2019 DS/ML digest 01

2019 DS/ML digest 01

Posted by snakers41 on December 19, 2018

Blogs / posts / articles

  • Does AI make strong tech companies stronger? - we should compare machine learning to SQL;
  • Airport disrupted by drones;
  • Netflix movie watched 45m times in a week?
  • Yet another randomness in a competition;
  • NIPS Conversational AI competition results - shake up, no technical details;
  • Scientific programming in SWIFT???
  • SkLearn drops python 2+ support;
  • Age discrimination in tech;
  • Ben Evans’ digest;
  • SRGAN + Max Payne textures;
  • Writing a cover letter to Google;

NLP

Sliced Recurrent Neural Networks

  • Link:
  • Sliced recurrent neural networks (SRNNs), which could be parallelized by slicing the sequences into many subsequences;
  • Without changing the recurrent units, SRNNs are 136 times as fast as standard RNNs and could be even faster when we train longer sequences;
  • Suitable only for classification;
  • Comprable on long sequences and sentiment analysis - but much faster;
  • Dilated CNNs are also comparable, but even 2x faster;

Datasets

  • Largest camera network and reliability of detection CNNs;
  • Tencent ML images updated - 17,609,752 training and 88,739 validation image URLs, which are annotated with up to 11,166 categories;

Trivia

Papers

  • Learning Not to Learn: Training Deep Neural Networks with Biased Data (link):
    • Interesting idea - if you have information on bias in your data, you can unlearn bias in data using adversarial training and KLD loss;
  • http://arxiv.org/abs/1901.02985

Competitions

Hardware

  • ML chip from Huawei?