Skip to content

Outlining Objectives for ML Systems


Defining the core objective of our task.
Goku Mohandas
· ·
Repository Β· Video

πŸ“¬  Receive new lessons straight to your inbox (once a month) and join 30K+ developers in learning how to responsibly deliver value with ML.

Intuition

Identifying the objective should always be the first step when solving any problem. It acts as the guide for all subsequent decision making and will prevent us from getting distracted along the way. However, identifying the objective isn't always straightforward, especially when we aren't analyzing the problem through the appropriate lens.

Problem to objective

What are some strategies to go from a problem to an objective?

Show answer
  1. Understand various aspects of the problem space (data sources, past efforts, etc.)
  2. Rephrase the problem from a user's perspective.
  3. Separate correlation from causation. What is main cause of the problem?

Application

In our application, we have a set of projects (with tags) that users search for (using tags).

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
{
    "id": 2427,
    "title": "Knowledge Transfer in Self Supervised Learning",
    "description": "A general framework to transfer knowledge from deep self-supervised models to shallow task-specific models.",
    "tags": [
        "article",
        "tutorial",
        "knowledge-distillation",
        "model-compression",
        "self-supervised-learning"
    ]
}

Our assigned problem is to:

Improve search and discoverability because users have complained that they are not able to discover the appropriate resources..

Based on key words in the problems, it's easy to prematurely jump to technological objectives such as:

  • we need a better search algorithm
  • we need better search infrastructure
  • we need a sleeker search interface

Though some of these objectives may be valid, they may not resolve the underlying issue. What we need to think about is why the user isn't able to discover the right resource.

  • What exactly are the user complaints?
  • Is it an issue of content presence or discoverability?
  • Any specifics on how exactly management wants to improve search?
  • What past data do we have to work with? Are the issues flagged?

Once we have a clear objective defined, we can start to think about potential solutions.

Resources


To cite this lesson, please use:

1
2
3
4
5
6
@article{madewithml,
    author       = {Goku Mohandas},
    title        = { Objective - Made With ML },
    howpublished = {\url{https://madewithml.com/}},
    year         = {2021}
}