AI Today Podcast: Insights into the future of ML Development: Interview with Gideon Mendels, CEO of Comet
Mar 17, 2021
auto_awesome
In this podcast, the CEO of Comet discusses challenges of integrating ML models into production and the importance of tracking and optimizing experiments. The discussion also explores the rising adoption of AI in various industries and the benefits of upskilling software engineers in machine learning.
ML models in production must align with business KPIs to succeed.
Tools for tracking, comparing, explaining, and optimizing ML experiments are essential for team success.
Deep dives
Overview of AI Today Podcast
The AI Today podcast hosted by Kathleen Mulch and Ronald Schmillzer covers various aspects of artificial intelligence and machine learning technologies. The podcast delves into emerging AI trends, technologies, and practical use cases across different industries. With over 200 episodes recorded in the past four years, the podcast features discussions with notable experts, CEOs, and CIOs from diverse backgrounds, shedding light on the application of AI in real-world scenarios and the evolving landscape of AI markets.
Challenges in Building Machine Learning Models
One of the significant challenges organizations face when transitioning machine learning models to production is building models that align with business key performance indicators (KPIs). Unlike traditional software engineering, machine learning involves an iterative process with various pitfalls such as choosing the wrong metric, data leakage, or inadequate dataset signals. The lack of robust processes and tools for developing models that justify deployment hampers many teams, necessitating a need for structured methods to ensure model quality.
Importance of Tools for Tracking and Optimizing Experiments
Having a system of record for tracking, comparing, explaining, and optimizing machine learning experiments and models is essential for team success. Similar to how software engineering relies on systems like GitHub, Salesforce, or HubSpot, data scientists and teams benefit significantly from experiment and model management platforms. These tools provide value by enabling data scientists to track experiments, compare model performance, identify bias, and optimize models, while also offering visibility to managers for tracking team progress and maintaining institutional knowledge within organizations.
As companies begin to move machine learning models into production there are a variety of factors that need to be addressed. In this podcast Cognilytica analysts Kathleen Walch and Ronald Schmelzer interview Gideon Mendels, CEO of Comet. He discusses challenges organizations face today with trying to bring ML models into production, why it’s important to have a tool for data scientists and teams to track, compare, explain and optimize experiments and models, and where ML Ops fits into the overall ML ecosystem.