Skip to content

Reward learning, and whether you can measure it from outside the skull.

Three positions, one question. Below: the benchmark that motivates all of it, then the projects, then everything written or presented.

Anhedonia classification accuracy, voice versus fMRIArea under the ROC curve for six classifiers, with 95 percent bootstrap confidence intervals, plotted against a chance line at 0.50. The three voice models, trained on acoustic features from clinical interview audio, score 0.61 to 0.65 and sit above chance. The three fMRI models, trained on nucleus accumbens activation, score 0.37 to 0.45 and sit at or below chance. The voice models therefore outperform the brain-imaging models on this benchmark.0.30.40.50.60.70.8CHANCEVOICERandom forest0.63Gradient boosted0.62Logistic regression0.56fMRILogistic regression0.58Random forest0.52Gradient boosted0.52

Voice cleared chance, barely (random forest, AUC 0.63, permutation p = .049). Ventral striatal BOLD did not (logistic regression, AUC 0.58, p = .057). The two are statistically non-inferior to each other, which is a statement about how modest both are.

Fig. — AUC-ROC with 95% bootstrap CI, stratified 5-fold CV. Primary preregistered analysis. Voice: eGeMAPSv02, diarized participant speech · DAIC-WOZ, n=142. fMRI: Nucleus accumbens BOLD, BART · ds000030, n=234. The voice result is borderline: p = .049 uncorrected, and it does not survive Bonferroni correction for three classifiers (α/3 = .017). Independent cohorts and different anhedonia instruments (PHQ-8 items 1–2 vs Chapman), so this is a benchmark across datasets, not a within-subject comparison. Both streams overfit heavily (train AUC 1.0, gap > 0.35) on 447 features and 32 positive cases. Stream B is also pipeline-dependent: fMRIPrep 23.x gives 0.45, below chance. Preregistered at osf.io/4d6ey.
ModelRead Lab · USC

A Rescorla-Wagner simulation of how reward learning collapses when the dopaminergic learning rate falls and effort stops being worth spending.

SignalItti Lab · USC Viterbi

Whether that collapse is audible. An open-source pipeline that pulls acoustic biomarkers out of clinical interviews without the audio ever leaving the room.

CircuitNIMH · Experimental Therapeutics

What the circuit is doing while it happens. MEG source localization, gamma-band power, and signal complexity in the insula as a marker of suicidal ideation.

  • Building source-localization pipelines for resting-state MEG analyses of mood-disorder neural biomarkers.
  • Working on gamma-band spectral power and signal complexity in the insula as candidate markers of suicidal ideation.

PRAXIS

Psychiatry Research, Analytics & eXperimental Innovation Society

From theory into tools that reach patients.

An undergraduate research group at USC working where machine learning meets clinical mental health. I founded it in February 2026 and lead it: we run original studies, preregister them, and publish what we find including the results that come back null.

Founded
Feb 2026
Host
USC Dornsife
Sponsor
Dr. Laurent Itti
My role
Founder, project lead
  • Original research applying machine learning to clinical populations, preregistered on OSF before the data are touched.
  • A twenty-paper journal-club curriculum taking undergraduates through reinforcement learning and active inference.
  • A biweekly speaker series, including researchers from the NIMH.
  • Open-source code and computational psychiatry resources, so the methods can be checked by anyone who wants to check them.

Software

ClinicalWhisperJan 2026 – present

Local-first, air-gapped clinical interview transcription and speech-biomarker pipeline. Whisper for transcription, pyannote for diarisation, OpenSMILE for eGeMAPS features.

  • Architected fully air-gapped so audio never leaves the machine, meeting IRB and HIPAA constraints.
  • Open-sourced; in use across PRAXIS research streams and adopted externally by the Rutledge Lab at Yale in August 2026.

Manuscripts and presentations

  1. 2026Preprint
  2. In preparationIn preparation

    Composable AI primitives for visual assistive devices: A needs-based review informed by 291 blind and visually impaired users

    Zhou, C., et al. · Manuscript in preparation

  3. Jul 2026Talk / poster

    Cross-Modal Benchmarking of Acoustic Prosody and Ventral Striatal BOLD for Depression-Related Anhedonia Classification

    Zhou, C. · Computational Psychiatry Conference, Yale University

  4. Apr 2026Talk / poster

    The dopaminergic voice: Acoustic prosody vs. ventral striatal BOLD activation for anhedonia classification

    Zhou, C. · 27th USC Undergraduate Symposium

  5. Nov 2025Talk / poster

    Attentive AI visual aid for persons with visual impairment

    Zhou, C. · Trojan Research Association Undergraduate Symposium