We propose a novel hierarchical soft mixture-of-experts for 3D medical image segmentation

A new segmentation framework and sampling mechanism to produce multiple consistent label maps across similar images on held-out datasets, outperforming baselines by a wide margin.

We present PolyPose, a simple locally-linear method for deformably aligning a high-resolution preoperative 3D volume to as few as two intraoperative 2D X-rays without explicit regularizers or hyperparameter optimization.

We introduce generative distribution embeddings (GDE), a framework that lifts autoencoders to the space of distributions.

State-of-the-art protein language models, under standard usage, cannot infer evolutionary relationships so we introduce Phyla, a model explicitly trained for evolutionary reasoning that achieves state-of-the-art phylogenetic performance.

We propose methods for studying behavior and neural dynamics in deep RL agents based on neuroscience methods and demonstrate them in a naturalistic simulated foraging environment

We introduce a brain decoding framework that maps fMRI data to interpretable semantic concepts for image reconstruction. This pipeline maintains high reconstruction quality, revealing voxel-level concept activations, enabling neuroscientific insight.

DQVis is a new dataset of natural language queries to biomedical visualizations and a framework for generating similar domain specific datasets.

We derive well-known learning rules from an objective that casts learning rules as policies for navigating uncertain loss landscapes