Model Optimization

7 Costly Surprises of ML Part VI: Input Data Changes

In this article we talk about how the algorithms ML models learn are determined by input data, rather than by a programmer’s handwritten rules—has profound system-level performance effects in ML and introduce the topics of covariance and prediction shift detection.

Jan 14, 2020

7 Costly Surprises of ML Part IV: Feedback Loops

In ML systems, data dependencies carry a lot of depth but are more difficult to predict. In this article, we explore the importance of quality input data and introduce a new danger: feedback loops.

Dec 5, 2019

7 Costly Surprises of ML Part III: Input Entanglement

The second costly surprise: machine learning often evolves into a complex web of interlinking systems. Changing anything in the system can have wide-reaching and potentially negative effects far from the source of the change. This post explores how to avoid the problems such as a system poses.

Nov 27, 2019

7 Costly Surprises of ML Part II: Leaky Abstractions

The first costly surprise: machine learning lacks the clean abstraction of software. Leaky abstractions mean we can’t always rely on the contract that ML offers. This post examines what abstraction is, why it’s important, and how ML’s weak contracts can result in an ineffective system.

Nov 25, 2019