LIVE FROM TWIMLCON! Encoding Company Culture in Applied AI Systems - #305

Topics covered
Popular Clips
Episode Highlights
Engineering Productivity
LinkedIn's approach to enhancing engineering productivity involves the strategic use of tools and infrastructure to support innovation. highlights the importance of maintaining productivity even as systems become more complex, emphasizing the role of the Proml initiative in standardizing processes and measuring success through the number of successful experiments 1. This initiative has led to a 30% improvement in successful experiments, fostering a unified company culture 1.
The industrial process will only work if the engineers are still productive. And in order to improve productivity, when you add more complexity, especially for such large scale distributed systems, if you really want them to run efficiently, if you want them to run in a reliable fashion, you have to make sure that the tooling and infrastructure can keep pace with the innovation that we are doing.
---
The focus on large-scale applications, such as recommender and search systems, ensures that the platform is tailored to LinkedIn's specific needs, maximizing return on investment 2.
Standardizing Tools
Standardizing tools for continuous integration and development is crucial for building a cohesive company culture at LinkedIn. explains that the tools used in machine learning reflect the company's culture, with LinkedIn opting for a standardized end-to-end approach to increase machine learning ROI 3. This approach emphasizes automation in model management, reducing the need for human intervention and allowing scientists to focus on innovation 4.
Continuous integration. Continuous development, if you look at different companies, they use different tools. And to me, that process, the tools you use, is actually a reflection of your culture.
---
The feature marketplace is a unique aspect of LinkedIn's platform, providing pre-built features to streamline the model creation process 4.
Cultural Impact
Company culture plays a pivotal role in driving innovation at LinkedIn. describes initiatives like the Ideas program and hackathons, which encourage grassroots innovation and energize the organization 5. By fostering a culture that values experimentation, LinkedIn ensures that innovation remains a core focus, with a centralized experimentation platform to measure and enhance experimental velocity 5.
We also do other things, like to encourage graphics route innovation. We have something called the Ideas program. So every quarter, anyone in the organization can actually submit an idea they want to work on.
---
This culture of innovation is crucial for LinkedIn's future in machine learning, as it seeks to balance productivity, efficiency, and innovation to maintain its competitive edge 6.
Related Episodes


Live from TWIMLcon! Culture & Organization for Effective ML at Scale (Panel) - #308
Answers 383 questions

Live from TWIMLcon! Operationalizing Responsible AI - #310
Answers 383 questions

Productive Machine Learning at LinkedIn with Bee-Chung Chen - TWiML Talk #200
Answers 383 questions

Live from TWIMLcon! Operationalizing ML at Scale with Hussein Mehanna - #306
Answers 383 questions

Engineering Production NLP Systems at T-Mobile - 600
Answers 383 questions

Productizing ML at Scale at Twitter with Yi Zhaung - TWIML Talk #271
Answers 383 questions

Live from TWIMLcon! Use-Case Driven ML Platforms with Franziska Bell - #307
Answers 383 questions

Industrializing Machine Learning at Shell with Daniel Jeavons - TWiML Talk #202
Answers 383 questions

Holistic Optimization of the LinkedIn News Feed - TWiML Talk #224
Answers 383 questions

AI for Content Creation with Debajyoti Ray - TWiML Talk #178
Answers 383 questions

Building an Autonomous Knowledge Graph with Mike Tung - #319
Answers 383 questions

Building AI Products with Hilary Mason - #11
Answers 383 questions

Applied Machine Learning for Publishers with Naveed Ahmad - TWiML Talk #182
Answers 383 questions

Live from TWIMLcon! Scaling ML in the Traditional Enterprise - #309
Answers 383 questions













