Generalizing to Unseen Domains via Adversarial Data Augmentation
Authors: Riccardo Volpi, Hongseok Namkoong, Ozan Sener, John C. Duchi, Vittorio Murino, Silvio Savarese
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our approaches on a simple digit recognition task, and a more realistic semantic segmentation task across different seasons and weather conditions. In both settings, we observe that our method allows to learn models that improve performance across a priori unknown target distributions that have varying distance from the original source domain. We evaluate our method for both classification and semantic segmentation settings, following the evaluation scenarios of domain adaptation techniques [9, 39, 14], though in our case the target domains are unknown at training time. |
| Researcher Affiliation | Collaboration | Riccardo Volpi , Istituto Italiano di Tecnologia Hongseok Namkoong Stanford University Ozan Sener Intel Labs John Duchi Stanford University Vittorio Murino Istituto Italiano di Tecnologia Università di Verona Silvio Savarese Stanford University |
| Pseudocode | Yes | Algorithm 1 Adversarial Data Augmentation |
| Open Source Code | Yes | Models were implemented using Tensorflow, and training procedures were performed on NVIDIA GPUs. Code is available at https://github.com/ricvolpi/generalize-unseen-domains |
| Open Datasets | Yes | Digit classification We train on MNIST [19] dataset and test on MNIST-M [9], SVHN [30], SYN [9] and USPS [6]. Semantic scene segmentation We use the SYTHIA[31] dataset for semantic segmentation. |
| Dataset Splits | No | We use 10, 000 digit samples for training and evaluate our models on the respective test sets of the different target domains, using accuracy as a metric. The paper mentions using a training set and test sets but does not provide specific details for a validation split for reproducibility. |
| Hardware Specification | No | Models were implemented using Tensorflow, and training procedures were performed on NVIDIA GPUs. The paper mentions 'NVIDIA GPUs' but does not specify exact models or other detailed hardware components like CPU or memory. |
| Software Dependencies | No | Models were implemented using Tensorflow, and training procedures were performed on NVIDIA GPUs. While 'TensorFlow' is mentioned, no specific version number is provided for it or any other software dependency. |
| Experiment Setup | Yes | We set the hyperparameters α = 0.0001, η = 1.0, Tmin = 100 and Tmax = 15. In the minimization phase, we use Adam [17] with batch size equal to 32. For semantic scene segmentation: We set the hyperparameters α = 0.0001, η = 2.0, Tmin = 500 and Tmax = 50. For the minimization phase, we use Adam [17] with batch size equal to 8. |