Spatial Dependency Networks: Neural Layers for Improved Generative Image Modeling
May 3, 2021
Speakers
Đorđe Miladinović
Řečník · 0 sledujících
Aleksandar Stanić
Řečník · 2 sledující
Stefan Bauer
Řečník · 0 sledujících
Jürgen Schmidhuber
Řečník · 2 sledující
Joachim Buhmann
Řečník · 0 sledujících
About
How to improve generative modeling by better exploiting spatial regularities and coherence in images? To answer this question, we introduce a novel neural layer for building image generators (decoders) and apply it to variational autoencoders (VAEs). Our spatial dependency networks (SDNs) compute feature maps in a spatially coherent way, utilizing a sequential gating-based mechanism to distribute contextual information across 2-D maps. Different SDN layers of a deep neural network represent spatial dependencies at different levels of abstraction. Augmenting the decoder of a hierarchical VAE by spatial dependencies considerably improves density estimation over baseline convolutional architectures and the state-of-the-art among the models within the same class. We show that SDN can be applied to large images by synthesizing samples of high quality and coherence. In a vanilla VAE setting, we find that a powerful SDN decoder also improves learning disentangled representations, indicating that neural architectures play an important role in this task. Our results suggest favoring SDNs over convolutional networks in various VAE settings.How to improve generative modeling by better exploiting spatial regularities and coherence in images? To answer this question, we introduce a novel neural layer for building image generators (decoders) and apply it to variational autoencoders (VAEs). Our spatial dependency networks (SDNs) compute feature maps in a spatially coherent way, utilizing a sequential gating-based mechanism to distribute contextual information across 2-D maps. Different SDN layers of a deep neural network represent spat…
About ICLR 2021
The International Conference on Learning Representations (ICLR) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence called representation learning, but generally referred to as deep learning. ICLR is globally renowned for presenting and publishing cutting-edge research on all aspects of deep learning used in the fields of artificial intelligence, statistics and data science, as well as important application areas such as machine vision, computational biology, speech recognition, text understanding, gaming, and robotics.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Oral Session 1 - QA 1
Zhlédnout později
Panel Discussion
Zhlédnout později
Eric Rice, …
Research Accountability in Machine Learning
Zhlédnout později
Ditto: Fair and Robust Federated Learning Through Personalization
Zhlédnout později
Tian Li, …
Decoupling Global and Local Representations via Invertible Generative Flows
Zhlédnout později
