Class Normalization for (Continual)? Generalized Zero-Shot Learning
Authors: Ivan Skorokhodov, Mohamed Elhoseiny
ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Then, we test our approach on 4 standard ZSL datasets and outperform sophisticated modern Sot A with a simple MLP optimized without any bells and whistles and having 50 times faster training speed. |
| Researcher Affiliation | Academia | 1King Abdullah University of Science and Technology (KAUST), Saudi Arabia 2Moscow Institute of Physics and Technology (MIPT), Russia |
| Pseudocode | No | The paper does not contain any explicitly labeled "Pseudocode" or "Algorithm" blocks. Procedures are described in paragraph text or high-level diagrams. |
| Open Source Code | Yes | The source code is available at https://github.com/universome/class-norm. |
| Open Datasets | Yes | We use 4 standard datasets: SUN (Patterson et al., 2014), CUB (Welinder et al., 2010), Aw A1 and Aw A2 and seen/unseen splits from Xian et al. (2018a). |
| Dataset Splits | Yes | To perform cross-validation, we first allocate 10% of seen classes for a validation unseen data (for Aw A1 and Aw A2 we allocated 15% since there are only 40 seen classes). Then we allocate 10% out of the remaining 85% of the data for validation seen data. This means that in total we allocate 30% of all the seen data to perform validation. |
| Hardware Specification | Yes | All runs are made with the official hyperparameters and training setups and on the same hardware: NVidia Ge Force RTX 2080 Ti GPU, 16 Intel Xeon Gold 6142 CPU and 128 GB RAM. |
| Software Dependencies | No | The paper mentions using "Adam optimizer" but does not specify its version or the versions of any other software libraries (e.g., Python, PyTorch, CUDA) used for implementation. |
| Experiment Setup | Yes | For all the datasets, we train the model with Adam optimizer for 50 epochs and evaluate it at the end of training. We also employ NS and AN techniques with γ = 5 for NS. Additional hyperparameters are reported in Appx D. Table 4: Hyperparameters for ZSL experiments SUN CUB Aw A1 Aw A2 Batch size 128 512 128 128 Learning rate 0.0005 0.005 0.005 0.002 Number of epochs 50 50 50 50 Lent weight 0.001 0.001 0.001 0.001 Number of hidden layers 2 2 2 2 Hidden dimension 2048 2048 1024 512 γ 5 5 5 5 |