I am a senior ML scientist at Diadia Health.
Before this, I was a postdoc in the Machine Learning & Analytics Group at Lawrence Berkeley National Laboratory, doing scientific ML (e.g., physics-informed neural networks) and interpretability (e.g., developing new ways to visualize loss landscapes and characterize learning).
I did my PhD in Biophysics at Stanford, where I developed new computational methods (using ML and topological data analysis) to study the brain’s dynamical organization during ongoing cognition.
C. Geniesse*, J. Chen*, T. Xie*, G. Shi, Y. Yang, D. Morozov, T. Perciano, M. Mahoney, R. Maciejewski, G.H. Weber. Visualizing Loss Functions as Topological Landscape Profiles. NeurIPS Workshop on Symmetry and Geometry in Neural Representations (NeurReps) Proceedings Track, 2024. [pdf] (*equal contribution)
T. Xie*, C. Geniesse*, J. Chen*, Y. Yang, D. Morozov, M. Mahoney, R. Maciejewski, G.H. Weber. Evaluating Loss Landscapes from a Topology Perspective. NeurIPS Workshop on Scientific Methods for Understanding Deep Learning (SciForDL), 2024. [pdf] (*equal contribution)
T. Xie*, J. Chen*, Y. Yang*, C. Geniesse*, G. Shi, A. Chaudhari, J.K. Cava, M.W. Mahoney, T. Perciano, G.H. Weber, R. Maciejewski. LossLens: Diagnostics for Machine Learning Through Loss Landscape Visual Analytics. IEEE Computer Graphics and Applications, 2024. [pdf] (*equal contribution)
C. Geniesse*, S. Chowdhury*, M. Saggar. NeuMapper: A Scalable Computational Framework for Multiscale Exploration of the Brain’s Dynamical Organization. Network Neuroscience, 2022. [code] (*equal contribution)
C. Geniesse, O. Sporns, G. Petri, M. Saggar. Generating Dynamical Neuroimaging Spatiotemporal Representations (DyNeuSR) Using Topological Data Analysis. Network Neuroscience, 2019. [code] [demos]
Z. Wu, B. Ramsundar, E. Feinberg, J. Gomes, C. Geniesse, A. Pappu, K. Leswing, V. Pande. MoleculeNet: A Benchmark for Molecular Machine Learning. Chemical Science, 2018.
NeuMapper. A scalable Mapper algorithm for neuroimaging data analysis. The Matlab implementation was designed specifically for working with complex, high-dimensional neuroimaging data and produces a shape graph representation that can be annotated with meta-information and further examined using network science tools.
Reciprocal Isomap. A reciprocal variant of Isomap for robust non-linear dimensionality reduction in Python. The ReciprocalIsomap transformer was inspired by scikit-learn’s implementation of Isomap, but the reciprocal variant enforces shared connectivity in the underlying k-nearest neighbors graph (i.e., two points are only considered neighbors if each is a neighbor of the other).
Landmark Cover. A Python implementation of NeuMapper’s landmark-based cover. The LandmarkCover transformer was designed for use with KeplerMapper, but rather than dividing an extrinsic space (e.g., low-dimensional projection) into overlapping hypercubes, the landmark-based approach directly partitions data points into overlapping subsets based on their intrinsic distances from pre-selected landmark points.
DyNeuSR. A Python visualization library for topological representations of neuroimaging data. The package combines visual web components with a high-level Python interface for interacting with, manipulating, and visualizing topological graph representations of functional brain activity.