One-Shot Generalization in Deep Generative Models

Abstract

Humans have an impressive ability to reason about new concepts and experiences from just a single example. In particular, humans have an ability for one-shot generalization: an ability to encounter a new concept, understand its structure, and then be able to generate compelling alternative variations of the concept. We develop machine learning systems with this important capacity by developing new deep generative models, models that combine the representational power of deep learning with the inferential power of Bayesian reasoning. We develop a class of sequential generative models that are built on the principles of feedback and attention. These two characteristics lead to generative models that are among the state-of-the art in density estimation and image generation. We demonstrate the one-shot generalization ability of our models using three tasks: unconditional sampling, generating new exemplars of a given concept, and generating new exemplars of a family of concepts. In all cases our models are able to generate compelling and diverse samples---having seen new examples just once---providing an important class of general-purpose models for one-shot machine learning.

8pgs, 1pg references, 1pg appendix, In Proceedings of the 33rd International Conference on Machine Learning, JMLR: W&CP volume 48, 2016

References in corpus (11)

• Auto-Encoding Variational Bayes
• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
• Variational Inference with Normalizing Flows
• Recurrent Models of Visual Attention
• Attend, Infer, Repeat: Fast Scene Understanding with Generative Models
• Neural GPUs Learn Algorithms
• Neural Variational Inference and Learning in Belief Networks
• Deep AutoRegressive Networks
• Learning Generative Models with Visual Attention
• Learning Wake-Sleep Recurrent Attention Models
• A Concept Learning Approach to Multisensory Object Perception

Cited by in corpus (28)

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• An Introduction to Variational Autoencoders
• Generalizing from a Few Examples: A Survey on Few-Shot Learning
• Learning What and Where to Draw
• Piecewise classifier mappings: Learning fine-grained learners for novel categories with few examples
• Sample Efficient Adaptive Text-to-Speech
• FIGR: Few-shot Image Generation with Reptile
• Distribution Matching in Variational Inference
• Few-Shot Learning with Global Class Representations
• DAWSON: A Domain Adaptive Few Shot Generation Framework
• Consistent Generative Query Networks
• Bringing Giant Neural Networks Down to Earth with Unlabeled Data
• Few-shot Learning for Time-series Forecasting
• Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video
• MetaCloth: Learning Unseen Tasks of Dense Fashion Landmark Detection from a Few Samples
• A Novel BiLevel Paradigm for Image-to-Image Translation
• Learning Task-General Representations with Generative Neuro-Symbolic Modeling
• SketchEmbedNet: Learning Novel Concepts by Imitating Drawings
• DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta
• Few-shot Learning for Spatial Regression
• Meta-learning One-class Classifiers with Eigenvalue Solvers for Supervised Anomaly Detection
• Meta-learning for Matrix Factorization without Shared Rows or Columns
• NP-DRAW: A Non-Parametric Structured Latent Variable Model for Image Generation
• MetalGAN: a Cluster-based Adaptive Training for Few-Shot Adversarial Colorization
• Few-shot Learning for Topic Modeling
• Deep Automodulators
• Meta-Active Learning for Node Response Prediction in Graphs
• Dynamic Flows on Curved Space Generated by Labeled Data