Enabling Dark Energy Science with Deep Generative Models of Galaxy Images
Authors: Siamak Ravanbakhsh, Francois Lanusse, Rachel Mandelbaum, Jeff Schneider, Barnabas Poczos
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we propose an alternative to the expensive acquisition of more high quality calibration data using deep conditional generative models. In recent years, these models have achieved remarkable success in modeling complex high-dimensional distributions, producing natural images that can pass the visual Turing test. Two prominent approaches for training these models are variational autoencoder (VAE) (Kingma and Welling 2013; Rezende, Mohamed, and Wierstra 2014) and generative adversarial network (GAN) (Goodfellow et al. 2014). Our aim is to train a coditional variation of these models using existing HST data and generate new galaxy images conditioned on statistics of interest such as the brightness or size of the galaxy. This will allow us to synthesize calibration datasets for specific galaxy populations, with objects exhibiting realistic morphologies. |
| Researcher Affiliation | Academia | 1. School of Computer Science, Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213, USA 2. Mc Williams Center for Cosmology, Department of Physics, Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213, USA |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | As our main dataset, we use the COSMOS survey to build a training and validation set of galaxy images and extract from the corresponding catalog a condition vector y with three features: half-light radius (measure of size), magnitude (measure of brightness) and redshift (cosmological measure of distance). ... The largest current survey being used for image simulation purposes is the COSMOS survey (Scoville et al. 2007), carried out using the Hubble Space Telescope (HST). ... We also use the GALAXY-ZOO dataset (Willett et al. 2013) to demonstrate the abilities of our alternative conditional adversarial objective. |
| Dataset Splits | No | The paper mentions using a "training and validation set" and shows validation results in figures (e.g., Figure 5), but it does not specify the exact percentages or counts for these dataset splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions software like "Gal Sim package" and optimization algorithms like "Adam", but it does not provide specific version numbers for any libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | For optimization, we use Adam (Kingma and Ba 2014) with reduced exponential decay rate of .5 for the first moment estimates. ... For this dataset we used 5-layer fully (de)convolutional generator and predictor, mini-batch discrimination, batch-normalization and tanh activation function for the final layer of the generator. ... For better mini-batch statistics, we use relatively larger mini-batches with 128/256 instances. |