Tips for Productive ML Teams

Reduce Iteration Time

Iterate on Subsets

  1. Less Classes
  2. Less Data
  3. Smaller Models

expand slowly from there

Parallelize

  1. Pay for 8 GPUs to get experiment done in an hour instead of 1 GPU for 8 hour

Make it easy to deploy

Tune Open Models

Cookiecutter Projects

Friction Points

Sanity Checks

Automate

Experiment Job Queue

Bayesian Optimization

One Click Train Runs

Set Baselines

Naive Baseline

Standard Architecture Baseline

Top Baseline

Human Baseline

Measure

Segment Your Data

Black box gold test set

Gamify

Internal Evals and Leaderboards

Share Learnings and Knowledge

Invest in Tooling - Especially Data

Recommendations

Look at the Data

Build Feedback Loops

Ship Early and Often

Build Demos and POCs

Logs Logs Logs

Data Data Data

Resist the urge to get off the beaten path

Don’t do research until you need to