Anisotropic Random Feature Regression in High Dimensions
Authors: Gabriel Mel, Jeffrey Pennington
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Simulations for m = 4000 (d,e; crosses) agree well with formulas. |
| Researcher Affiliation | Collaboration | Gabriel C. Mel Neurosciences Graduate Program Stanford University meldefon@gmail.com; Jeffrey Pennington Brain Team, Google Research jpennin@google.com |
| Pseudocode | No | The paper describes theoretical derivations and analyses but does not include any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any specific links or statements regarding the availability of open-source code for the methodology described. |
| Open Datasets | No | The paper uses synthetic data generated with specific parameters for simulations, such as 'covariates are Gaussian, xi N(0, Σ)' and '2-scale LJSD (Eq. (10))', rather than referencing publicly available datasets. |
| Dataset Splits | No | The paper describes theoretical analysis and simulations based on synthesized data, using terms like 'training error' and 'test error' related to the model's performance on generated data, but it does not specify dataset splits for training, validation, and testing as commonly found in empirical studies. |
| Hardware Specification | No | The paper mentions 'Simulations for m = 4000' but does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for these computations. |
| Software Dependencies | No | The paper mentions 'The NCAlgebra Mathematica package' was used for some derivations, but it does not specify any version numbers for this or any other software dependencies. |
| Experiment Setup | Yes | Figure 1: Test error, bias, and variance as a function of the overparameterization ratio (φ/ψ = n1/m) and the alignment θ for the 2-scale LJSD (Eq. (10)) with φ = n0/m = 10/9, σ = tanh, γ = 10 8 and σ2 ε = 1/100. Figure 2: All panels refer to a 3-scale covariance model with α = 103, and ψ = 1/2, s = ρ = 1, γ = 10 13 and σ2 ε = 10. |