I study representation learning for dynamical systems with applications to the life sciences, from the genetic to the behavioral levels. Some major currents include the unsupervised inference of governing equations and bifurcation prediction in cell cycle dynamics. Other recent work has explored machine learning methods for optimal control, particularly of systems of coupled oscillators, which we have used to model the cortical processes underlying higher visual cognition. Here is an overview of ongoing and former work:

Current Projects
Large-scale representation learning
for dynamical systems
images-static/dummy image
Can we learn a uiversal feature dictionary for the classification, design and control of dynamical systems? [Github]

Related Work:
Learning the latent
structure of biological PDEs
images-static/dummy image
Can we learn latent features (left) for the automated design (right) of biological patterning systems?

Related Work:
  • Ricci, M.G., Moriel, N., Nitzan, M. Spatial Phase2vec: Learning a PDE embedding space with applications to reaction-diffusion models (In preparation)
Embedding ethological dynamics
images-static/dummy image
Can we learn compact representation of ethological dynamics that tell us about animal injury and disease? This is joint work with the lab of Victoria Abraira .

Related Work:

Earlier Projects
Machine learning for
adaptive network dynamics
images-static/dummy image
Given a population of networked agents with random features, can we predict
how to connect them to achieve an optimal dynamics? [Github]

Related Work:
Understanding the neural dynamics
of visual reasoning
How do machines reason about images compared to humans, and what
principles of neural dynamics underlie this ability?

Related Work: