Frederike H. Petzschner
A name the wiki had been carrying secondhand for two ingests before reading her first-hand. “Petzschner et al. 2017” is cited on computational-psychiatry and allostasis as the source of computational psychosomatics and of the precision-sets-the-force-of-regulation claim — both arriving through Khalsa et al. (2018). Petzschner et al. (2021) is the first paper in raw/ she leads, and it is created under the same load-bearing-figure convention used for Schandry, Katkin, Ehlers and Craske.
The Zurich lineage, and what it brings
She writes from the Translational Neuromodeling Unit — Klaas Enno Stephan’s group, thanked in the paper’s acknowledgments along with Lilian Weber — which is a different intellectual address from every other computational source in this wiki. Seth and Friston come at interoception from consciousness science and the free-energy principle; Barrett from the psychology of emotion; Khalsa and Paulus from clinical psychiatry. The TNU’s programme is generative modelling as a clinical instrument: fit a model to an individual’s data, read the parameters, use them for differential diagnosis. That orientation is visible in what the 2021 review chooses to be careful about.
The methodological signature: keep Marr’s levels apart
Her distinctive move, and the thing that makes the 2021 paper more than a survey, is insisting on the separation between what a system computes and how it computes it. Bayesian inference is a computational-level claim and functions as an ideal-observer benchmark; predictive-coding is one algorithmic proposal for approximating it, among others. Most of the wiki’s predictive material blurs these, treating “the brain does predictive coding” and “the brain approximates Bayesian inference” as one claim. Her review holds them apart and reports, drily, that the interoceptive network implementing the second has not been identified.
The same discipline produces the paper’s most useful result: HRL and interoceptive active-inference are equivalent at the computational level (drive re-expresses as surprise) and differ at the implementational one — so the way to arbitrate between them is anatomical, not mathematical.
Computational psychosomatics
Her own programme, and the reason the wiki cited her before it read her. The claim is that psychosomatic phenomena — symptoms without proportionate peripheral pathology, and placebo effects — fall out of the precision structure of inference over bodily states: a body-state prior held with pathologically high precision drives over-vigorous regulatory action, because “the tighter the expected range of bodily state, the more vigorous the elicited regulatory action.” The deliverable is a computational biomarker: a parameter estimate, read like a blood test, pointing at an individual’s specific maladaptive computation.
She is notably unwilling to oversell it. The 2021 review states that there is “little empirical work testing such models’ predictions” and that clinical relevance requires a drug-development-style translational pipeline with the biomarker demonstrating utility “ideally in individual patients.” One of the few papers she can cite as such a test is her own (the heartbeat-evoked-potential/attention study), which makes the point about scarcity rather than refuting it.
Placement
The modeller of the wiki’s computational cluster, as Khalsa is its convener and Paulus its clinician. She is the source of the wiki’s two most deflationary computational claims — that predictive coding is an implementation hypothesis with live alternatives, and that body regulation may not require an interocept at all — and she is on record making both while working inside the framework they deflate. That combination (committed practitioner, careful about the framework’s actual evidential standing) is the same posture the roadmap takes with its “indirect so far,” and it is why both documents are usable here as status reports rather than advocacy.