The brain as a probabilistic feedback controller

What gets you out of bed in the morning? Everything we do in life has a cost to our body budget, and yet we do so much more with our lives than survive as minimally as possible. Our situations change moment to moment, and ensuring our well-being across those changes requires predictive regulation: allostasis.

In every moment, our brains need to infer, from nothing but experience and sensorimotor data: who am I, how do I feel, what’s going on, and what can I do about it? Solving this problem requires the brain to simulate your body and sensorium, and infer how to control the body by predictive coding.

I am a Postdoctoral Research Fellow in applied deep learning at Vanderbilt University. I work in the Bastos Lab with the titular Andre Bastos, after finishing my PhD in the Probabilistic Modeling Lab with Jan-Willem van de Meent and the Interdisciplinary Affective Science Lab with Lisa Feldman Barrett and Karen Quigley.

Check out the equation that makes it happen! \(\varepsilon_{z} := \nabla_{z} \log \gamma_{\theta}(z \mid \mathbf{z}_{\setminus z}) = \nabla_{z} \log p_{\theta}(z \mid \mathrm{Pa}(z)) + \sum_{v \in \mathrm{Ch}(z)} \nabla_{z} \log p_{\theta}(v \mid \mathrm{Pa}(v))\)