Too many ideas, too little data | Markus Nutz & Thomas Pawlitzki | ML Conference 2018

Leverage data-driven insights to overcome data scarcity and validate ideas, optimizing model selection, deployment, and communication with stakeholders, while aligning projects with business goals and measuring impact.

Key takeaways

Always start by defining a clear problem and setting goals.
Data availability is a major challenge, and it’s essential to have a data strategy.
Use small, incremental experiments to validate ideas and reduce risk.
Agile project management and iterative development can help overcome cold start problems.
Use random forests and other ensemble methods to improve model selection and reduce overfitting.
Ensure data quality and preprocessing steps are robust and reproducible.
Use serverless architectures and cloud-based services to deploy models quickly and efficiently.
Consider using open data sources and public datasets to augment internal data.
Always have a clear understanding of the problem domain and the data being used to solve it.
Use clear and simple language to communicate data science insights to non-technical stakeholders.
Divide complex data science projects into smaller, manageable tasks and prioritize them based on business goals.
Use data to validate and iterate on business decisions, rather than using intuition or anecdotal evidence.
Ensure data science projects are aligned with business objectives and are measurable in terms of impact.
Use cloud-based services and developer tools to accelerate experimentation and iteration.

Too many ideas, too little data | Markus Nutz & Thomas Pawlitzki | ML Conference 2018

More talks