3 Things you’ll learn about at Labelbox Accelerate 2022

There's just one week until Labelbox Accelerate 2022 on November 9th, and the agenda is full of AI builders and leaders sharing practical insights that you can put to use the next day at work as you build better AI products. Read on to see a few of the insights you'll get at this event. Register today to save your spot!

Why finding great datasets is key to building transformative products

The opening session of Labelbox Accelerate is a fireside chat featuring Anthony Goldbloom, founder of leading data science and ML community Kaggle. They'll be discussing a few topics during the 45-minute session, including key practices for building transformative AI products.

Data-centric AI is no longer a new concept, with AI teams of every maturity level taking a data-first approach to training their models. While many are taking steps to improve the quality of their labels with benchmarking, consensus, and other QA workflows, the practice of curating a dataset that will significantly improve model performance is still largely underused. Join the session to learn why finding great datasets is a vital aspect of building the next generation of AI products and solutions.

How to cut your annotation costs in half with automation

Labelbox customers already leverage automation throughout their labeling processes with our best-in-class queueing system, customized workflows, and the SDK that enables seamless data transfers. AI teams are going beyond these measures, however, to realize 50-70% savings on annotation time and costs by using model-assisted labeling and AI-assisted annotation tools. Hear from data scientists at Nayya, Cape Analytics, and Move.ai as they discuss how their teams incorporate automation to increase efficiency.

How to pair active learning with weak supervision to build better models, fast

Using active learning — essentially, using a model’s weak points to inform what kind of data it’s trained on in the next iteration — is a highly effective data-centric approach to building AI. It can also save teams significant amounts of time and costs if they refrain from trying to get as much of their available unstructured data labeled as possible, and instead label smaller datasets as required for each iteration. Even so, teams that prioritize data and data labeling quality can experience longer labeling times, especially when review queues and other QA processes are involved. That’s why some leading AI teams are pairing their active learning practices with weak supervision, in which more general or lower-quality labels generated by models or software programs are used a starting point for labelers, saving them time on each labeling task.

In How to improve model performance with active learning and weak supervision, learn how the teams at Edelman, Deque, and Advent Health Partners are incorporating both active learning and weak supervision into their data engine, the results they’ve experienced, and get best practices for how to best combine the two methods.

Learn more about what’s on the Labelbox Accelerate 2022 agenda and register for the event here. See you on November 9th at 8:30 am PT!