Embedding Space Interpolation Beyond Mini-Batch, Beyond Pairs and Beyond Examples
Authors: Shashanka Venkataramanan, Ewa Kijak, laurent amsaleg, Yannis Avrithis
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically show that our solutions yield significant improvement over state-of-the-art mixup methods on four different benchmarks, despite interpolation being only linear. |
| Researcher Affiliation | Academia | 1Inria, Univ Rennes, CNRS, IRISA 2Institute of Advanced Research on Artificial Intelligence (IARAI) |
| Pseudocode | No | The paper describes the methods using mathematical equations and diagrams (e.g., Figure 2) but does not include a formally labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper does not explicitly state that source code for the methodology is being released or provide a link to a code repository. |
| Open Datasets | Yes | We use Pre Act Resnet-18 (R-18) [21] and WRN16-8 [63] as encoder on CIFAR-10 and CIFAR-100 datasets [29]; R-18 on Tiny Imagenet [60] (TI); and Resnet-50 (R-50) and Vi T-S/16 [13] on Image Net [44]. |
| Dataset Splits | No | The paper mentions using a 'mini-batch of size b = 128' and refers to training and test sets of datasets like CIFAR-100, but it does not explicitly provide the specific percentages or sample counts for training, validation, and test splits needed to reproduce the data partitioning. |
| Hardware Specification | Yes | Table 3 also shows the training speed as measured on NVIDIA V-100 GPU including forward and backward pass. |
| Software Dependencies | No | The paper does not specify version numbers for any software dependencies or libraries used for implementation (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | We use a mini-batch of size b = 128 examples in all experiments. Following manifold mixup [51], for every mini-batch, we apply Multi Mix with probability 0.5 or input mixup otherwise. For Multi Mix, the default settings are given in subsection 4.6. ... Our default setting is to draw uniformly at random from [0.5, 2] for every interpolation vector (column of ). |