A Generative Adversarial Network (GAN) involves Generator (G) and Discriminator (D) networks whose purpose, respectively, is to map random noise to samples and discriminate real and generated samples. Formally, the GAN objective, in its ori…With the default suggested $\gamma$ strength of 10, training becomes stable and improves the smoothness and boundedness of spectra in both G and D, but performance severely degrades, resulting in a 45% reduction in IS. Reducing the penalty …Published as a conference paper at ICLR 2019
Caption: Figure 11: Nearest neighbors in ResNet-50-avgpool (He et al., 2016) feature space. The generated image is in the top left.
Caption: Figure 12: Nearest neighbors in pixel space. The gener…
Ishaan Gulrajani
scientist · 3 mentions across 1 reading
In this course
Gulrajani is a machine learning researcher known for work on stabilizing and improving generative adversarial networks, particularly through techniques like gradient penalty that address training instability in GANs. The course readings cite him in the context of tuning hyperparameters (like penalty strength gamma) that balance training stability against model performance, illustrating the practical tensions in adversarial training. His ICLR 2019 publication appears in the material examining how generated images relate to learned feature spaces, making him relevant to discussions of what GANs learn and how to evaluate their outputs.
Mentioned in 1 reading
Appears alongside
People mentioned in the same passages — sorted by co-occurrence weight.