Paths to the Future: A Year at Google Brain
google-braintensorflowresearch-culturecareerconvex-optimization
Abstraction: Personal account of working at Google Brain on TensorFlow development
Key points:
- Author worked at Google Brain 2017–2018 as a software engineer on TensorFlow, then returned to Stanford PhD under Stephen Boyd
- Brain's culture paralleled Xerox PARC: high autonomy, self-directed researchers, multiple competing tools (TF 2.0, JAX, Swift for TF, Dex, MLIR) on the same vision
- TensorFlow is the 5th most-starred project on GitHub; Brain also produced BERT, TPUs, Google Translate, and Cloud AI/AutoML
- The parallel exploration culture was a double-edged sword: feature proliferation frustrated users
- Author co-wrote the TF 2.0 academic paper (arXiv:1903.01855) and became a core CVXPY developer
- 2018: Google renamed "Google Research" to "Google AI," signaling a narrowing focus toward deep learning and reinforcement learning
Connections: Google Brain · Tensorflow · Stanford · Deep Learning · Convex Optimization
Source: https://www.debugmind.com/2020/01/04/paths-to-the-future-a-year-at-google-brain/