Max-Sliced Mutual Information
Authors: Dor Tsur, Ziv Goldfeld, Kristjan Greenewald
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present experiments that demonstrate the utility of m SMI for several tasks, encompassing independence testing, multi-view representation learning, algorithmic fairness, and generative modeling. |
| Researcher Affiliation | Collaboration | Dor Tsur Ben-Gurion University Ziv Goldfeld Cornell University Kristjan Greenewald MIT-IBM Watson AI Lab |
| Pseudocode | No | The paper describes methods and processes but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any concrete access information (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described. |
| Open Datasets | Yes | We next explore m SMI as an informationtheoretic generalization of CCA by examining its utility in multi-view representation learning a popular CCA application. Without using class labels, we obtain m SMI-based k-dimensional representations of the top and bottom halves of MNIST images (considered as two separate views of the digit image)." and "Table 2 shows results on the US Census Demographic dataset extracted from the 2015 American Community Survey, which has 37 features collected over 74,000 census tracts. |
| Dataset Splits | No | The paper mentions using datasets and running multiple trials/seeds but does not provide specific training/validation/test split percentages, sample counts, or citations to predefined splits. |
| Hardware Specification | Yes | We used an NVIDIA V100 GPU. |
| Software Dependencies | No | The paper mentions software components and methods (e.g., PyTorch, Adam optimizer, Re LU) but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | We employ the LIPO algorithm from [55] with a stopping criterion of 1000 samples." and "the learned Z is 80-dimensional." and "We consider 3 latent codes (C1, C2, C3) |