Lukas Biewald (Weights & Biases)

Lukas is co-founder and CEO of Weights & Biases, an ML tooling company. He previously co-founded and led data labeling company Figure Eight (acquired by Appen).

Deep Learning In The Wild

• Machine Learning can be unpredictable and opaque.

• Deep Learning can be vulnerable to hacking.

• Machine Learning requires tons of clean training data.

• Deep Learning and GPUs break a lot of assumptions.

• Machine Learning can look at far more data than humans.

• The combination of humans and computers is powerful.

• What's coming?

  • Better tools and platforms.

  • More medical applications.

  • New solutions to training data.

• How to (successfully) ship deep learning projects:

  • Pay a lot of attention to your training data.

  • Get something working end-to-end right away, then improve one thing at a time.

  • Look for graceful ways to handle the inevitable cases where the algorithm fails.

Mentioned Resources:

• The State of Machine Intelligence (curated by Shivon Zillis)

• Read Lukas's article: "Why are Machine Learning Projects so Hard to Manage?"

• Datasets over Algorithms (credit to Alex Wissner-Gross)