The Fifth Elephant 2023 Monsoon

On AI, industrial applications of ML, and MLOps

Tickets

Loading…

Aditi Tuli

@aditituli

Algorithm Friendly Design: Aligning product design with algorithmic requirements.

Submitted Jun 30, 2023

In the ever-changing landscape of digital products, algorithms play a vital role in providing personalized experiences for users. This talk will delve into the intersection of product and algorithmic design, addressing the need to create products that optimize algorithmic capabilities. For example, at Glance, a lock screen experience, we have 200m+ daily active users and need to design our user experience with our recommendation system in mind.

Product design from an algorithmic perspective:
The term algorithm friendly product design was coined by Eugene Wei and addresses the unique challenge faced when building a product that uses an algorithm: how do you allow your algorithm to see what it needs to see? This section of the talk will cover the challenges of creating training data, with a particular focus on implicit data collection – uncovering user preferences without explicitly asking the user what they like. It will also discuss ways to improve algorithmic feedback loops through intentional product design, such as isolated feedback loops, creating user friction, and separating exploration and exploitation.

Algorithmic design from an HCI perspective:
Given the mutual dependance between algorithms and UX design, it may not be beneficial to evaluate the performance of an algorithm and the performance of UX design in isolation from each other. This section will discuss how subjective system aspects like user perception mediates the relationship between the objective system (the algorithm itself) and objective measures of user experience. It will also touch on the ways in which user sentiment towards a product’s algorithm can affect user experience as a whole. This will build to the argument that evaluation of performance needs to be more holistic and consider the user experience and the algorithm together, moving beyond traditional measures of algorithmic performance like accuracy.

Best practices for algorithm friendly design:
Drawing on the above two sections and on examples from industry, this section of the talk will present some best practices for algorithm friendly design and sum up the challenges and opportunities involved in designing products that work in conjunction with algorithms, rather than parallel to them. It will make the case for moving away from the traditional siloed structure to a deeper integration between teams that design user experiences and teams that develop algorithms. This will be one concrete takeaway from the talk; it is important to align product design with algorithmic requirements through effective collaboration between user experience and data science teams, right from the start of the design process.

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hybrid access (members only)

Hosted by

All about data science and machine learning

Supported by

E2E Cloud is India's first AI hyper scaler, a cloud computing platform providing accelerated cloud-based solutions at maximum optimization and lowest pricing