Ομιλία Δασκαλάκη Παρασκ 23/01,17:30 (Αμφ.3 Νέα Κτίρια Ηλεκ)
Δημοσιεύτηκε: Τρί Ιαν 20, 2015 10:44 pm
Την Παρασκευή 23/01 στις 17:30 θα δώσει ομιλία ο Kαθηγητής Κωνσταντίνος Δασκαλάκης (ΜΙΤ και απόφοιτος ΣΗΜΜΥ ΕΜΠ) με θέμα:
"Beyond Berry Esseen: Structure and Learning of Sums of Random Variables"
Abstract: It is well-known that ~1/\epsilon^2 coin tosses are necessary and sufficient to estimate the bias of a coin. But how many samples are needed to learn the distribution of the sum of n independent Bernoullis? I will show that the answer is still O(1/\epsilon^2), independent of n, and will generalize this bound to sums of independent bounded integer random variables (SIIRVs). I will argue that the celebrated Berry-Esseen theorem and its variants fall short from characterizing the general structure of SIIRVs, and offer stronger finitary central limit theorems, which tightly characterizing their structure and have the afore-mentioned implications to learning.
H ομιλία θα γίνει στο Αμφιθέατρο 3 στα Νέα Κτίρια Ηλεκτρολόγων.
ΠΗΓΗ: https://shmmy.ntua.gr/forum/viewtopic.php?t=19680
"Beyond Berry Esseen: Structure and Learning of Sums of Random Variables"
Abstract: It is well-known that ~1/\epsilon^2 coin tosses are necessary and sufficient to estimate the bias of a coin. But how many samples are needed to learn the distribution of the sum of n independent Bernoullis? I will show that the answer is still O(1/\epsilon^2), independent of n, and will generalize this bound to sums of independent bounded integer random variables (SIIRVs). I will argue that the celebrated Berry-Esseen theorem and its variants fall short from characterizing the general structure of SIIRVs, and offer stronger finitary central limit theorems, which tightly characterizing their structure and have the afore-mentioned implications to learning.
H ομιλία θα γίνει στο Αμφιθέατρο 3 στα Νέα Κτίρια Ηλεκτρολόγων.
ΠΗΓΗ: https://shmmy.ntua.gr/forum/viewtopic.php?t=19680