Designing for Machine Learning Data Labeling
jeu. 7 février à 00:45
Machine Learning at its most basic is the practice of using algorithms to parse data, learn from it, and then make a determination or prediction about something in the world. - NVIDIA Algorithmic performance can be highly impacted by the quality of input training data and the labels that are applied to that data. As they say, “Garbage in, garbage out”. However, getting high-quality inputs is easier said than done and is one large limitation in machine learning today. How can you get a high-quality labeled dataset? There are different approaches that depend on the project complexity, the training data, the size of the data science team, and the resources a company can allocate to data science projects. Different forms include labeling using internal teams, outsourcing, crowdsourcing, synthetic labeling and data programming. In this talk, Sabrina Siu will give a brief overview of these different techniques and how the Alluvium team has used a novel approach to get high quality labels for messy datasets. She will also explore using data visualizations to explain machine learning results and lessons learned throughout this process. Speaker Bio
Sabrina Siu is a user experience designer based in New York City. She currently works as a UX Designer and Researcher for Alluvium, a machine learning startup, where she focuses on combining principles of design and user behavior to create a compelling predictive platform. Agenda:
Pizza and socializing begins @ 6:45
Talk begins @ 7:00 followed by Q&A
A little extra socializing after the event
Nous avons temporairement désactivé la possibilité de naviguer vers les tags.