Cost-efficient Gaussian tensor network embeddings for tensor-structured inputs
Authors: Linjian Ma, Edgar Solomonik
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct multiple experiments to demonstrate the efficacy of our proposed embeddings. Below we first justify the theoretical analysis in Theorem 4.1 and Theorem 4.3 via testing the sketching performance on tensor train inputs and Kronecker product inputs. We then perform experiments to demonstrate that the accuracy of our proposed sketching algorithms is comparable to that of state-of-the-art sketching techniques for CP decomposition and tensor train rounding. Our experiments are carried out on an Intel Core i7 2.9 GHz Quad-Core machine using Num Py [34] routines in Python. |
| Researcher Affiliation | Academia | Linjian Ma, Edgar Solomonik Department of Computer Science, University of Illinois at Urbana-Champaign {lma16, solomon2}@illinois.edu |
| Pseudocode | Yes | Algorithm 1 Sketching algorithm |
| Open Source Code | Yes | Our code is included in the supplemental materials. |
| Open Datasets | Yes | For CP-ALS, we conduct experiments on a Time-Lapse hyperspectral radiance image [32], which is a 3-D tensor with dimensions of 1024 1344 33. |
| Dataset Splits | No | The paper describes the input datasets and their dimensions but does not specify how they were split into training, validation, or test sets, or if standard splits were used without explicit mention. |
| Hardware Specification | Yes | Our experiments are carried out on an Intel Core i7 2.9 GHz Quad-Core machine using Num Py [34] routines in Python. |
| Software Dependencies | No | Our experiments are carried out on an Intel Core i7 2.9 GHz Quad-Core machine using Num Py [34] routines in Python. ... We use the Tensor Ly [23] library to truncate the input tensor... The paper mentions specific libraries but does not provide version numbers for them. |
| Experiment Setup | Yes | 10 ALS iterations are performed for all algorithms before the final fitness are calculated. |