Weekend long read suggestions 2019/51

This may easily be the last one for this year. Thank you everybody for you support, wish you a great holidays and see you in an amazing year 2020.

Just interesting stuff

• The #1 bug predictor is not technical, it’s organizational complexity

  Article summarizes findings of Miscrosoft Reasearch, after the unsuccessful launch of Windows Vista in 2007. Original full report is available at the Miscrosoft’s site. As title spoiled, the main predictor is not technical complexity, but company culture and setup.

• Anki, Jibo, and Kuri: What We Can Learn from Social Robots That Didn’t Make It

  I am not building a social robot, but those findings are primarily about human interactions after all. So these failures are an interesting for human-to-human interactions (especially when they are remote).

• Interactive Core Memory Shield Using LED Matrix

  This project is just a beautiful teaching tool, if you find yourself in need of explaining how core memory works.

• TMAI #192: The Perfect Rose Bush Distraction

  This week’s write from Avinash Kaushik is excellent. He talks about rose bush culture, which is basically about not paying attention to the bigger and more important picture.

Machine Learning

• Deep Learning

  Classical book from MIT Press by Ian Goodfellow and Yoshua Bengio and Aaron Courville released as a free web book.

• Decoding the Best Machine Learning Papers from NeurIPS 2019

  • Distribution-Independent PAC Learning of Halfspaces with Massart Noise
  • Nonparametric Density Estimation & Convergence Rates for GANs under Besov IPM Losses
  • Fast and Accurate Least-Mean-Squares Solvers
  • Uniform convergence may be unable to explain generalization in Deep Learning
  • Putting An End to End-to-End: Gradient-Isolated Learning of Representations
  • Scene Representation Networks: Continuous 3D-Structure-Aware Neural Scene Representations
  • Dual Averaging Method for Regularized Stochastic Learning and Online Optimization

• Future of Data Engineering

  Talk goes through 6 stages of building a working data engineering platform. Chris Riccomini from WePay shows how are also data platforms shifting from monoliths into microservices / distributed architectures.

• Pretraining BERT with Layer-wise Adaptive Learning Rates

  Article also includes example implementation of LAMB (called NVLAMB) in their BERT repositories. One based on PyTorch and the other one on TensorFlow.

• Accelerate OpenCV: Optical Flow Algorithms with NVIDIA Turing GPUs

  NVIDIA has released hw accelerated OpticalFlow SDK and it’s integration with OpenCV.

Building Software

• Idiomatic Rust

  A peer-reviewed collection of articles/talks/repos which helps you to write elegant Rust programs.

• Elixir Boilerplate

  The project template which is used at Mirego to build Phoenix projects. In has full GraphQL, gettext and code-quality setup. Definitely a lot of inspiration for our own projects there.

• Elixir Styleguide

  A community drive style guide for Elixir.

Infrastructure

• PostgreSQL at low level: stay curious!

  Article describes ways how to observe (yeah some BPF greatness there as well) some of the low-level behavior of PostgreSQL.

• Building a Kubernetes Platform at Pinterest

  Notes from Pinterest Kubernetes team about their experience of scaling k8s for their production.

• BeyondProd: How Google moved from perimeter-based to cloud-native security

  Article is, unfortunately, not very detailed. But provides links to key projects you should consider when trying to adopt BeyondProd. BeyondProd is set of the following security principles:

  • Protection of the network at the edge
  • No inherent mutual trust between services
  • Trusted machines running code with known provenance
  • Choke points for consistent policy enforcement across services, for example, ensuring authorized data access
  • Simple, automated, and standardized change rollout, and
  • Isolation between workloads