Adaptive Sampling for SGD by Exploiting Side Information
Authors: Siddharth Gopal
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments on highly multiclass datasets show that our proposal converge significantly faster than existing techniques. |
| Researcher Affiliation | Industry | Siddharth Gopal SIDDHARTHG@GOOGLE.COM Google Inc, 1600 Amphitheatre Parkway |
| Pseudocode | Yes | The complete pseudocode is given in Algorithm 1. |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the methodology is openly available. |
| Open Datasets | Yes | We used three datasets for experimentation ALOI, CIFAR100 and IPC, 1. ALOI 1 : An image database... (footnote: http://www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/multiclass.html) ... 3. IPC 2 : A set of 75,250 patents... (footnote: http://gcdart.blogspot.com/2012/08/datasets 929.html) |
| Dataset Splits | Yes | Unless otherwise noted, all learning rates were carefully tuned (using the scheme in (Bottou, 2010)) to achieve the lowest objective at the cutoff point and the regularization was set using a 20% validation set. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers, only referencing algorithms or general concepts like 'Ada Grad'. |
| Experiment Setup | Yes | For all the experiments, we define ci to be the class-label associated with the instance. We generically set J = N/4 and δ = 0.5. The choice for J was made so as to ensure no noticeable increase in the computational cost and δ was set to a midpoint value between the two distributions. Unless otherwise noted, all learning rates were carefully tuned (using the scheme in (Bottou, 2010)) to achieve the lowest objective at the cutoff point and the regularization was set using a 20% validation set. |