TY - GEN
T1 - On the Lower Bound of Minimax Error for Crowdsourcing
AU - Li, Zhuqing
AU - Mohajer, Soheil
AU - Touri, Behrouz
N1 - This work is supported by the AFOSR FA9550-23-1-0057 grant.
PY - 2025
Y1 - 2025
N2 - We consider a binary crowdsourcing problem in which independent tasks, such as fake-news detection or binary classification tasks, are assigned to n imperfect agents, each of which may misclassify/mislabel the tasks with some unknown probability. We revisit a result in [1], presented at NeurIPS 2017, regarding a lower bound for the minimax error of this problem, and then investigate a more general lower bound under a broader parameter set. We demonstrate that for any estimator using a sufficiently large number of independent observations T of the labeling results, the probability of having an estimation error of at least 1 / √T is bounded away from zero by a constant that depends on a structural feature of the parameter set. Additionally, we derive a local version of this result, which essentially asserts that for any r-ball around any parameter point and any estimator based on T ≥ 4 / r2 samples, there always exists a point in that ball that cannot be estimated accurately to within o(1 / √T).
AB - We consider a binary crowdsourcing problem in which independent tasks, such as fake-news detection or binary classification tasks, are assigned to n imperfect agents, each of which may misclassify/mislabel the tasks with some unknown probability. We revisit a result in [1], presented at NeurIPS 2017, regarding a lower bound for the minimax error of this problem, and then investigate a more general lower bound under a broader parameter set. We demonstrate that for any estimator using a sufficiently large number of independent observations T of the labeling results, the probability of having an estimation error of at least 1 / √T is bounded away from zero by a constant that depends on a structural feature of the parameter set. Additionally, we derive a local version of this result, which essentially asserts that for any r-ball around any parameter point and any estimator based on T ≥ 4 / r2 samples, there always exists a point in that ball that cannot be estimated accurately to within o(1 / √T).
UR - https://www.scopus.com/pages/publications/105021918739
UR - https://www.scopus.com/pages/publications/105021918739#tab=citedBy
U2 - 10.1109/ISIT63088.2025.11195234
DO - 10.1109/ISIT63088.2025.11195234
M3 - Conference contribution
AN - SCOPUS:105021918739
T3 - IEEE International Symposium on Information Theory - Proceedings
BT - ISIT 2025 - 2025 IEEE International Symposium on Information Theory, Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2025 IEEE International Symposium on Information Theory, ISIT 2025
Y2 - 22 June 2025 through 27 June 2025
ER -