Individual Assignment Topic
个人作业主题
You are sitting in the review panel for making grant allocation decisions for the Innovation and Technology Fund, administrated by the Innovation and Technology Commission of Hong Kong. Below are brief introductions of two grant proposals.
您正在参加香港创新科技署管理的创新及科技基金拨款评审小组的评审工作。以下是两项拨款提案的简介。
Project 1. Hollo is a platform that assists mental health professionals to improve their processes, making them better informed and faster. Hollo collects personal-level data on mood fluctuation, sleep quality and everyday behaviour to obtain a more holistic and objective input to facilitate more accurate assessment and diagnosis. It allows stakeholders to collect data continuously instead of requiring the clients to recall and report on what they have done in the past week in the counselling room. See the following links for details about Hollo: https://www.scifac.hku.hk/prospective/ug/minor-in-science-entrepreneurship/student-setup-hollo; https://hollo.hk/
项目一： Hollo 是一个帮助心理健康专业人士改进流程的平台，使他们能够更及时、更有效地了解情况。Hollo 收集个人层面的情绪波动、睡眠质量和日常行为数据，以获得更全面、更客观的反馈，从而进行更准确的评估和诊断。它允许利益相关者持续收集数据，而无需客户回忆并报告过去一周在咨询室的活动。有关 Hollo 的详细信息，请参阅以下链接： https://www.scifac.hku.hk/prospective/ug/minor-in-science-entrepreneurship/student-setup-hollo ； https://hollo.hk/
Consider the following hypothetical scenario: Hollo is now seeking funding of 3 million to upscale their platform by collecting more data on the clients (people with mental health issues). These include mobility (location) information, random audio recording, facial expression, eye movement, and linked electronic health records.
假设以下场景：Hollo 正在寻求 300 万美元的融资，以通过收集更多客户（有心理健康问题的人）的数据来升级其平台。这些数据包括移动性（位置）信息、随机音频记录、面部表情、眼球运动以及关联的电子健康记录。
Project 2. HINCare is a software platform for suggesting and matching potential helpers from social networks. It collects basic information about the older persons and the helpers, including gender, age, address, request, whether the older person is cognitively impaired, etc.
项目二： HINCare 是一个从社交网络推荐和匹配潜在帮助者的软件平台。它收集老年人和帮助者的基本信息，包括性别、年龄、地址、需求、老年人是否有认知障碍等。
See the following links for details about HINCare: https://www.hincare.hku.hk/
有关 HINCare 的详细信息，请参阅以下链接： https://www.hincare.hku.hk/
Consider the following hypothetical scenario: HINCare is now seeking funding of 5 million to improve its platform by collecting more data on both the older persons and potential helpers. To better understand older persons' needs, HINCare proposes to provide wearable devices to a sample of 150 older persons. The devices will collect mobility, sleep quality, heart rate, mood, and random audio sample data. These personal data will also be linked to the person’s eHealth and electronic health records.
设想以下场景：HINCare 目前正在寻求 500 万美元的融资，以收集更多关于老年人及其潜在帮助者的数据，从而改进其平台。为了更好地了解老年人的需求，HINCare 计划为 150 名老年人样本提供可穿戴设备。这些设备将收集他们的活动能力、睡眠质量、心率、情绪和随机音频样本数据。这些个人数据还将与用户的电子健康记录 (eHealth) 关联。
Will you support the application, and why?
您会支持该申请吗？为什么？
Please choose one project and write an essay (between 1800 and 2000 words; bibliography excluded) to motivate your decision. You can consider the following aspects when arguing for your decision: novelty, significance, feasibility, and risk. We will assess your essay on the following aspects:
请选择一个项目，并撰写一篇论文（1800 至 2000 字， 不含参考文献 ）来论证您的决策。您可以考虑以下几个方面来论证您的决策： 新颖性、重要性、可行性和风险 。我们将根据以下方面评估您的论文：
· Originality
· 原创性
· Understanding of the topic
· 理解主题
· Literature coverage and appraisal
· 文献覆盖与评价
· Analysis, argument & critique
· 分析、论证和批评
· Structure of the presentation
· 演示文稿的结构
· Language use
· 语言运用
· Referencing
· 引用
Format requirement: 格式要求：
Please use the individual essay template to submit.
请使用个人论文模板提交。
1) spacing: 1.5; 2) font: Times New Roman; 3) font size: 12 pts.; 4) citation style: APA
1）行距：1.5；2）字体：Times New Roman；3）字号：12 pts.；4）引用格式：APA
Due Date: 30 April, 2025 (Wed), 23:59
到期日：2025年4月30日（周三）23:59
Mental health disorders are a leading cause of disability worldwide, demanding innovative solutions. Mood disorders such as depression and bipolar disorder contribute significantly to global disease burden and are associated with elevated premature mortality ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). In Hong Kong and beyond, rising rates of depression, anxiety, and youth suicide underscore an urgent need for improved mental health strategies. Traditionally, diagnosis and monitoring rely on infrequent clinical visits, interviews, and questionnaires – methods that are subjective, episodic, and prone to recall bias ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Patients often suffer “between the appointments,” as subtle warning signs of deterioration go undetected by the healthcare system. The lack of clear objective biomarkers for mental illnesses further hampers early intervention ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ).
Amid this challenge, advances in technology offer hope. Ubiquitous smartphones and wearables continuously generate digital traces of human behavior – from GPS location pings to social media usage – which can be mined for health insights ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). The emerging field of digital phenotyping seeks to harness these real-time behavioral indicators as objective measures of mental state ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ) (Frontiers | Issues and opportunities of digital phenotyping: ecological momentary assessment and behavioral sensing in protecting the young from suicide). In other words, patterns in our daily digital lives (how far we travel, sleep rhythms, communication patterns, etc.) can serve as proxies for our psychological well-being. Machine learning and big data analytics can analyze these patterns at scale, offering the potential for continuous mental health monitoring and early warning systems outside of traditional clinical settings ( The State of Digital Biomarkers in Mental Health - PMC ).
Hollo, a Hong Kong-based mental health technology initiative, has put forward a proposal to the Innovation and Technology Fund (ITF) to develop a cutting-edge digital phenotyping platform. The project plans to collect multimodal behavioral data – including users’ location, facial expressions, voice tone, eye movement patterns, and health records – and apply artificial intelligence (AI) to detect early signs of mental distress. This essay, written from the perspective of a grant review panel, argues in support of funding Hollo’s HK$3 million proposal. We will detail the project’s academic and practical merits: its novelty in integrating emotion and attention data (a frontier concept in mental health AI), its significance in addressing the mental health crisis (with implications for suicide prevention), its technical feasibility through cloud-based architecture and privacy-preserving learning, and its consideration of risks (privacy, fairness, and potential surveillance concerns). In doing so, we highlight how Hollo aligns with modern big data methodologies – leveraging digital traces, real-time inference, and predictive analytics – and why it stands to substantially advance public mental health infrastructure.
Mental health conditions impose a vast global burden. According to the World Health Organization, approximately 13% of the world’s population lives with a mental disorder ( A systematic survey on the application of federated learning in mental state detection and human activity recognition - PMC ), and hundreds of millions suffer from depression or anxiety. In Hong Kong, local surveys and healthcare reports indicate increasing prevalence of depression and high stress levels, especially in the wake of social unrest and the COVID-19 pandemic. Suicide remains a leading cause of death among young people in many regions. This societal toll – measured in lost lives, disability, and economic costs – makes mental health a top public health priority. Yet, as noted, our current systems catch problems late. Clinical assessments are typically based on what a patient recalls and reports during a visit ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Such assessments miss the day-to-day fluctuations of mood and behavior that might signal relapse or crisis. For instance, a depressed individual’s withdrawal from social activities or a subtle change in their speaking voice might precede a suicide attempt, but clinicians would not know until it’s too late if relying solely on scheduled appointments. The significance of Hollo’s project lies in its potential to fill these critical gaps using big data.
By continuously analyzing a person’s digital life footprint, Hollo aims to provide objective, real-time measures of mental state. This aligns with the vision of moving mental health care beyond the clinic walls. Research shows that digital behavior markers can reflect mental well-being: for example, depression is associated with social isolation and reduced mobility, which can be detected via smartphone communication logs and GPS data ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ) ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Fatigue and low energy (core symptoms of depression) manifest in physical inactivity, something wearable sensors or phone accelerometers can capture ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Even sleep patterns (too little or too much sleep) are telltale signs of mood disorders ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). By tapping into these rich data streams, Hollo could alert users and caregivers to concerning trends – such as sharply decreased daytime movement or unusual midnight phone use – that traditionally would go unnoticed. This proactive monitoring is especially significant for suicide prevention. Studies in digital psychiatry have begun linking behavioral “red flags” (like erratic daily routines or spikes in despair in one’s voice or text sentiment) with heightened suicide risk (Frontiers | Issues and opportunities of digital phenotyping: ecological momentary assessment and behavioral sensing in protecting the young from suicide). An app like Hollo could detect these red flags and prompt an intervention (e.g. notifying a clinician or sending the user a supportive message/hotline resource) well before a crisis peaks. The ability to reach individuals in real time, wherever they are, represents a sea change in mental health support – from reactive to preventative care. Funding Hollo would thus support a project of tremendous public health significance, leveraging big data to tackle the pressing burden of mental illness in a way not possible with traditional tools.
Hollo’s approach is highly novel, pushing the frontier of digital phenotyping through its use of multimodal data and sophisticated AI analytics. While prior studies have used smartphones to passively track behaviors (like steps or call frequency), most have been limited to a single modality at a time ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Hollo, in contrast, proposes to integrate multiple data sources simultaneously to create a more holistic picture of mental health. This includes not just physical and phone usage data, but also facial and vocal indicators of emotion, and even eye movement patterns. Such a comprehensive approach to behavioral sensing is cutting-edge and relatively unseen in deployed mental health apps.
Importantly, research suggests that combining modalities can yield more powerful and personalized insights. One recent review found that studies focusing on one data type often see only modest differences between people with depression and healthy individuals, whereas a multimodal approach could capture subtle, otherwise hidden patterns ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ) ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ). Hollo’s system will, for example, correlate location data with sleep data and social interaction data to detect complex behavioral signatures of worsening mood. This “multimodal digital phenotyping” is a novel contribution, moving beyond simple step counters or mood diaries toward an AI that understands the user on multiple levels. The inclusion of health record data further adds context (e.g. knowing a user’s history of depression or past medication can help the AI personalize its analysis), something innovative for a consumer-facing mental health platform.
A standout novel feature is Hollo’s exploration of emotion–attention coupling. This concept refers to how a person’s emotional state might be linked with where they direct their attention. Hollo plans to use the smartphone’s camera to monitor facial expressions (as a proxy for emotion) and eye gaze patterns (as a proxy for attention). By coupling these, the system might detect, say, that a user who is frequently looking away and showing a flat affect during video journaling exercises is experiencing disengagement indicative of depression. Early academic work supports the promise of such analytics. For instance, a study on children found that depressive symptoms correlated with how long they looked at sad images, suggesting attention bias towards negative stimuli (Frontiers | Exploring a multimodal approach for utilizing digital biomarkers for childhood mental health screening). In adults, similarly, eye-tracking studies show anxious individuals may fixate on threat cues, while depressed individuals often have difficulty sustaining attention on external tasks. Hollo’s innovation is to bring these experimental findings into a real-world, real-time tool. If successful, it could, for the first time, validate emotion-attention coupling as a digital biomarker for mental health in a large user base. This represents a novel scientific contribution as well as a unique value proposition for the product. No existing mental health app on the market offers this kind of nuanced AI insight into the interplay between how we feel and where we focus our attention.
Beyond emotion-attention analysis, Hollo’s use of advanced AI/ML algorithms to fuse multimodal data is itself innovative. The platform will employ machine learning models (potentially deep neural networks) that can take in streams of disparate data – combining geolocation sequences with voice tone features, for example – to detect patterns humans might miss. By developing algorithms that can weigh hundreds of behavioral features simultaneously, Hollo could discover new digital biomarkers of mental illness. Such biomarkers might include complex patterns like “reduced location entropy coupled with monotonic voice pitch” as a signature of impending depressive relapse – insights not found in current clinical practice. This kind of data-driven knowledge discovery is a hallmark of big-data innovation, and funding Hollo would place Hong Kong at the forefront of exploring these novel biomarkers. In summary, the novelty of Hollo’s proposal lies in its comprehensive sensor approach and its pioneering AI analytics (especially emotion-attention coupling and multimodal pattern recognition), which go well beyond the state-of-the-art in digital mental health.
While ambitious, Hollo’s plan is grounded in feasible technology and methodologies. The project smartly combines a cloud-based infrastructure with edge computing (on smartphones) to handle the data flows. Here is how the system is envisioned and why it’s practical:
Any initiative collecting personal and sensitive data must contend with serious ethical responsibilities and potential risks. The grant panel has carefully evaluated these issues in Hollo’s proposal – notably privacy, fairness, and concerns about surveillance – and our assessment is that Hollo not only acknowledges these risks but has a robust plan to mitigate them, aligning with best practices in digital health ethics.
1. Privacy and Data Protection: Hollo will be privy to intimate details of a person’s life – where they go, how they sound, how they look, and elements of their health history. If mishandled, such data could cause harm (e.g., stigma or discrimination if leaked). Hollo’s design, however, places privacy at its core. As described, the use of federated learning means the bulk of raw data stays on-device, never pooling in a centralized repository ( A systematic survey on the application of federated learning in mental state detection and human activity recognition - PMC ). This drastically reduces the risk of a mass data breach because there is no single trove of all user data to steal. Additionally, all data communications will be encrypted in transit and storage; Hollo will employ industry-standard security (e.g. HTTPS, encryption of sensitive fields in databases, secure enclaves for any cloud computation involving personal data). The proposal also mentions exploring differential privacy, which would add noise to any aggregated data Hollo shares or publishes, ensuring individuals cannot be re-identified. Users will be given a detailed privacy policy and must provide informed consent for each data type (GPS, microphone, etc.) that the app uses – with the option to opt out of any particular sensor if they choose. These steps follow the consensus recommendations from ethics research: be transparent, require consent, protect data, and only use it with permission (Frontiers | Issues and opportunities of digital phenotyping: ecological momentary assessment and behavioral sensing in protecting the young from suicide). We are satisfied that Hollo’s approach to privacy is rigorous and compliant with regulations (e.g., GDPR-equivalent standards and Hong Kong’s Personal Data Privacy Ordinance). The platform will also undergo security audits and have a clear data deletion policy (users can withdraw and have their data wiped). Overall, while no system is without risk, Hollo’s architecture and policies greatly minimize the privacy risks, making it an acceptable endeavor from a data protection standpoint.
2. Fairness and Algorithmic Bias: AI models can inadvertently perpetuate biases – a known issue in health algorithms. In mental health, there is a risk that an AI trained mostly on data from one group (say, young tech-savvy users) might under-detect issues in another group (elderly users or those from different cultural backgrounds with different behavior patterns). The Hollo team is aware of this and has incorporated fairness checks in the project. First, by collecting large and diverse datasets, the model can learn the normal behavior baselines for different ages, genders, etc. The project will collaborate with local clinics and perhaps NGOs to recruit a diverse user base (ensuring representation of various communities). Second, the machine learning models will be evaluated for performance across subgroups. If the accuracy for one subgroup is significantly lower, developers will investigate and retrain the model to address the gap (for example, by adding subgroup-specific model components or additional training data). Such practices ensure the AI does not predominantly serve only a subset of users. Fairness is identified as a major ethical concern in digital phenotyping literature, and experts advise to “establish effective methods to minimize bias or discrimination” (Frontiers | Issues and opportunities of digital phenotyping: ecological momentary assessment and behavioral sensing in protecting the young from suicide). Hollo’s plan to iteratively improve model fairness aligns with this. Moreover, the federated learning framework itself can be tuned for fairness (recent research explores techniques like weighted updates to ensure minority data contributions are not ignored). The proposal also states that an ethics board or advisory panel will periodically review the system’s outcomes for any signs of bias. With these measures, we believe Hollo will deliver a tool that is fair and equitable, maximizing benefit and minimizing harm across society.
3. User Autonomy and Surveillance Concerns: A potential psychological risk is that users might feel uneasy or “watched” knowing an app is continuously monitoring them. This could cause anxiety or alter their behavior (the Hawthorne effect – people changing behavior because they know they’re observed). Hollo intends to mitigate this in several ways. Transparency is key: users will know exactly what is being collected and for what purpose. The app interface will have a dashboard showing, for example, “This app is using your location to detect changes in your routine that relate to your mental well-being.” Education within the app will make clear that the goal is therapeutic, not surveillance. Additionally, Hollo will allow users to pause data collection at will (e.g., a “privacy pause” button if they want to temporarily not be tracked). Giving users this control can alleviate feelings of helplessness or invasion. Another strategy is to primarily use passive, in-the-background data collection that does not require user action; studies have found that such passive monitoring can actually be less intrusive and more accurate than self-reports, since it doesn’t rely on the user constantly remembering they are being assessed ( Development of Digital Biomarkers of Mental Illness via Mobile Apps for Personalized Treatment and Diagnosis - PMC ) ( Development of Digital Biomarkers of Mental Illness via Mobile Apps for Personalized Treatment and Diagnosis - PMC ). Over time, as users see the app delivering value – for instance, alerting them “you seem to be withdrawing, consider reaching out to friends or using coping strategies” – they may come to view it as a helpful guardian rather than a spy. Nonetheless, Hollo will be cautious not to cross the line into undue surveillance. All analyses will be done for the user’s benefit and, by default, data will not be shared with third parties (e.g., not with employers or insurers, unless a user explicitly opts to share a summary with a doctor). By adhering to the ethical principle of accountability (Frontiers | Issues and opportunities of digital phenotyping: ecological momentary assessment and behavioral sensing in protecting the young from suicide), Hollo’s team accepts responsibility for protecting users from harm. They will continuously seek user feedback to ensure the app’s presence remains a welcome one in users’ lives. In summary, while the notion of continuous monitoring can raise surveillance concerns, Hollo’s commitment to user-centric design, transparency, and control will keep users informed and empowered, thereby maintaining trust.
In evaluating Hollo’s risk management, the grant panel finds that the project not only meets but exceeds typical ethical standards for big data health research. The proposal demonstrates a clear understanding that technology in mental health must be humane, privacy-preserving, and just. By integrating state-of-the-art solutions like federated learning for privacy and explicitly planning for fairness and accountability, Hollo is poised to become a model for how to do data-driven innovation responsibly. These robust safeguards give us confidence that funding Hollo will yield a tool that users and stakeholders can embrace rather than fear.
After thorough review, we strongly recommend funding Hollo’s HK$3M grant proposal. This endorsement rests on multiple compelling factors that we have explored:
In conclusion, funding Hollo represents a strategic investment in innovation that aligns with both public health needs and technological advancement. The proposal exemplifies how big data can be harnessed for social good – turning everyday digital traces into life-saving knowledge ( Multimodal Digital Phenotyping Study in Patients With Major Depressive Episodes and Healthy Controls (Mobile Monitoring of Mood): Observational Longitudinal Study - PMC ) ( Digital Phenotyping of Emotion Dysregulation Across Lifespan Transitions to Better Understand Psychopathology Risk - PMC ). By approving this grant, the Innovation and Technology Commission would enable the development of a platform with the capacity to detect silent cries for help hidden in data and provide timely support to those in psychological distress. The ripple effects of such a project are far-reaching: improved patient outcomes, reduced burden on healthcare systems, new research insights, and the growth of a homegrown health-technology enterprise. Hollo’s vision of AI-driven, personalized mental health care is precisely the kind of high-impact, data-informed innovation that the ITF was established to nurture. We, therefore, enthusiastically recommend full funding for Hollo. This project holds promise not only to advance technology but, more importantly, to advance the well-being of our community – a goal worth every dollar of the investment.