I am a theorist – a postdoctoral fellow at the Computational Neuroscience Initiative at the University of Pennsylvania – who is perplexed by the nature and behavior of space, time, and brains. As of September 2019, I will be an assistant professor in the Department of Physics at New York Institute of Technology! Here’s my CV.
- the neural basis of pattern generation (how intrinsically chaotic neurons can – via network connectivity – be reined into delivering reliable output, where the network activity results in a stable macroscopic behavior);
- predictive models of functional biological neurons and networks that are associated with the creation and reception of auditory signals;
- the information content of acoustic communication signals;
- nonlinear processes in neutrino astrophysics.
- dynamical systems, inference, statistical physics, information theory.
Regarding pattern generation by the central nervous system for acoustic signal processing: I’m interested in ferreting out fundamental organizing principles of the central nervous system, particularly those that give rise to reliable patterned neural activity associated with acoustic information. Nearly all research on neuronal circuits controlling pattern generation (or CPG activity) has been done on small (~ 30-cell) circuits in crustaceans, because these circuits can be identified and isolated from the animal and the relatively-large cell size facilitates whole-cell recording. Extremely little modeling has explored how principles governing CPG activity are effected when the number of neurons involved is much larger. Nucleus HVC of the songbird brain is an example of a larger-scale (~ 10^5-cell) circuit with a well-demonstrated ability to generate reliable patterned activity. I have created a toy model of the pattern-generating mechanism (Armstrong & Abarbanel, J. Neurophysiol. 2016), where I used state and parameter estimation to infer electrophysiological properties. Currently I am working with colleagues who study zebra finch learning (Ofer Tchernichovski – Hunter College, CUNY; Julia Hyland Bruno – Columbia U.; Tiberiu Teliseanu – Flatiron Institute), to expand this model in terms of its learning capacity and connections to other areas of the song system associated with song timing.
Regarding the information content of acoustic signals: Female cowbirds display preferences for songs of particular males, given song as the only information about each male (preferences are measured via copulation solicitation display). More intriguing is the observation that isolation from other females will each rank a set of songs in an identical order. Further, following lesions to the female song circuit, these preferences dissolve. The failure of traditional acoustic analysis tools to probe female preferences in unsurprisin: vocal production is a nonlinear process, and meanwhile acoustic studies are based mostly on linear spectral analysis. We approach this problem by assuming that some known nonlinear dynamical system produced the song. We employ time-delay embedding to reconstruct the song’s attractor, where the dimension of the attractor space is the number of time-delayed coordinates required for unfolding. Currently we are working to classify cowbird songs based on neural network training upon these attractors. The learning algorithm is able to synthesize the waveforms of the songs, given the time-delayed coordinates as input data. We are building a list of candidate synthesized songs for playback to females, with the aim of identifying predictors of female song preferences. Collaborators: Alicia Zeng, David White, and Andrew Gersick.
Regarding inference for unveiling the role of song in a behavioral setting: During mating season, most songbird species engage in a societal evolution wherein monogamous pairs “freeze out”, presumably for successful procreation. The means by which all individuals “agree” on this structure is unknown, although the role of song is known to be significant. Further, electrophysiological manipulations of the song circuit disrupt pair bonding. We aim to characterize a relationship between vocalizations and pair bonding, in advance of more targeted electrophysiological manipulations that are planned for spring 2019. We tackle this problem with a maximum-entropy approach, using an Ising model. The inferred parameters, trained on instances of song, outperforms the correlations themselves in indicating which song-related interactions contain information about the pair-bonding structure. The Ising model fails to capture all of the important structure in the data, suggesting that triadic interactions are important; this would be a quantification of a longstanding speculation within the experimental community that triadic interactions guide social dynamics. Moreover, we are finding that the language of statistical physics can offer new insight into the biological motivations for songbird social structure. Finally, we are comparing the Ising model results to a method of model identification. We find that a coordinate basis involving certain bird triads captures more of the structure in the data than does a basis in which the vectors are simply the individual birds. Collaborators: Marc Schmidt, David White, Vijay Balasubramanian, Ammon Perkes, Clelia de Mulatier, Luke Anderson.
In astrophysics, I apply inference to nonlinear problems in neutrino flavor evolution. We seek to ascertain what information a earth-based detector must receive in order to infer the flavor evolution history of neutrinos that have emanated from a supernova (SN) event. This is an important question for cosmology, as flavor evolution following SN core collapse in part sets the heavy-element abundances. We have found that when considering forward-scattering events only (and not energy-changing events), a method of state and parameter estimation can infer the complete flavor evolution history, given measurements only at the detector location. Currently we are examining how the result is affected by the addition to the model of direction-changing scattering. In parallel, we are designing a more complicated model of spin-spin interactions to investigate the possibility of frustration in the neutrinos field, and whether the flavor spectrum may have multiple final states (in the low-temperature, or vacuum limit). We use a Monte Carlo search within a maximum-entropy inference framework to sample the probabilities of final flavor states given multiple initial conditions. Preliminary results indicate that there indeed exist multiple minima whose probabilities are similar to that of the global minimum. This finding suggests that a measurement made at an earth-based detector has multiple (degenerate) histories. Collaborators: George Fuller, Amol Patwardhan, Baha Balantekin, Chad Kishimoto, Luke Johns, Shashank Shalgar, Mark Paris.
Previously I was a postdoc at the BioCircuits Institute at the University of California, San Diego. There I worked with Henry Abarbanel on a dynamical model of the avian song nucleus HVC. We tested models via methods of statistical data assimilation, using experimental data from collaborators in the laboratory of Daniel Margoliash at the University of Chicago.