Open-Set Recognition with Gaussian Mixture Variational Autoencoders
Authors: Alexander Cao, Yuan Luo, Diego Klabjan6877-6884
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | With this, our Gaussian mixture variational autoencoder (GMVAE) achieves more accurate and robust open-set classification results, with an average F1 increase of 0.26, through extensive experiments aided by analytical results.Finally in 5, we conduct open-set classification experiments on three standard datasets. |
| Researcher Affiliation | Academia | Alexander Cao,1 Yuan Luo,2 Diego Klabjan1 1Department of Industrial Engineering and Management Sciences 2Department of Preventive Medicine Northwestern University a-cao@u.northwestern.edu, {yuan.luo, d-klabjan}@northwestern.edu |
| Pseudocode | Yes | Algorithm 1: Nearest centroid thresholding on distance to the nearest centroid |
| Open Source Code | No | We will publish our code upon acceptance of this paper. |
| Open Datasets | Yes | Finally in 5, we conduct open-set classification experiments on three standard datasets. Our findings from experiments are two-fold. First, GMVAE outperforms a state-of-the-art classification-reconstruction-based, deep open-set classifier both in terms of accuracy and robustness to an increasing number of unknown classes.Second, the use of extreme value theory (EVT) to infer classbelongingness (Bendale and Boult 2016; Yoshihashi et al. 2019) may be ill-suited in this classification-reconstruction open-set framework as we find that ours and another simple algorithm consistently beat it. |
| Dataset Splits | Yes | The training data has only labeled samples from the C known classes. The validation set also only has samples from the same C classes. The validation set is used to determine the threshold τ.Finally, the test set has samples from the C known classes and samples from additional Q unknown classes, which are all treated as class C + 1. For each of the experiments below, we perform an ablation study. Four combinations of model and classification algorithms were applied: (i) CROSR with CROSR s EVT (CROSR+EVT), (ii) CROSR with Algorithm 1 (CROSR+NC-D), (iii) GMVAE with Algorithm 1 (GMVAE+NC-D), and (iv) GMVAE with Algorithm 2 (GMVAE+NC-U). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments. It only mentions general aspects of training. |
| Software Dependencies | No | The paper mentions using 'Adam' for optimization but does not provide specific software dependencies with version numbers (e.g., Python version, library versions like PyTorch or TensorFlow) required for reproducibility. |
| Experiment Setup | Yes | The latent space dimension of z equals 10, 50, 5, and 20 for the four experiments. A table of GMVAE network architectures for each experiment can be found in the technical appendix. We optimize over the training set using Adam until the loss, evaluated on the known validation set, plateaus. |