Robust generalization to new concepts has long remained a distinctive feature of human intelligence. However, recent progress in deep generative models has now led to neural architectures capable of synthesizing novel instances of unknown visual concepts from a single training example. Yet, a more precise comparison between these models and humans is not possible because existing performance metrics for generative models (i.e., FID, IS, likelihood) are not appropriate for the one-shot generation scenario. Here, we propose a new framework to evaluate one-shot generative models along two axes: sample recognizability vs. diversity (i.e., intra-class variability). Using this framework, we perform a systematic evaluation of representative one-shot generative models on the Omniglot handwritten dataset. We first show that GAN-like and VAE-like models fall on opposite ends of the diversity-recognizability space. Extensive analyses of the effect of key model parameters further revealed that spatial attention and context integration have a linear contribution to the diversity-recognizability trade-off. In contrast, disentanglement transports the model along a parabolic curve that could be used to maximize recognizability. Using the diversity-recognizability framework, we were able to identify models and parameters that closely approximate human data.

Y et, a more precise comparison between these models and humans is not possible because existing performance metrics for generative models (i.e., FID, IS, likelihood) are not appropriate for

Robust generalization to new concepts has long remained a distinctive feature of human intelligence. However, recent progress in deep generative models has now led to neural architectures capable of synthesizing novel instances of unknown visual concepts from a single training example. Yet, a more precise comparison between these models and humans is not possible because existing performance metrics for generative models (i.e., FID, IS, likelihood) are not appropriate for the one-shot generation scenario. Here, we propose a new framework to evaluate one-shot generative models along two axes: sample recognizability vs. diversity (i.e., intra-class variability). Using this framework, we perform a systematic evaluation of representative one-shot generative models on the Omniglot handwritten dataset.

The ability to generate synthetic sequences is crucial for a wide range of applications, and recent advances in deep learning architectures and generative frameworks have greatly facilitated this process. Particularly, unconditional one-shot generative models constitute an attractive line of research that focuses on capturing the internal information of a single image or video to generate samples with similar contents. Since many of those one-shot models are shifting toward efficient non-deep and non-adversarial approaches, we examine the versatility of a one-shot generative model for augmenting whole datasets. In this work, we focus on how similarity at the subsequence level affects similarity at the sequence level, and derive bounds on the optimal transport of real and generated sequences based on that of corresponding subsequences. We use a one-shot generative model to sample from the vicinity of individual sequences and generate subsequence-similar ones and demonstrate the improvement of this approach by applying it to the problem of Unmanned Aerial Vehicle (UAV) identification using limited radio-frequency (RF) signals. In the context of UAV identification, RF fingerprinting is an effective method for distinguishing legitimate devices from malicious ones, but heterogenous environments and channel impairments can impose data scarcity and affect the performance of classification models. By using subsequence similarity to augment sequences of RF data with a low ratio (5%-20%) of training dataset, we achieve significant improvements in performance metrics such as accuracy, precision, recall, and F1 score.