HOPES: Digital Phenotyping for Mental Health
Since 2019, I have been involved with the HOPES Project at the MOH Office for Healthcare Transformation (under the Ministry of Health) and the Institute of Mental Health (IMH). HOPES is a remote monitoring clinical service at IMH, where we continuously and passively measure behavioral data from the smartphone and a wrist-wearable for patients with schizophrenia and major depression. The data drives machine learning algorithms that prioritze patients and raise alerts to clinical staff with the goal of avoiding or reducing relapses, including hospitalizations.
We developed the machine learning algorithms on data collected through an observational study that we ran at IMH from 2019 - 2023:
-
Evaluating the utility of digital phenotyping to predict health outcomes in schizophrenia: protocol for the HOPE-S observational study.
NAA Rashid, W Martanto, Z Yang, X Wang, C Heaukulani, N Vouk, T Buddhika, Y Wei, W Verma, C Tang, RJT Morris, and J Lee. BMJ open, 11(10), e046552. Oct 2021. [paper] -
HOPES -- An integrative digital phenotyping platform for data collection, monitoring and machine learning.
X Wang, N Vouk, C Heaukulani, T Bhuddika, W Martanto, J Lee, and RJT Morris. Journal of Medical Internet Research, 23(3), e23984. Mar 2021. [paper]
mindline.sg and Let's Talk: Digital Mental Wellness
Since 2020, I have been involved with two of MOHT's digital mental wellness platforms, mindline.sg (https://mindline.sg) and Let's Talk (https://letstalk.mindline.sg). mindline.sg is a one-stop mental health and wellness resource website targeting users in Singapore, currently focusing on self-care and maintenance in the community setting (we are moving into the clinical adjunct space next). Features of the site include over 500 curated and localized resources, an AI mental health chatbot from Wysa, self-assessment tools, and direction to local service touchpoints. In addition to visiting the site, you can find a descriptive article about the platform along with some of our early learnings from implementation here:
-
Mental Wellness Self-care in Singapore with mindline.sg: A Framework for the Development of a Digital Mental Health Platform for Behaviour Change.
JH Weng JH, Y Hu, C Heaukulani, C Tan, JK Chang, YS Phang, P Rajendram, WM Tan, WC Loke, and RJT Morris. In Review. JMIR Preprints 17/01/2023:45761. Jan 2023. [paper]
Let's Talk is a digital mental health forum for youth in Singapore, and unlike mindline.sg (which is purely about self-care) Let's Talk is purely about human-based support. Peer-support through anonymous accounts on carefully monitored forums forms the bulk of the strategy, and we additionally provide the "Ask-A-Therapist" service where users may post a question that will be answered by our on-staff professional therapists within 24 hours. A blog post describing our motivation and design principles for the platform can be found here:
-
Building a youth-centric mental health solution.
Janice Weng & Caleb Tan, MOHT blog post, 5 May 2023.
Lecturing at NUS: Machine Learning
I teach a graduate course on statistical machine learning at the National University of Singapore (NUS), ST5227 Applied Statistical Learning. The course is hosted by the Department of Statistics and Data Science, where I am adjunct faculty, and is usually attended by graduate students. The course largely follows the level and pedagogy of The Elements of Statistical Learning (Hastie, Tibshirani, and Friedman), with equal parts laboratory work in Python based on a series of data science case studies. My goals for the course are to produce versatile students with a balance of strengths in statistical inference, machine learning algorithms, and practical data science.
Molecular Biology and Epigenetics
I'm currently undertaking a part-time masters degree in the Department of Biological Sciences at NUS, where I am training in molecular biology in Greg Tucker--Kellogg's lab. I don't have any public material to show for this work yet, but in short I'm developing spatial epigenetic methods in single cells to detect certain epigenetic biomarkers for cancer. My current project utilizes techniques in immunoflurescence and DNA nanostructures.
Finance
I spent 3 years as a machine learning quant on an equities trading desk at Goldman Sachs in Hong Kong. The following two papers summarize the style of models I was thinking about at the time:
-
Modeling financial volume curves with hierarchical Poisson processes.
C. Heaukulani, Abhinav Pandey, and Lancelot F. James. Arxiv, June 2024. [paper] -
Scalable Bayesian dynamic covariance modeling with variational Wishart and inverse Wishart processes.
C. Heaukulani and Mark van der Wilk. NeurIPS, 2019. [paper]
Bayesian nonparametrics
My PhD work focused on Bayesian nonparametric modeling and inference in machine learning. Zoubin Ghahramani and Carl Rasmussen were the PIs in my lab at the time, and if you are familiar with their work then that summarizes much of my training. Equally as much of my training occurred under people such as Daniel M. Roy, David Knowles, and Lancelot F. James. The following two papers exemplify my work at the time:
-
Gibbs-type Indian buffet processes.
C. Heaukulani and Daniel M. Roy. Bayesian Analysis, Vol. 15, No. 3, 683--710, 2020. [paper] -
The combinatorial structure of beta negative binomial processes.
C. Heaukulani and Daniel M. Roy. Bernoulli, Vol. 22, No. 4, 2301--2324, 2016. [paper]
Graphs and Network Models
Much of my work looks at applying probabilistic modeling and inference techniques to models for random graphs or networks. The following papers are highlights:
-
Dynamic probabilistic models for latent feature propagation in social networks.
C. Heaukulani and Zoubin Ghahramani. ICML, 2013. [paper] -
Bayesian inference on random simple graphs with power law degree distributions.
Juho Lee, C. Heaukulani, Zoubin Ghahramani, Lancelot F. James, and Seungjin Choi. ICML, 2017. [paper] -
Variational inference for neural network matrix factorization and its application to stochastic blockmodeling.
Onno Kampman and C. Heaukulani. ICML 2019 Workshop on Learning and Reasoning with Graph-Structured Data. [paper]