ANC Workshop - Amos Storkey and Arno Onken Speaker: Arno Onken Title: Non-negative matrix factorization for identifying brainstem ensembles and their interactions with hippocampus Abstract: Non-negative matrix factorization (NMF) is a simple and versatile method for identifying component-based representations of large datasets. The algorithm can be adapted for taking into account various data distributions and for identifying invariant components across selected dimensions. Here, we adjust NMF for identifying ensembles of neurons from the brainstem of mice. We explore the relation of these ensembles to hippocampal CA1 neurons and illuminate coordination between brainstem and hippocampus across sleep states. This is joint work with Tomomi Tsunematsu, Amisha Patel and Shuzo Sakata. Speaker: Amos Storkey Title: Foundations of Learning Representations - How sturdy are they? Abstract: Learning Representations has been a key aspect of modern deep learning, so much so there is a major conference with that name. In this talk I will summarise a number of current methods of learning representations. However I will also focus on the lack of fundamental basis for most representation learning algorithms; how there is a built-in implicit prior via the models used and learning approaches that go beyond the target objective and are absolutely critical for the outcome. This difficulty will hopefully prompt discussion on whether there are better ways of understanding what makes a good approach for learning representations. We can discuss the idea of a universal representation for a given modality e.g. natural images. Mar 23 2021 11.00 - 12.00 ANC Workshop - Amos Storkey and Arno Onken Tuesday, 23rd March 2021 online
ANC Workshop - Amos Storkey and Arno Onken Speaker: Arno Onken Title: Non-negative matrix factorization for identifying brainstem ensembles and their interactions with hippocampus Abstract: Non-negative matrix factorization (NMF) is a simple and versatile method for identifying component-based representations of large datasets. The algorithm can be adapted for taking into account various data distributions and for identifying invariant components across selected dimensions. Here, we adjust NMF for identifying ensembles of neurons from the brainstem of mice. We explore the relation of these ensembles to hippocampal CA1 neurons and illuminate coordination between brainstem and hippocampus across sleep states. This is joint work with Tomomi Tsunematsu, Amisha Patel and Shuzo Sakata. Speaker: Amos Storkey Title: Foundations of Learning Representations - How sturdy are they? Abstract: Learning Representations has been a key aspect of modern deep learning, so much so there is a major conference with that name. In this talk I will summarise a number of current methods of learning representations. However I will also focus on the lack of fundamental basis for most representation learning algorithms; how there is a built-in implicit prior via the models used and learning approaches that go beyond the target objective and are absolutely critical for the outcome. This difficulty will hopefully prompt discussion on whether there are better ways of understanding what makes a good approach for learning representations. We can discuss the idea of a universal representation for a given modality e.g. natural images. Mar 23 2021 11.00 - 12.00 ANC Workshop - Amos Storkey and Arno Onken Tuesday, 23rd March 2021 online
