<div dir="ltr"><div><br></div>Hi Everyone,<br><div><br></div><div>This coming Friday Yu Lu will talk about:<br><br></div><div><div><div>Exact Exponent in Optimal Rates for Crowdsourcing<br></div><div><div></div><div></div><div><div><div><div></div><div>Crowdsourcing
has become a popular tool for labeling large datasets. This paper
studies the optimal error rate for aggregating crowdsourced labels
provided by a collection of amateur workers. Under the Dawid-Skene
probabilistic model, we establish matching upper and lower bounds with
an exact exponent $mI(\pi)$, where $m$ is the number of workers and
$I(\pi)$ is the average Chernoff information that characterizes the
workers' collective ability. Such an exact characterization of the error
exponent allows us to state a precise sample size requirement
$m>\frac{1}{I(\pi)}\log\frac{1}{\epsilon}$ in order to achieve
an $\epsilon$ misclassification error. In addition, our results imply
the optimality of various forms of EM algorithms given accurate
initializers of the model parameters. This is a joint work with Chao Gao and Dengyong Zhou from Microsoft Research. </div></div></div><div><div><div><br></div><div>See you Friday at 11am in the Stat's classroom.<br><br></div><div>Regards,<br></div><div>sekhar<br><br></div></div></div></div></div></div></div></div>