2018 DS/ML digest 22
NLP:
- New Google Char RNN https://arxiv.org/abs/1808.09943;
- NLP generalization https://thegradient.pub/frontiers-of-generalization-in-natural-language-processing/;
Notable papers
- Fooling CNNs may be easier than you think;
- New Google Char RNN https://arxiv.org/abs/1808.09943;
- Creating poems from Images in Chinese - hierarchical LSTMs;
Market / libraries / articles
- Received a mixed review on DVC - a library to manage your datasets:
- Multiple storage options;
- Very slow calculation of hashes;
- Received a positive feedback about sacred - a library to thouroughly track your experiments;
- AdamW soon to be integrated into upstream PyTorch?
- Someone managed to use PlaidML with Radeon GPU - https://habr.com/post/420989/ - but if you go to their website - they are dead or acquihired http://vertex.ai/;
- A library for interpreting Sklearn’s trees;
- Google’s framework for RL
ML; - Understanding surroundings in 3D https://thegradient.pub/beyond-the-pixel-plane-sensing-and-learning-in-3d/;
- Approx integrals https://habr.com/post/420867/;
- Interpreting the black boxes (RU);
- How FBI uses face recognition - 50-400m images are searchable;
- Yet another article about PHD opportunity costs:
- From fast.ai;
- From some random Russian dude;
Tech / news
- Man in the disk attack on Android;
- Chinese Orwelline citizen surveillance;
- The age of privacy nihilism
- MS runs an underwater data center;
- ML optimized data center cooling (no details);