Tag
This paper presents 3D masked autoencoders for volumetric microscopy data, demonstrating that 3D modeling outperforms 2D max-projection and slice-based variants on downstream single-cell tasks, with cross-modal alignment to a protein language model further improving performance.
OpenAI Microscope is an open-source tool that systematically visualizes every neuron in commonly studied vision models with fast feedback loops and linkable neurons to support interpretability research. The platform reduces visualization time from minutes to seconds and aims to make neural network analysis more accessible to the research community.