Research topics

How does the body's changing internal state shape behavior?
..and can we read it continuously, as it unfolds?

Physiological state is never static: blood chemistry, temperature, and activity fluctuate continuously across seconds to days, tracking health, sickness, and the demands of behavior. To study this directly, I designed and led the development of a chronic microdialysis pipeline that continuously measures blood chemistry in freely moving mice. This allows us to turn blood chemistry into a multi-dimensional time-series. From these recordings and movement data collected over days I reconstruct a physiological state space and ask how it shapes, and is shaped by, behavior. (See here: https://www.biorxiv.org/content/10.64898/2025.12.08.692974v1.full)

Current interests span methods from dynamical systems and statistics, such as state-space reconstruction, time-delay embedding, dimensionality reduction, and Bayesian inference to make sense of noisy high-D biosignals, including microdialysis Mass-Spec readings, but also Oxygen/CO2 ratios, glucose (CGM) and body-temperature. I'm increasingly interested in how these approaches carry over to human physiology where continuous monitoring is opening a similar window onto sleep, circadian rhythms, metabolism, and stress.

What's driving >90% of animal behavior?
Or, why do we need a brain that consumes much of our energy budget? 

Neuroscience has been trying to describe behavior as a mind-centered activity. But minds are not disembodied agents. From a survival perspective, physiological needs are one of the most fundamental behavioral drives, suggesting that internal physiological signals play a huge role in shaping neural dynamics. An intricate interplay exists between the brain and the body, acting over widely different timescales. From fast interoceptive spikes down to ultradian, circadian, and infradian hormonal cycles, various types of internal signals have been shown to affect behavior deeply. This suggests that the nervous system evolved together with the endocrine and immune systems to orchestrate physiological balance. 

One of the main questions that drive my research deals with the formalization of behavior and physiological balance within a formal framework. 

How are neural interactions within and across brain areas shaped?
And how do we measure/study them?

Local circuit interactions shape neural dynamics and computations. These are hard to study: direct synaptic connections can be detected only in a few cases from spiking data. Statistics can help characterize effective interactions representing the summed action of multiple synapses, neuromodulators, and long-range neural oscillations. These interactions can be measured from data by analyzing the second-order moments of the population activity. The traditional way to measure them is by partitioning neural co-variability into an explainable component, driven by external stimuli, and an unexplained component, or “noise” correlations, representing the amount of co-variability that cannot be explained by stimulus selectivity.

From an information theory perspective, these interactions can profoundly affect the encoding and transfer of information within and across brain areas. Moreover, they can change across experience, shedding light on internal novelty mechanisms and local-processing of stimulus-dependent inputs.

Relevant publications: https://doi.org/10.1523/JNEUROSCI.0194-23.2023 and https://doi.org/10.1016/j.celrep.2023.113015 

Theory of spiking neural networks
How do we make them more biologically realistic?

Biological neural networks carry out nonlinear computations crucial for survival. Moreover, biological cells can send signals over different distances and different timescales. From short to long, neurons within a local network can communicate via gap junctions, synaptic neurotransmitters, and intrinsic neuromodulation. More globally, network dynamics can be shaped by extrinsic neuromodulation (i.e., coming from the outside) and hormones. 

Current artificial neural networks can implement almost any nonlinear computation using only one type of synapse (usually linear integration of inputs passed through a non-linear threshold function); nonetheless, this requires extensive training, and the link between network connectivity and computation may be unclear.

How do we make this link clearer? One option is to ask directly, "What type of neural computation requires using a particular type of synapse, neuromodulator, or other forms of communication"?

Relevant publication: https://doi.org/10.24072/pcjournal.69, 

Take a look at the Publications page for previous projects.