher, “Stochastic Frank-Wolfe
    for composite convex minimization,” in <i>Advances in Neural Information Processing
    Systems</i>, Vancouver, Canada, 2019, vol. 32, pp. 14291–14301.
  ista: 'Locatello F, Yurtsever A, Fercoq O, Cevher V. 2019. Stochastic Frank-Wolfe
    for composite convex minimization. Advances in Neural Information Processing Systems.
    NeurIPS: Neural Information Processing Systems vol. 32, 14291–14301.'
  mla: Locatello, Francesco, et al. “Stochastic Frank-Wolfe for Composite Convex Minimization.”
    <i>Advances in Neural Information Processing Systems</i>, vol. 32, 2019, pp. 14291–14301.
  short: F. Locatello, A. Yurtsever, O. Fercoq, V. Cevher, in:, Advances in Neural
    Information Processing Systems, 2019, pp. 14291–14301.
conference:
  end_date: 2019-12-14
  location: Vancouver, Canada
  name: 'NeurIPS: Neural Information Processing Systems'
  start_date: 2019-12-08
date_created: 2023-08-22T14:09:35Z
date_published: 2019-12-29T00:00:00Z
date_updated: 2023-09-12T08:48:45Z
day: '29'
department:
- _id: FrLo
extern: '1'
external_id:
  arxiv:
  - '1901.10348'
intvolume: '        32'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://arxiv.org/abs/1901.10348
month: '12'
oa: 1
oa_version: Preprint
page: 14291–14301
publication: Advances in Neural Information Processing Systems
publication_identifier:
  isbn:
  - '9781713807933'
publication_status: published
quality_controlled: '1'
scopus_import: '1'
status: public
title: Stochastic Frank-Wolfe for composite convex minimization
type: conference
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 32
year: '2019'
...
---
_id: '14193'
abstract:
- lang: eng
  text: "A disentangled representation encodes information about the salient factors\r\nof
    variation in the data independently. Although it is often argued that this\r\nrepresentational
    format is useful in learning to solve many real-world\r\ndown-stream tasks, there
    is little empirical evidence that supports this claim.\r\nIn this paper, we conduct
    a large-scale study that investigates whether\r\ndisentangled representations
    are more suitable for abstract reasoning tasks.\r\nUsing two new tasks similar
    to Raven's Progressive Matrices, we evaluate the\r\nusefulness of the representations
    learned by 360 state-of-the-art unsupervised\r\ndisentanglement models. Based
    on these representations, we train 3600 abstract\r\nreasoning models and observe
    that disentangled representations do in fact lead\r\nto better down-stream performance.
    In particular, they enable quicker learning\r\nusing fewer samples."
article_processing_charge: No
arxiv: 1
author:
- first_name: Sjoerd van
  full_name: Steenkiste, Sjoerd van
  last_name: Steenkiste
- first_name: Francesco
  full_name: Locatello, Francesco
  id: 26cfd52f-2483-11ee-8040-88983bcc06d4
  last_name: Locatello
  orcid: 0000-0002-4850-0683
- first_name: Jürgen
  full_name: Schmidhuber, Jürgen
  last_name: Schmidhuber
- first_name: Olivier
  full_name: Bachem, Olivier
  last_name: Bachem
citation:
  ama: 'Steenkiste S van, Locatello F, Schmidhuber J, Bachem O. Are disentangled representations
    helpful for abstract visual reasoning? In: <i>Advances in Neural Information Processing
    Systems</i>. Vol 32. ; 2019.'
  apa: Steenkiste, S. van, Locatello, F., Schmidhuber, J., &#38; Bachem, O. (2019).
    Are disentangled representations helpful for abstract visual reasoning? In <i>Advances
    in Neural Information Processing Systems</i> (Vol. 32). Vancouver, Canada.
  chicago: Steenkiste, Sjoerd van, Francesco Locatello, Jürgen Schmidhuber, and Olivier
    Bachem. “Are Disentangled Representations Helpful for Abstract Visual Reasoning?”
    In <i>Advances in Neural Information Processing Systems</i>, Vol. 32, 2019.
  ieee: S. van Steenkiste, F. Locatello, J. Schmidhuber, and O. Bachem, “Are disentangled
    representations helpful for abstract visual reasoning?,” in <i>Advances in Neural
    Information Processing Systems</i>, Vancouver, Canada, 2019, vol. 32.
  ista: 'Steenkiste S van, Locatello F, Schmidhuber J, Bachem O. 2019. Are disentangled
    representations helpful for abstract visual reasoning? Advances in Neural Information
    Processing Systems. NeurIPS: Neural Information Processing Systems vol. 32.'
  mla: Steenkiste, Sjoerd van, et al. “Are Disentangled Representations Helpful for
    Abstract Visual Reasoning?” <i>Advances in Neural Information Processing Systems</i>,
    vol. 32, 2019.
  short: S. van Steenkiste, F. Locatello, J. Schmidhuber, O. Bachem, in:, Advances
    in Neural Information Processing Systems, 2019.
conference:
  end_date: 2019-12-14
  location: Vancouver, Canada
  name: 'NeurIPS: Neural Information Processing Systems'
  start_date: 2019-12-08
date_created: 2023-08-22T14:09:53Z
date_published: 2019-05-29T00:00:00Z
date_updated: 2023-09-12T09:02:43Z
day: '29'
department:
- _id: FrLo
extern: '1'
external_id:
  arxiv:
  - '1905.12506'
intvolume: '        32'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://doi.org/10.48550/arXiv.1905.12506
month: '05'
oa: 1
oa_version: Preprint
publication: Advances in Neural Information Processing Systems
publication_identifier:
  isbn:
  - '9781713807933'
publication_status: published
quality_controlled: '1'
status: public
title: Are disentangled representations helpful for abstract visual reasoning?
type: conference
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 32
year: '2019'
...
---
_id: '14197'
abstract:
- lang: eng
  text: "Recently there has been a significant interest in learning disentangled\r\nrepresentations,
    as they promise increased interpretability, generalization to\r\nunseen scenarios
    and faster learning on downstream tasks. In this paper, we\r\ninvestigate the
    usefulness of different notions of disentanglement for\r\nimproving the fairness
    of downstream prediction tasks based on representations.\r\nWe consider the setting
    where the goal is to predict a target variable based on\r\nthe learned representation
    of high-dimensional observations (such as images)\r\nthat depend on both the target
    variable and an \\emph{unobserved} sensitive\r\nvariable. We show that in this
    setting both the optimal and empirical\r\npredictions can be unfair, even if the
    target variable and the sensitive\r\nvariable are independent. Analyzing the representations
    of more than\r\n\\num{12600} trained state-of-the-art disentangled models, we
    observe that\r\nseveral disentanglement scores are consistently correlated with
    increased\r\nfairness, suggesting that disentanglement may be a useful property
    to encourage\r\nfairness when sensitive variables are not observed."
article_processing_charge: No
arxiv: 1
author:
- first_name: Francesco
  full_name: Locatello, Francesco
  id: 26cfd52f-2483-11ee-8040-88983bcc06d4
  last_name: Locatello
  orcid: 0000-0002-4850-0683
- first_name: Gabriele
  full_name: Abbati, Gabriele
  last_name: Abbati
- first_name: Tom
  full_name: Rainforth, Tom
  last_name: Rainforth
- first_name: Stefan
  full_name: Bauer, Stefan
  last_name: Bauer
- first_name: Bernhard
  full_name: Schölkopf, Bernhard
  last_name: Schölkopf
- first_name: Olivier
  full_name: Bachem, Olivier
  last_name: Bachem
citation:
  ama: 'Locatello F, Abbati G, Rainforth T, Bauer S, Schölkopf B, Bachem O. On the
    fairness of disentangled representations. In: <i>Advances in Neural Information
    Processing Systems</i>. Vol 32. ; 2019:14611–14624.'
  apa: Locatello, F., Abbati, G., Rainforth, T., Bauer, S., Schölkopf, B., &#38; Bachem,
    O. (2019). On the fairness of disentangled representations. In <i>Advances in
    Neural Information Processing Systems</i> (Vol. 32, pp. 14611–14624). Vancouver,
    Canada.
  chicago: Locatello, Francesco, Gabriele Abbati, Tom Rainforth, Stefan Bauer, Bernhard
    Schölkopf, and Olivier Bachem. “On the Fairness of Disentangled Representations.”
    In <i>Advances in Neural Information Processing Systems</i>, 32:14611–14624, 2019.
  ieee: F. Locatello, G. Abbati, T. Rainforth, S. Bauer, B. Schölkopf, and O. Bachem,
    “On the fairness of disentangled representations,” in <i>Advances in Neural Information
    Processing Systems</i>, Vancouver, Canada, 2019, vol. 32, pp. 14611–14624.
  ista: 'Locatello F, Abbati G, Rainforth T, Bauer S, Schölkopf B, Bachem O. 2019.
    On the fairness of disentangled representations. Advances in Neural Information
    Processing Systems. NeurIPS: Neural Information Processing Systems vol. 32, 14611–14624.'
  mla: Locatello, Francesco, et al. “On the Fairness of Disentangled Representations.”
    <i>Advances in Neural Information Processing Systems</i>, vol. 32, 2019, pp. 14611–14624.
  short: F. Locatello, G. Abbati, T. Rainforth, S. Bauer, B. Schölkopf, O. Bachem,
    in:, Advances in Neural Information Processing Systems, 2019, pp. 14611–14624.
conference:
  end_date: 2019-12-14
  location: Vancouver, Canada
  name: 'NeurIPS: Neural Information Processing Systems'
  start_date: 2019-12-08
date_created: 2023-08-22T14:12:28Z
date_published: 2019-12-08T00:00:00Z
date_updated: 2023-09-12T09:37:22Z
day: '08'
department:
- _id: FrLo
extern: '1'
external_id:
  arxiv:
  - '1905.13662'
intvolume: '        32'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://arxiv.org/abs/1905.13662
month: '12'
oa: 1
oa_version: Preprint
page: 14611–14624
publication: Advances in Neural Information Processing Systems
publication_identifier:
  isbn:
  - '9781713807933'
publication_status: published
quality_controlled: '1'
scopus_import: '1'
status: public
title: On the fairness of disentangled representations
type: conference
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 32
year: '2019'
...