Select-and-Sample for Spike-and-Slab Sparse Coding
Authors: Abdul-Saboor Sheikh, Jörg Lücke
NeurIPS 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Numerical Experiments. We first investigate the accuracy and convergence properties of our method on ground-truth data. we turned to verifying the approach on a denoising benchmark. Fig. 1C,D show the obtained results and a comparison to alternative approaches. Here we applied S5C with H = 10 000 hidden dimensions to demonstrate scalability of the method |
| Researcher Affiliation | Collaboration | Abdul-Saboor Sheikh Technical University of Berlin, Germany, and Cluster of Excellence Hearing4all University of Oldenburg, Germany, and SAP Innovation Center Network, Berlin EMAIL Jörg Lücke Research Center Neurosensory Science and Cluster of Excellence Hearing4all and Dept. of Medical Physics and Acoustics University of Oldenburg, Germany EMAIL |
| Pseudocode | Yes | Algorithm 1: Select-and-sample for spike-and-slab sparse coding (S5C) |
| Open Source Code | No | The paper does not provide an explicit statement or link for open-sourcing the code for the methodology described. |
| Open Datasets | Yes | For our application we used the standard van Hateren database [22], extracted N = 10^6 image patches of size 16x16, and applied pseudo-whitening following [21]. We applied S5C using a noisy house image [following 11, 4, 8, 7]. |
| Dataset Splits | No | The paper describes the total number of data points used for training, but does not specify explicit training, validation, or test dataset splits (e.g., percentages, sample counts, or cross-validation setup). |
| Hardware Specification | No | The paper states 'We used a multi-core parallelized implementation and executed the algorithm on up to 1000 CPU cores' but does not specify the exact CPU models, types, or other hardware details. |
| Software Dependencies | No | The paper does not mention any specific software dependencies with version numbers. |
| Experiment Setup | Yes | In all the experiments, the initial values of π were drawn from a uniform distribution on the interval [0.1, 0.5]..., µ was initialized with normally distributed random values, ψh was set to 1 and σd was initialized with the standard deviation of yd. The elements of W were iid drawn from a normal distribution with zero mean and a standard deviation of 5.0. We apply the S5C algorithm (Alg. 1) with H = 10 latents and M = 40 samples per data point and use two settings for preselection: (A) no preselection (H = H = 10) and (B) subspace preselection using H = 5. We applied the S5C algorithm with H = 256, select subspaces with H = 40 and used M = 100 samples per subspace. applied S5C for 50 EM iterations to the data using H = 20 dimensional subspaces and M = 50 samples per data point. |