Rademacher Observations, Private Data, and Boosting
Authors: Richard Nock, Giorgio Patrini, Arik Friedman
ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on domains with up to millions of examples, backed up by theoretical arguments, display that learning over a small set of random rados can challenge the state of the art that learns over the complete set of examples. and 4. Basic experiments with RADOBOOST, Table 1. Comparison of RADOBOOST (n random rados), ADABOOST (Schapire & Singer, 1999) (full training fold) and ADABOOST(n) (n random examples in training fold) |
| Researcher Affiliation | Collaboration | Richard Nock RICHARD.NOCK@NICTA.COM.AU Giorgio Patrini GIORGIO.PATRINI@NICTA.COM.AU Arik Friedman ARIK.FRIEDMAN@NICTA.COM.AU NICTA & { The Australian National University, The University of New South Wales}, Sydney, Australia |
| Pseudocode | Yes | Algorithm 1 Rado boosting (RADOBOOST) and Algorithm 2 Feature-wise DP rados (DP-FEAT) |
| Open Source Code | No | The paper mentions a companion ArXiv paper (Nock et al., 2015) for proofs and additional experiments, but it does not provide an explicit statement about the release of their source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | We have performed comparisons with 10 folds stratified cross-validation (CV) on 16 domains of the UCI repository (Bache & Lichman, 2013) of varying size. |
| Dataset Splits | Yes | We have performed comparisons with 10 folds stratified cross-validation (CV) on 16 domains of the UCI repository (Bache & Lichman, 2013) of varying size. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU or CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using ADABOOST (Schapire & Singer, 1999) and states it implemented Step 2 in Algorithm DP-FEAT, but it does not specify any software libraries or their version numbers used in the experiments. |
| Experiment Setup | Yes | Each algorithm was ran for a total number of T = 1000 iterations; furthermore, the classifier kept for testing is the one minimizing the empirical risk throughout the T iterations; in doing so, we also assessed the early convergence of algorithms. We fixed n = min{1000, train fold size/2}. |