Top-k eXtreme Contextual Bandits with Arm Hierarchy
Authors: Rajat Sen, Alexander Rakhlin, Lexing Ying, Rahul Kidambi, Dean Foster, Daniel N Hill, Inderjit S. Dhillon
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we implement our algorithm using a hierarchical linear function class and show superior performance with respect to well-known benchmarks on simulated bandit feedback experiments using e Xtreme multi-label classification datasets. |
| Researcher Affiliation | Collaboration | Rajat Sen 1 Alexander Rakhlin 2 3 Lexing Ying 4 3 Rahul Kidambi 3 Dean Foster 3 Daniel Hill 3 Inderjit S. Dhillon 5 3 1Google Research, Mountain View (work done while at Amazon) 2Massachusetts Institute of Technology, Boston 3Amazon 4Stanford University, Palo Alto 5Department of Computer Science, University of Texas, Austin. |
| Pseudocode | Yes | Algorithm 1 Top-k Contextual Bandits with IGW, Algorithm 2 Beam search, Algorithm 3 e Xtreme Top-k Contextual Bandits with IGW |
| Open Source Code | Yes | We provide our implementation here. |
| Open Datasets | Yes | We use 6 XMC datasets for our experiments. Table 1 provides some basic properties of each dataset. We use the eurlex-4k XMC dataset (Bhatia et al., 2016) in Table 1... Our experiments are performed under simulated bandit feedback using real-world e Xtreme multi-label classification datasets (Bhatia et al., 2016). |
| Dataset Splits | No | We first form the tree and the routing functions from the held out portion of each dataset. The paper does not provide explicit train/validation/test splits for the main bandit feedback experiments, only an "Initialization Size" for a held-out set used for initial setup. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running experiments. |
| Software Dependencies | No | The paper mentions software like "Liblinear" (Fan et al., 2008) and "Eigen v3" (Guennebaud et al., 2010) via citations, but does not specify version numbers for other key software components, libraries, or programming languages used in the implementation. |
| Experiment Setup | Yes | The experiment is done with k = 50, r = 25 and b = 10. In our experiments the number of arms allowed to be chosen each time is k = 5. In Algorithm 3 we set the number of explore slots r = 3 and b = 10 (unless otherwise specified). The hyper-parameters in all the algorithms are tuned on the eurlex-4k datasets and then held fixed. |