NADS: Neural Architecture Distribution Search for Uncertainty Awareness
Authors: Randy Ardywibowo, Shahin Boluki, Xinyu Gong, Zhangyang Wang, Xiaoning Qian
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform multiple Oo D detection experiments and observe that our NADS performs favorably, with up to 57% improvement in accuracy compared to state-of-the-art methods among 15 different testing configurations. |
| Researcher Affiliation | Academia | 1Department of Electrical and Computer Engineering, Texas A&M University, College Station, Texas, USA 2Department of Computer Science and Engineering, Texas A&M University, Col lege Station, Texas, USA. |
| Pseudocode | No | No explicit pseudocode or algorithm blocks were found. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | We applied our architecture search on five datasets: Celeb A (Liu et al.), CIFAR-10, CIFAR-100, (Krizhevsky et al., 2009), SVHN (Netzer et al., 2011), and MNIST (Le Cun). |
| Dataset Splits | No | We can gradually remove the continuous relaxation and sample discrete architectures by annealing the tem perature parameter τ, allowing us to perform architecture search without using a validation set. |
| Hardware Specification | Yes | With this setup, we found that we are able to identify neural architectures in less than 1 GPU day on an Nvidia RTX 2080 Ti graphics card. |
| Software Dependencies | No | The paper mentions using 'Adam optimizer' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | In all experiments, we used the Adam optimizer with a fixed learning rate of 1 10 5 with a batch size of 4 for 10000 iterations. We approximate the WAIC score using M = 4 architecture samples, and set the tempera ture parameter τ = 1.5. We then retrained each architecture for 150000 iterations using Adam with a learning rate of 1 10 5 . |