ML Review 6

This Review gives an overview of interesting stuff I stumbled over which is related to machine learning.

News

• Hardware
  • Inside Volta: The World’s Most Advanced Data Center GPU
  • NVIDIA DGX SYSTEMS
• Core-Software
  • CUDA 9
  • cuDNN 7
  • Scaling Keras Model Training to Multiple GPUs
• Applications:
  • Facebook: Transitioning entirely to neural machine translation
  • nVidia: Microsoft Sets New Speech Recognition Record
  • nature: Artificial intelligence identifies plant species for science
  • nVidia: AI Writes Believable Fake Yelp Reviews
• Other
  • OpenAI: Robust Adversarial Examples
  • Where does this hyped news come from? Facebook shut down AI that invented its own language.
  • Wired: Machines Taught by Photos Learn a Sexist View of Women

Live Demos, Websites and Blogs

• Anish Athalye: A Step-by-Step Guide to Synthesizing Adversarial Examples
• Aran Lunzer and Amelia McNamara: What's so hard about histograms?
• Kaggle:
  • ImageNet Object Localization Challenge
  • ImageNet Object Detection Challenge
  • ImageNet Object Detection from Video Challenge
• Martin Thoma: Document Classification
• Metadata: A Comparison of Distributed Machine Learning Platforms (Spark, PMLS, TensorFlow)
• Hu, Peiyun and Ramanan, Deva: Finding Tiny Faces (MATLAB)
• Mozilla: Common Voice
• Slav Ivanov: 37 Reasons why your Neural Network is not working
• Humble Book Bundle - Data Science
• Deep Learning Is Not Good Enough, We Need Bayesian Deep Learning for Safe AI
• DeepL: Neues Tool übersetzt viel besser als Google und Microsoft
• TensorFlow Debugger Screencast

Publications

• Martin Thoma: Analysis and Optimization of Convolutional Neural Network Architectures
• Abhinav Gupta (Google): Revisiting the Unreasonable Effectiveness of Data

Software

• mozilla/DeepSpeech (Tensorflow)
• dennybritz/cnn-text-classification-tf

Interesting Questions

• Why are Machine Learning models called black boxes?
• What is an appropriate way to compare classifiers with different sets of classes?