Tighter bounds lead to improved classifiers
Authors: Nicolas Le Roux
ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Section 3 experiments this iterative scheme using generalized linear models over a variety of datasets to estimate its impact. We experimented the impact of using tighter bounds to the expected misclassification rate on several datasets |
| Researcher Affiliation | Industry | Nicolas Le Roux Criteo Research nicolas@le-roux.name |
| Pseudocode | Yes | Algorithm 1: Iterative supervised learning |
| Open Source Code | No | The paper does not contain any explicit statements about providing source code or links to a code repository for the methodology described. |
| Open Datasets | Yes | The Covertype binary dataset (Collobert et al., 2002) has 581012 datapoints... The Alpha dataset is a binary classification dataset... The MNist dataset is a digit recognition dataset... The IJCNN dataset is a dataset with 191681 samples. |
| Dataset Splits | Yes | We first set aside part of the dataset to compose the test set. We then performed k-fold cross-validation, using a generalized linear model, on the remaining datapoints for different values of T... We used the first 90% for the cross-validation and the last 10% for testing (Covertype). |
| Hardware Specification | No | The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | Each update was computed on a randomly chosen minibatch of 50 datapoints using the SAG algorithm (Le Roux et al., 2012). No specific software versions are mentioned. |
| Experiment Setup | Yes | For a fair comparison, each internal optimization was run for Z updates so that ZT was constant. Each update was computed on a randomly chosen minibatch of 50 datapoints using the SAG algorithm (Le Roux et al., 2012). For each value of T, we then selected the set of hyperparameters (λ and the number of iterations) which achieved the lowest validation classification error. |