Boosting for Online Convex Optimization
Authors: Elad Hazan, Karan Singh
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | While the primary contribution of this work is theoretical, empirically testing our proposal serves as a sanity check for the theoretical results. |
| Researcher Affiliation | Collaboration | Elad Hazan 1 2 Karan Singh 3 1Google AI Princeton, Princeton, NJ, USA 2Princeton University, Princeton, NJ, USA 3Microsoft Research, Redmond, WA, USA. |
| Pseudocode | Yes | Algorithm 1 Bo OCO = Boosting Online Convex Opt. Algorithm 2 Bo BLO = Boosting Bandit Linear Opt. Algorithm 3 B4CO = Boosting for Contextual Opt. |
| Open Source Code | Yes | Further experimental details and code are detailed in the supplement. |
| Open Datasets | Yes | We evaluated the weak learners and the boosting algorithm on 3 publicly available datasets3 for regression with square loss... The California Hosuing dataset (Pace & Barry, 1997) and the Boston dataset (Harrison Jr & Rubinfeld, 1978) can be found at http://lib.stat.cmu.edu/datasets/. The Diabetes dataset used here is present in UCI ML Repository (https:// archive.ics.uci.edu/ml/datasets/Diabetes). |
| Dataset Splits | No | The paper states it evaluated on 3 publicly available datasets, but does not explicitly detail the training, validation, or test splits for these datasets within the main text. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper states, 'We use machine learning models from Scikit-Learn(Pedregosa et al., 2011) to implement base learners,' but does not specify a version number for Scikit-Learn or any other software dependencies. |
| Experiment Setup | Yes | The step size η was set to 0.01, and γ was chosen to be 0.1 these values were not tuned. Three online weak learners were considered: decision stumps, ridge regression, and a tiny multi-layer perceptron with one hidden unit trained via online gradient descent. |