Reproduce “NeuralCubes: Deep Representations for Visual Data Exploration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhe Wang, Dylan Cashman, Mingwei Li, Jixian Li, Matthew Berger, Joshua A. Levine, Remco Chang, Carlos Scheidegger
NeuralCubes introduces a method to replace traditional data cubes with deep neural networks, enabling real-time, client-side visual exploration of large datasets. The approach compresses multi-gigabyte datasets into small, sub-megabyte models (e.g., 1.8GB to 798KB) while maintaining low prediction error (RAE often below 5%) and revealing semantically meaningful latent spaces for novel interactions.
Key insights
- Aggregate database queries can be effectively learned as functions by deep neural networks.
- Deep neural networks can serve as compact, client-side proxies for large, server-side data infrastructure.
- Incorporating autoencoders allows the network to learn semantically meaningful 2D latent spaces of attributes, enabling new forms of data exploration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.