Skip to content

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

Three positions, one question. The benchmark figure is on the home page; Table 1 gives its numbers. Then the projects, then everything written or presented.

Table 1. Classification of anhedonia, AUC-ROC with 95 percent bootstrap CI.
StreamClassifierAUC95% CIPermutation p
VOICEn = 142Random forest0.630.51 to 0.74.049
Gradient boosted0.620.50 to 0.73
Logistic regression0.560.45 to 0.67
fMRIn = 234Logistic regression0.580.51 to 0.66.057
Random forest0.520.45 to 0.60
Gradient boosted0.520.45 to 0.60
Table 1 Primary preregistered analysis. Permutation p for each stream's best model only, 1,000 label permutations, uncorrected. Chance AUC = 0.50.
Cite the preprint
@article{zhou2026crossmodal,
  author  = {Zhou, C. and Wu, M. and Xiang, Y. and Itti, L.},
  title   = {Cross-Modal Benchmarking of Acoustic Prosody and Ventral Striatal BOLD for Depression-Related Anhedonia Classification: A Pre-Registered Study with the ClinicalWhisper Pipeline},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.64898/2026.06.08.728970},
  note    = {Preprint, not peer reviewed}
}
ModelRead Lab · USC

A Rescorla-Wagner simulation of how anhedonia breaks reward learning: choices that stop following learned value cost far more than slower learning.

  • Built a Rescorla-Wagner reinforcement learning simulation with self-derived parameters (α as dopaminergic learning rate; β as inverse temperature, how strongly learned values drive choice), modelling how reward learning collapses under anhedonia.
  • Developed a formal parameter framework mapping circuit-level variables (LHb-VTA pathway) to RL parameters and their behavioural predictions, and simulated circuit-level treatment effects (ketamine, pramipexole, rTMS, psilocybin).
  • Synthesised literature on the LHb-VTA pathway and mTOR signalling to support hypothesis generation for studies on rapid-acting antidepressants such as ketamine.
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.

  • Designed and deployed a fully accessible survey pipeline using Qualtrics logic flows and custom scripting, ensuring WCAG compliance and collecting 493 responses from blind and visually impaired adults (291 retained after a two-stage quality-control filter).
  • First author of a review paper in preparation on vision-based assistive technology, informed by the survey, with collaborators now at Georgia Tech.
  • Led a pre-registered benchmark of acoustic prosody versus ventral-striatal fMRI for anhedonia classification, resulting in a first-author bioRxiv preprint (2026).
  • Coordinate and moderate IRB-approved Zoom focus groups with blind and visually impaired adults (pilot May 2026; weekly tier-based sessions in Fall 2026). Findings will inform live prototype testing of the AI assistive device.
CircuitNIMH · Experimental Therapeutics

What the circuit is doing while it happens. MEG source localization and Lempel-Ziv signal complexity in the insula, explored as a correlate of suicidal thoughts.

  • Built a Lempel-Ziv complexity pipeline for source-localized MEG data from 57 patients with major depressive disorder (MNE-Python on NIH Biowulf, SWARM-parallelized), testing broadband and six frequency bands at the right subgenual ACC and right anterior insula.
  • Exploratory analyses, not corrected for multiple comparisons, linked insula broadband complexity to MADRS severity and insula delta-band complexity to a suicidal-thoughts factor. The work contributes to a planned multimodal NIMH study (manuscript in preparation; middle author).

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, director
  • Original research applying machine learning to clinical populations, preregistered on OSF before the data are touched.
  • A twenty-paper journal-club curriculum taking undergraduates from the founding papers of the field through to algorithmic bias.
  • 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 pipeline for sensitive clinical interview audio: joint transcription and diarization, OpenSMILE eGeMAPSv02 acoustic features, and local LLM scoring that is documented as not yet clinically validated.

  • Designed for fully on-device processing, so audio, transcripts, and results never leave the machine, supporting IRB data-privacy requirements.
  • Open-sourced on GitHub (Zenodo DOI 10.5281/zenodo.20559786). Used across the PRAXIS research streams and being trialled by a Yale lab on long clinical interviews.

Manuscripts and presentations

  1. 2026Preprint

    Cross-Modal Benchmarking of Acoustic Prosody and Ventral Striatal BOLD for Depression-Related Anhedonia Classification: A Pre-Registered Study with the ClinicalWhisper Pipeline

    Zhou, C., Wu, M., Xiang, Y., & Itti, L. · bioRxiv preprint, not peer reviewed · doi 10.64898/2026.06.08.728970

  2. In preparationIn preparation

    Vision-based assistive devices: A review informed by a survey of blind and visually impaired users [working title]

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

  3. Jul 2026Talk / poster

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

    Zhou, C., Wu, M., Xiang, Y., & Itti, L. · Poster, Computational Psychiatry Conference, Yale University

  4. Aug 2026Talk / poster

    Characterizing suicidality using band-specific resting-state MEG neural complexity

    Zhou, C., Kumar, S., Gilbert, J., Zarate, C., Jr., Ballard, E., & Kvarta, M. · Poster, NIH Summer Poster Day, Bethesda, MD

  5. Apr 2026Talk / poster

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

    Zhou, C. · Poster, 27th USC Undergraduate Symposium

  6. Nov 2025Talk / poster

    Attentive AI visual aid for persons with visual impairment

    Zhou, C. · Poster, Trojan Research Association Undergraduate Research Symposium