Publications

Grouped by theme. For a complete list see Google Scholar.

Foundation models for biology

Models of single-cell sequencing data, and what they let you predict about perturbations.

Influence functions and robustness

What would this fit look like without that data point, or under a slightly different model? Answered without refitting.

Uncertainty quantification and calibration

Making a model's confidence mean something, and checking when it does not.

Bayesian neural networks and horseshoe priors

Heavy-tailed priors that shrink most units to nothing and let a few through, which turns model selection into inference.

Large language models: explanation, safety, adaptation

Explaining what a generative model did, catching what it should not do, and moving adapters between models.

Trans-LoRA: Towards Data-Free Transferable Parameter Efficient Finetuning

Runqian Wang, Soumya Ghosh, David Cox, Diego Antognini, Aude Oliva, Rogerio Feris, Leonid Karlinsky

NeurIPS, 2024

Health and disease progression

Latent-variable models for noisy clinical registries, and what they say about how disease unfolds.

Bayesian nonparametrics and segmentation

Letting the data decide how many parts there are, in images, in 3D meshes, and on lattices.

Federated learning and model fusion

Combining separately trained neural networks by matching their parameters.

Machine learning for materials science

Design and literature mining for lower-carbon concrete.