读NeurIPS 2018 accepted paper list的十点感想

• 一.图模型graph很火，至少有45篇graph文章和4篇topic方面的文章。
• 二.变分(Variational)非常火 ，至少有35篇的文章。
• 三.现在机器学习慢慢朝着推理（39篇Inference和6篇reason）方向发展。
• 四.强化学习的文章也比较多，41篇reinforce，5篇reward，22篇policy和5篇Imitation Learning。
• 五.11篇元学习（meta-learning）的文章，7篇vc维（pac-learning）理论方面的文章，9篇可解释性（interpret）方面的文章。
• 六.至少63篇关于GAN网络方面的文章。
• 七.8篇video和11篇3D方面的文章，video understanding可能是下一个需要突破的task。
• 八.5篇脉冲神经网络（spiking Neural Networks）方面的文章，3篇Capsule方面的文章。
• 九.有大概19篇跨媒体（文本与语音，文本与视觉以及视觉关系）方面的文章，朝着reason，knowledge和graph方向发展。
  • 1.Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding
  • 2.LinkNet: Relational Embedding for Scene Graph
  • 3.Disentangling Learning for CaptionBot and DrawingBot
  • 4.Dialog-based Interactive Image Retrieval
  • 5.Hybrid Retrieval-Generation Reinforced Agent for Medical Image Report Generation
  • 6.Hybrid Knowledge Routed Modules for Large-scale Object Detection
  • 7.Partially-Supervised Image Captioning
  • 8.Learning to Specialize with Knowledge Distillation for Visual Question Answering
  • 9.Chain of Reasoning for Visual Question Answering
  • 10.Learning Conditioned Graph Structures for Interpretable Visual Question Answering
  • 11.Out of the Box: Reasoning with Graph Convolution Nets for Factual Visual Question Answering
  • 12.Overcoming Language Priors in Visual Question Answering with Adversarial Regularization
  • 13.Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis
  • 14.Answerer in Questioner’s Mind: Information Theoretic Approach to Goal-Oriented Visual Dialog
  • 15.Speaker-Follower Models for Vision-and-Language Navigation
  • 16.A Neural Compositional Paradigm for Image Captioning
  • 17.Weakly Supervised Dense Event Captioning in Videos
  • 18.Text-Adaptive Generative Adversarial Networks: Manipulating Images with Natural Language
  • 19.Unsupervised Cross-Modal Alignment of Speech and Text Embedding Spaces
• 十.Yoshua Bengio，Yann LeCu和Geoffrey E Hinton三位大仙共有7篇文章，有必要关注一下meta-learning。
  • 1.Image-to-image translation for cross-domain disentanglement， Yoshua Bengio。
  • 2.Dendritic cortical microcircuits approximate the backpropagation algorithm ， Yoshua Bengio。
  • 3.Bayesian Model-Agnostic Meta-Learning， Yoshua Bengio。
  • 4.Sparse Attentive Backtracking: Temporal Credit Assignment Through Reminding， Yoshua Bengio。
  • 5.MetaGAN: An Adversarial Approach to Few-Shot Learning， Yoshua Bengio。
  • 6.GLoMo: Unsupervised Learning of Transferable Relational Graphs， Yann LeCun。
  • 7.Assessing the Scalability of Biologically-Motivated Deep Learning Algorithms and Architectures， Geoffrey E Hinton。