Deep Learning Systems at Scale

Jeff Dean's February 2015 University of Washington lecture on large-scale deep learning at Google, covering data-rich domains, neural representations, distributed training, and the production launch decisions behind speech, vision, and search. For software and ML builders who know basic ML and want sharper judgment about scale, tooling, and deployment.

4 sections ยท 10 lessons

Course outline

Why Scale Matters

  1. Data Becomes a Product Primitive
  2. Neural Networks Learn Abstractions
  3. Find the Learning Loop

Training Systems

  1. Frameworks Hide Distributed Messiness
  2. Parallelism Buys Experiment Time
  3. Asynchrony Creates New Failure Modes
  4. Choose a Scaling Strategy

Product Applications

  1. Speech Shows Production Leverage
  2. Vision Benefits from Shared Structure
  3. Search Learns from Behavior
  4. Ship or Keep Researching

Modern Trajectory

  1. Platforms Follow Workloads
  2. From Task Models to Foundations
  3. Modernize the 2015 Playbook

Start learning with Wondering