Charting the Course to Graduation
This Dojo is a living document for my thesis progress, my hLDS approach (hierarchical fast/slow dynamics), and how I translate it into practical routines for sustainable performance.
Explore the sections below for structure, status, and actionable takeaways.
Thesis Chapter Outline
Proposed structure; status indicated by diamonds.
1. Introduction & Background
Why hierarchical models for multi-timescale dynamics; limits of flat approaches.
2. Methods: hLDS Model
Generative process, identifiability, core inference procedure.
3. Methods: Inference Deep Dive
EM, variational inference, compute considerations.
4. Application: Rats & RNNs
Slow belief updating in temporal wagering (rats, simulated RNNs).
5. Application: Multi-Area Primate
Correlated slow drifts (e.g., V4, PFC) in visual decision tasks.
6. Application: Mouse ACC
Multiple slow processes: disengagement, block bias, performance markers.
7. Discussion
Implications, limitations, future directions.
Appendices
Derivations, controls, dataset descriptions.
Hierarchical Latent Dynamical Systems (hLDS)
Modeling fast (within-trial) and slow (across-trial) factors to separate noise from signal and target the levers that matter.
Core Argument & Design
Cognitive and physiological processes evolve over multiple timescales. hLDS explicitly represents this hierarchy.
Across-Trial (Slow)
Minutes–hours: motivation, sleep pressure, engagement.
Within-Trial (Fast)
Milliseconds–seconds: stimulus processing, decisions, autonomic shifts.
Observations
Wearables, logs, tasks, and other digital health signals.
Application Findings
Visual Decision Tasks
- Strong held-out performance and interpretable slow drifts.
- Saccade decoding aligned with within-trial latents.
- Slow latents tracked cognitive fluctuation (e.g., false alarms).
Future Work & Collaborations
Continuing to extend theory and applied analytics, with collaborations that reinforce translational impact.
Academic Extensions
Comparative analyses across tasks and species; attention to identifiability, controls, and robust inference.
xSync: Digital Health Platform
Integrating heterogeneous digital health data into a structured timeline for applying hierarchical reasoning outside the lab.
Timeline to Graduation
Month-by-month milestones toward a May 2026 degree conferral.
Fall 2025
Finalize Research & Early Submissions
- Present poster at Neuroscience 2025.
- Submit abstract to CoSyNe 2026 (by Nov 16).
- Freeze figures for hLDS manuscript.
Winter 2025–2026
Manuscript & Chapter Writing
- Submit hLDS manuscript.
- Draft Methods & Applications chapters.
March 2026
Internal Deadline
- Full thesis draft to committee; incorporate feedback.
April 2026
Defense & Final Submission
- Target defense: late April.
- Final submission to Graduate School.
May 2026
Degree Conferral
Official Ph.D. graduation.