Latent Functional Maps: a spectral framework for representation alignment
Authors: Marco Fumero, Marco Pegoraro, Valentino Maiorca, Francesco Locatello, Emanuele Rodolà
NeurIPS 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our framework on various applications, ranging from stitching to retrieval tasks, and on multiple modalities, demonstrating that Latent Functional Maps can serve as a swiss-army knife for representation alignment. |
| Researcher Affiliation | Academia | Marco Fumero IST Austria EMAIL Marco Pegoraro Sapienza, University of Rome EMAIL Valentino Maiorca Sapienza, University of Rome EMAIL Francesco Locatello IST Austria EMAIL Emanuele Rodolà Sapienza, University of Rome EMAIL |
| Pseudocode | No | The paper describes methods and processes in paragraph form, but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks or figures. |
| Open Source Code | No | We will share the code after the paper acceptance. All the data we use are open-source and publically available. |
| Open Datasets | Yes | We train 10 CNN models (the architecture is depicted in Appendix B.1) on the CIFAR-10 dataset [24], changing the initialization seed. |
| Dataset Splits | Yes | We use 2K random corresponding samples to construct the k-NN graphs and evaluate the retrieval performance on the remaining 18K word embeddings. |
| Hardware Specification | Yes | In all our experiments we used gpu rtx 3080ti and 3090. |
| Software Dependencies | No | The paper mentions general software like PyTorch but does not provide specific version numbers for any key software dependencies. |
| Experiment Setup | Yes | For each encoder, we compute a graph of 3,000 points with 300 neighbors per node. We optimize the problem in Equation 2 using the first 50 eigenvectors of the graph Laplacian and consider two different descriptors: the distance functions defined from the anchors (LFM+Ortho) and the labels (LFM+Ortho (Labels)). |