:orphan:


Losses
*********************

Examples of how to train models with different objective functions.



.. raw:: html

    <div class="sphx-glr-thumbnails">

.. thumbnail-parent-div-open

.. thumbnail-parent-div-close

.. raw:: html

    </div>


Supervised
----------------------

Examples of how to use supervised objective functions that are trained with examples of OOD data.



.. raw:: html

    <div class="sphx-glr-thumbnails">

.. thumbnail-parent-div-open

.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="We train a model with  Outlier Exposure &lt;pytorch_ood.loss.OutlierExposureLoss&gt; on the CIFAR10.">

.. only:: html

  .. image:: /auto_examples/loss/supervised/images/thumb/sphx_glr_outlier_exposure_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_supervised_outlier_exposure.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Outlier Exposure</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="We train a model with Virtual Outlier Synthesizer Loss&lt;pytorch_ood.loss.vos.VirtualOutlierSynthesizingRegLoss&gt; on the CIFAR10.">

.. only:: html

  .. image:: /auto_examples/loss/supervised/images/thumb/sphx_glr_cifar_vos_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_supervised_cifar_vos.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Virtual  Outlier Synthesizer Loss</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="Multi Class Hypersphere Anomaly Detection &lt;pytorch_ood.loss.MCHADLoss&gt; (MCHAD) can be seen as a multi-class generalization of Deep One-Class Learning, where there are several centers \{\mu_1, \mu_2, ..., \mu_y\} in the output space of the model, one for each class. During training, the representation f_{\theta}(x) from class y is drawn towards the corresponding center \mu_y.">

.. only:: html

  .. image:: /auto_examples/loss/supervised/images/thumb/sphx_glr_mchad_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_supervised_mchad.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">MCHAD</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="We train a model with  Energy Margin &lt;pytorch_ood.loss.EnergyMarginLoss&gt; on the CIFAR10.">

.. only:: html

  .. image:: /auto_examples/loss/supervised/images/thumb/sphx_glr_scone_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_supervised_scone.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Scone</div>
    </div>


.. thumbnail-parent-div-close

.. raw:: html

    </div>


.. toctree::
   :hidden:

   /auto_examples/loss/supervised/outlier_exposure
   /auto_examples/loss/supervised/cifar_vos
   /auto_examples/loss/supervised/mchad
   /auto_examples/loss/supervised/scone


Unsupervised
----------------------

Examples of how to use unsupervised objective functions that are trained **without** examples of OOD data.



.. raw:: html

    <div class="sphx-glr-thumbnails">

.. thumbnail-parent-div-open

.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="We train a One-Class model (that is, a model that does not need class labels) on MNIST, using Deep SVDD &lt;pytorch_ood.loss.DeepSVDDLoss&gt;. SVDD places a single center \mu in the output space of a model f_{\theta}. During training, the parameters  \theta are adjusted to minimize the (squared) sum of the distances of representations f_{\theta}(x) to this center. Thus, the model is trained to map the training samples close to the center. The idea is that the model learns to map only IN samples close to the center, and not OOD samples. The distance to the center can be used as outlier score.">

.. only:: html

  .. image:: /auto_examples/loss/unsupervised/images/thumb/sphx_glr_svdd_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_unsupervised_svdd.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Deep One-Class Learning</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="Class Anchor Clustering &lt;pytorch_ood.loss.CACLoss&gt; (CAC) can be seen as a multi-class generalization of Deep One-Class Learning, where there are several centers \{\mu_1, \mu_2, ..., \mu_y\} in the output space of the model, one for each class. During training, the representation f_{\theta}(x) from class y is drawn towards the corresponding center \mu_y.">

.. only:: html

  .. image:: /auto_examples/loss/unsupervised/images/thumb/sphx_glr_cac_thumb.png
    :alt:

  :ref:`sphx_glr_auto_examples_loss_unsupervised_cac.py`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Class Anchor Clustering</div>
    </div>


.. thumbnail-parent-div-close

.. raw:: html

    </div>


.. toctree::
   :hidden:

   /auto_examples/loss/unsupervised/svdd
   /auto_examples/loss/unsupervised/cac


.. only:: html

  .. container:: sphx-glr-footer sphx-glr-footer-gallery

    .. container:: sphx-glr-download sphx-glr-download-python

      :download:`Download all examples in Python source code: loss_python.zip </auto_examples/loss/loss_python.zip>`

    .. container:: sphx-glr-download sphx-glr-download-jupyter

      :download:`Download all examples in Jupyter notebooks: loss_jupyter.zip </auto_examples/loss/loss_jupyter.zip>`


.. only:: html

 .. rst-class:: sphx-glr-signature

    `Gallery generated by Sphinx-Gallery <https://sphinx-gallery.github.io>`_