Latent Space Translation via Semantic Alignment
Authors: Valentino Maiorca, Luca Moschella, Antonio Norelli, Marco Fumero, Francesco Locatello, Emanuele Rodolà
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We extensively validate the adaptability of this translation procedure in different experimental settings: across various trainings, domains, architectures (e.g., Res Net, CNN, Vi T), and in multiple downstream tasks (classification, reconstruction). Notably, we show how it is possible to zero-shot stitch text encoders and vision decoders, or vice-versa, yielding surprisingly good classification performance in this multimodal setting. |
| Researcher Affiliation | Academia | Valentino Maiorca1, Luca Moschella1, Antonio Norelli1 Marco Fumero1 Francesco Locatello2 Emanuele Rodolà1 1Sapienza University of Rome 2Institute of Science and Technology Austria (ISTA) |
| Pseudocode | No | No pseudocode or clearly labeled algorithm blocks were found in the paper. |
| Open Source Code | Yes | Moreover, we release a modular Py Torch codebase2 implementing the various translation methods and scaling techniques. 2https://github.com/Flegyas/latent-translation |
| Open Datasets | Yes | We consider a variety of Computer Vision (MNIST [Lecun et al., 1998], Fashion MNIST [Xiao et al., 2017], N24News, CIFAR10, CIFAR100 Krizhevsky [2009]) and Natural Language Processing (TREC [Hovy et al., 2001, Li and Roth, 2002], DBpedia [Auer et al., 2007] , N24News [Wang et al., 2022], AG News [Zhang et al., 2015], IMDB [Maas et al., 2011] ) datasets. |
| Dataset Splits | No | The paper uses various datasets and mentions training SVM/MLP classifiers, but it does not provide specific details on how the datasets were split into training, validation, and test sets (e.g., percentages, sample counts, or explicit standard split citations for all datasets used). While it mentions 'test set', the validation split information is not provided. |
| Hardware Specification | Yes | All the experiments were conducted using a machine equipped with an Intel Core i7-9700k CPU, 64 GB of RAM, and an NVIDIA 2080TI GPU. |
| Software Dependencies | No | The paper lists several software and tools used (e.g., NN-Template, DVC, Py Torch Lightning, Weights and Biases, Hugging Face Transformers, Hugging Face Datasets, scikit-learn) with citations. However, it does not provide explicit version numbers for these software components (e.g., 'PyTorch 1.9', 'scikit-learn 0.24') but rather provides publication years for their respective papers or project initiations, which are not precise software versions for reproducibility. |
| Experiment Setup | Yes | In each instance, we use the same parallel anchors, that are uniformly chosen, in a quantity comparable with the dimensionality of the absolute representation. |