Cascading Bandits: Learning to Rank in the Cascade Model

Authors: Branislav Kveton, Csaba Szepesvari, Zheng Wen, Azin Ashkan

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experiment with our algorithms on several problems. The algorithms perform surprisingly well even when our modeling assumptions are violated.
Researcher Affiliation Collaboration Branislav Kveton KVETON@ADOBE.COM Adobe Research, San Jose, CA Csaba Szepesv ari SZEPESVA@CS.UALBERTA.CA Department of Computing Science, University of Alberta Zheng Wen ZHENGWEN@YAHOO-INC.COM Yahoo Labs, Sunnyvale, CA Azin Ashkan AZIN.ASHKAN@TECHNICOLOR.COM Technicolor Research, Los Altos, CA
Pseudocode Yes The pseudocode of both algorithms is in Algorithm 1.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The paper describes generating data from the BLB(L, K, p, ) problem class and the DBN model, rather than using a publicly available, named dataset with access information. For example: "We experiment with the class of problems BLB(L, K, p, ) in Section 4.3." and "We generate data from the dynamic Bayesian network (DBN) model of Chapelle & Zhang (2009)"
Dataset Splits No The paper does not specify traditional training/validation/test dataset splits. It describes experiments for online learning algorithms (bandits) which are evaluated on cumulative regret over a number of steps.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments (e.g., specific GPU/CPU models, memory, or cloud instance types).
Software Dependencies No The paper does not list any specific software dependencies with version numbers.
Experiment Setup Yes We set p = 0.2; and vary L, K, and . ... We run Cascade UCB1 and Cascade KL-UCB for n = 105 steps. ... The attraction probability of item e is (e) = w(e), where w(e) is given in (6). We set = 0.15. The satisfaction probabilities (e) of all items are the same. We experiment with two settings of (e), 1 and 0.7; and with two settings of persistence γ, 1 and 0.7. We run Cascade KL-UCB for n = 105 steps.