Hessian Eigenspectra of More Realistic Nonlinear Models

Authors: Zhenyu Liao, Michael W. Mahoney

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We also demonstrate our results empirically by showing that: 4. for a given response model (1), the Hessian eigenvalue distribution depends on the choice of loss function and the data/feature statistics in an intrinsic manner, e.g., bounded versus unbounded support and singleversus multi-bulk in Fig 2; and 5. there may exist two qualitatively different spikes one due to data signal µ and the other due to w or w and thus the underlying model which may appear on different sides of the main bulk, and their associated phase transition behaviors are characterized (Fig 4 versus 5).
Researcher Affiliation Academia Zhenyu Liao School of Electronic Information and Communications Huazhong University of Science and Technology, China zhenyu_liao@hust.edu.cn Michael W. Mahoney ICSI and Department of Statistics University of California, Berkeley, USA mmahoney@stat.berkeley.edu
Pseudocode No The paper does not contain pseudocode or a clearly labeled algorithm block.
Open Source Code No The paper does not provide an unambiguous statement or link for the open-source code of the methodology described.
Open Datasets No The paper uses synthetic data generated with parameters such as 'xi i.i.d. N(µ, C)' and specified dimensions (e.g., 'p = 800, n = 6 000'), but does not mention or provide access information for a publicly available or open dataset in the traditional sense.
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits, as its empirical components are demonstrations based on theoretical models and generated data rather than typical machine learning experiments with distinct data splits.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments.
Software Dependencies No The paper does not provide a reproducible description of ancillary software with specific version numbers.
Experiment Setup Yes Fig 1: 'eigenspectral properties of the Hessian of G-GLMs with p = 800, n = 6 000 and C = Ip.' Fig 2: 'p = 800, n = 6 000, logistic model in (2) with µ = 0, C = Ip, w = 0 and w = [ 1p/2, 1p/2]/ p.' Fig 4: 'p = 512 and n = 2 048.' Fig 5: 'p = 800 and n = 8 000.'