《科学》封面文章 贝叶斯学习课件.pptx

Human-level Concept Learning through Probabilistic Program Induction Human-level Classification Generalization Problems Many of the leading approaches in machine learning are data-hungry. Machines learn poorer representations than people. Goals One-shot Learning Ability Good Performances on Generalization Solutions Bayesian Program Learning with 3 Key Ideas: Compositionality, Causality, and Learning to Learn. To reuse the pieces of existing ones, capturing the causal and compositional properties of real-world generative processes operating on multiple scales. Learning Task To learn 1623 handwritten characters from 50 writing systems Data set are from Omniglot Strokes were collected as detailed of online supplementary materials. Learning Task Parsing and generation of new examples after one example of a new concept To compare with 2 varieties of deep convolutional networks, a hierarchical deep model, a probabilistic model Learning Task The model learns to learn by fitting each conditional distribution to a background set of characters from30 alphabets, using both the image and the stroke data, and this image set was also used to pretrain the alternative deep learning models. Neither the production data nor any alphabets from this set are used in the subsequent evaluation tasks. Data Set Concrete Steps Parsing Characters Parsing Characters Generalization Results Source Code Influences Comments “Every week it seems we read about machines that can perform tasks in object recognition face recognition or speech recognition seemingly as well as humans do. Yet to scientists like me who study the mind, the gap between machine learning and human learning capacities remains vast. We want to close that gap, and that’s our long term goal.” -- Joshu Tenenbaum Comments  “Very impressive, the model’s ability to pass a visual Turing test is significant. It’s quite an achievement to make that work. It’s the death one more argument about why computer models aren’t good models of hum