Dan Cher

I'm a PhD student at Washington University in Computational Data Sciences, where I have the honor of being an AI-ACCESS and Olin Chancellor's fellow. I'm working in the Multimodal Vision Research Laboratory led by Dr. Nathan Jacobs.

Prior to the PhD, I worked as a Data Scientist (6 YoE) tackling geospatial problems focusing on using machine learning to quantify the risk of where people live and drive. Also built some aerial imagery models. During that time I received my MS in Data Science from Worcester Polytechnic Institute.

Email CV GitHub Google Scholar LinkedIn

Research

I am interested in computer vision, multi-modal modeling and its applications to geospatial problems. My work has focused on satellite image synthesis and multimodal representation learning for geospatial and ecological problems.

A location encoder that replaces fixed bases (e.g., spherical harmonics) with a learnable Spherical Voronoi partition of the sphere, concentrating capacity where the world is complex; global semantic tokens share concepts across distant sites. New state of the art for location encoders.

Generalizes point-conditioned control to any geospatial primitive — points, bounding boxes, polylines, and polygons — through Geometry-Aware Local Attention (GALA).

Generates satellite imagery conditioned on point queries — spatial points paired with textual descriptions — enabling precise, annotation-efficient control without dense maps.

We introduce VectorSynth, a diffusion-based model that generates pixel-accurate satellite imagery from polygonal geographic annotations with semantic attributes, enabling spatially grounded, geometry- and language-guided image synthesis and editing with superior semantic and structural fidelity.

We introduce ProM3E, a probabilistic masked multimodal embedding model that enables any-to-any modality generation, fusion feasibility analysis, and superior cross-modal retrieval and representation learning through masked modality reconstruction in the embedding space.