Crowdsourcing for Machine Learning
Details
Y-DATA Meetup #14
Crowdsourcing for Machine Learning
Third online Y-DATA meetup.
https://us02web.zoom.us/j/88376071814
Intro:
This time we will talk about how the new generation of methods and tools allows to collect high quality human labelled data on a large scale and why every ML specialist should know how to use crowdsourcing.
AI stands on three pillars: algorithms, hardware and training data. While the first two have already become commodities on the market, the latter - reliable labelled data - is still a bottleneck in the industry.
Need to add twice as much data to the training set to improve your model? Want to validate the accuracy of a new classificator in an hour? Or maybe you are building a human-in-the-loop process with 90% of cases processed automatically and the trickiest 10% of cases fine-tuned by people in real time. You can do it all with crowdsourcing, but only with crowdsourcing done right.
During the meetup Olga Megorskaya, CEO of one of the largest crowdsourcing platforms Yandex.Toloka (https://toloka.ai/), will share best practices for tapping into the wisdom of the crowd and show real-life applications for search evaluation, voice assistants and self-driving cars.
Agenda:
Speaker: Olga Megorskaya, CEO of Yandex.Toloka
18:00 - 18:45 Best practices for tapping into the wisdom of the crowds.
19:00 - 19:45 Real-life applications for search evaluation, voice assistants and self-driving cars.
Bio:
Olga is responsible for providing human-labeled data for all AI projects at Yandex. She is in charge of integrating crowdsourcing into other business processes, such as customer support, product localization, software testing, etc. Olga helped Yandex to grow the number of crowd performers involved in data labeling from several dozens in 2009 up to 1M in 2018.
She graduated from the Saint Petersburg State University as a specialist in Mathematical Methods and Modeling in Economics. Also, she is a co-author of research papers and tutorials on efficient crowdsourcing and quality control at SIGIR, CVPR, KDD, WSDM, and SIGMOD.