Timezone: »
Learning generative models that span multiple data modalities, such as vision and language, is often motivated by the desire to learn more useful, generalisable representations that faithfully capture common underlying factors between the modalities. In this work, we characterise successful learning of such models as the fulfilment of four criteria: i) implicit latent decomposition into shared and private subspaces, ii) coherent joint generation over all modalities, iii) coherent cross-generation across individual modalities, and iv) improved model learning for individual modalities through multi-modal integration. Here, we propose a mixture-of-experts multi-modal variational autoencoder (MMVAE) for learning of generative models on different sets of modalities, including a challenging image <-> language dataset, and demonstrate its ability to satisfy all four criteria, both qualitatively and quantitatively.
Author Information
Yuge Shi (University of Oxford)
Siddharth N (Unversity of Oxford)
Brooks Paige (UCL)
Philip Torr (University of Oxford)
More from the Same Authors
-
2021 : MSA-Conditioned Generative Protein Language Models for Fitness Landscape Modelling and Design »
Alex Hawkins-Hooker · David Jones · Brooks Paige -
2022 : Using domain-domain interactions to probe the limitations of MSA pairing strategies »
Alex Hawkins-Hooker · David Jones · Brooks Paige -
2022 : Towards Healing the Blindness of Score Matching »
Mingtian Zhang · Oscar Key · Peter Hayes · David Barber · Brooks Paige · Francois-Xavier Briol -
2020 Workshop: Machine Learning for Molecules »
José Miguel Hernández-Lobato · Matt Kusner · Brooks Paige · Marwin Segler · Jennifer Wei -
2020 Poster: Barking up the right tree: an approach to search over molecule synthesis DAGs »
John Bradshaw · Brooks Paige · Matt Kusner · Marwin Segler · José Miguel Hernández-Lobato -
2020 Poster: Goal-directed Generation of Discrete Structures with Conditional Generative Models »
Amina Mollaysa · Brooks Paige · Alexandros Kalousis -
2020 Spotlight: Barking up the right tree: an approach to search over molecule synthesis DAGs »
John Bradshaw · Brooks Paige · Matt Kusner · Marwin Segler · José Miguel Hernández-Lobato -
2018 Workshop: Machine Learning for Molecules and Materials »
José Miguel Hernández-Lobato · Klaus-Robert Müller · Brooks Paige · Matt Kusner · Stefan Chmiela · Kristof Schütt -
2018 Poster: Faithful Inversion of Generative Models for Effective Amortized Inference »
Stefan Webb · Adam Golinski · Rob Zinkov · Siddharth N · Thomas Rainforth · Yee Whye Teh · Frank Wood -
2018 Poster: With Friends Like These, Who Needs Adversaries? »
Saumya Jetley · Nicholas Lord · Philip Torr -
2017 Workshop: Learning Disentangled Features: from Perception to Control »
Emily Denton · Siddharth Narayanaswamy · Tejas Kulkarni · Honglak Lee · Diane Bouchacourt · Josh Tenenbaum · David Pfau -
2017 Poster: Learning Disentangled Representations with Semi-Supervised Deep Generative Models »
Siddharth Narayanaswamy · Brooks Paige · Jan-Willem van de Meent · Alban Desmaison · Noah Goodman · Pushmeet Kohli · Frank Wood · Philip Torr