Journey

Personal Reflections

The promotion itself was validating, but what means more to me is what it represents: that the work I care about—prototyping, advising, helping researchers achieve their goals—is genuinely valued here. In academia-adjacent roles, there’s often pressure to publish or perish, but my path has been different. I get to focus on practical impact and delivery, which aligns much better with how I want to contribute.

What I’ve learned most in this role is that technical expertise only gets you so far. The real work is listening carefully, translating between different domains, and building trust with collaborators who may not share your technical background. I’m proud that people trust my recommendations, but that trust was earned through many conversations, failed experiments, and the willingness to say “I don’t know, but let me find out.”

The autonomy is both liberating and sometimes isolating. There’s no one looking over my shoulder, but that also means I need to be self-directed and proactive about seeking feedback. I’m learning to balance independence with connection, to know when to forge ahead and when to pause and collaborate.

Looking ahead, I’m excited about the expanded scope and influence. But I’m also mindful that leadership isn’t just about having more responsibility—it’s about creating space for others to do their best work too.

Personal Reflections

Coming from the intensity of startup life, this transition felt like learning to breathe differently. The slower pace initially made me anxious, was I doing enough? But I’ve come to appreciate the space it creates for deeper thinking and more intentional work.

What surprised me most was how energizing it is to work on multiple projects with quick turnaround. Instead of being stuck in one long slog, I get to prototype ideas, see them through to completion, and move on to the next challenge. This variety keeps me engaged in a way I didn’t expect.

The opportunity to help shape a relatively new center feels meaningful. There’s real ownership here—not just in execution, but in defining how we work and what we prioritize. I’m learning that influence doesn’t always require authority; sometimes it’s about being thoughtful, persistent, and building trust over time.

From Research Questions to Product Requirements

Joining Min Learning System was a leap into unfamiliar territory, taking years of academic research on how children learn to read and translating it into a practical product. The work involved developing custom cognitive reading models to simulate children’s reading development and building deep learning systems to assess vocabulary knowledge.

The transition from academia to a startup required a fundamental shift in mindset. In research, the goal was understanding and publishing work that advanced scientific knowledge. In a startup, the goal was utility, building something that helped children learn to read better. Success was defined not by statistical significance but by whether the product worked for real users.

This required different tradeoffs. Models needed to be accurate, fast, and scalable. Evaluation focused less on controlled experiments and more on performance in real-world conditions. The standard for “good enough” shifted from theoretical rigor to practical impact.

Building with Constraints

Working in a startup environment meant operating under tighter resource constraints than in academia. Decisions involved balancing accuracy, speed, cost, and development time. Prioritization became crucial. The focus was on identifying the core value proposition and allocating effort where it mattered most.

The experience reinforced the importance of rigorous model evaluation and experimental design in product settings. Assessing accuracy and reliability was not just academic. It directly affected whether the system genuinely supported children’s learning.

This chapter demonstrated how research skills can extend beyond academia when combined with product thinking and a focus on solving real problems.

Moving to Haskins Labs at Yale represented a deliberate shift in my research approach—from primarily using neuroimaging to observe the brain during reading, to building computational models that could simulate the reading process itself. Working with Jay Rueckl and collaborating across multiple research groups, I immersed myself in the world of computational cognitive neuroscience.

From Observation to Simulation

The post-doc fundamentally changed how I thought about research. Rather than just measuring what the brain does during reading, I was now building models that attempted to capture the underlying mechanisms. Developing custom RNN architectures meant making explicit commitments about how reading development might work—what representations matter, how learning unfolds over time, what factors create individual differences.

This shift made theoretical assumptions concrete. A vague hypothesis about “phonological processing” had to become specific architectural choices and learning algorithms. If the model couldn’t learn to read in a way that matched human development, it meant our theory was incomplete.

Interdisciplinary Collaboration

The environment at Haskins Labs emphasized collaboration across methodologies. I worked with teams using fMRI, EEG, behavioral experiments, and Bayesian statistics—each bringing different perspectives on understanding reading. Learning to contribute to projects involving Bayesian latent-mixture models and neuroimaging data analysis expanded my methodological range considerably.

Grant writing and manuscript preparation became central to the work. Learning to communicate computational modeling research to audiences more familiar with traditional neuroscience methods sharpened my ability to bridge different research communities and explain technical concepts accessibly.

This experience set the foundation for thinking about reading research as fundamentally computational, preparing me for the applied work that would come next.

Completing my Ph.D. represented the culmination of years of intensive research on reading development in Chinese children. Under the supervision of Catherine McBride and Urs Maurer, I explored questions at the intersection of genetics, brain development, and literacy acquisition—examining not just how children learn to read, but why some children find it easier than others.

Bridging Multiple Disciplines

The dissertation work required integrating methodologies from genetics, neuroscience, and educational psychology. Studying heritability meant understanding behavioral genetics; examining neurobehavioral correlates meant working with EEG and understanding neural oscillations; and focusing on Chinese word reading meant deeply engaging with orthographic and linguistic theory specific to Chinese.

This interdisciplinary approach taught me that the most interesting research questions often exist at the boundaries between fields. No single methodology could fully answer the questions I was asking—it took combining multiple perspectives to build a coherent picture of reading development.

From Individual Studies to Research Programs

One of the most valuable lessons from the doctoral process was learning to think in terms of research programs rather than isolated studies. Each study needed to build on previous work and set up future questions. The dissertation wasn’t just an end point. It established a foundation and raised new questions that would continue to drive my research interests.

Working with children and families throughout the research also reinforced the real-world implications of this work. Understanding individual differences in reading wasn’t just academically interesting, it had practical implications for how we support children’s literacy development and identify those who need additional help.

My three-month exchange at USC was a pivotal learning experience that introduced me to deep learning and its applications in cognitive research. Working on a CNN to read Chinese characters, I began to see how computational models could complement traditional neuroscience approaches to understanding reading.

From Brain Signals to Neural Networks

Coming from a background in EEG and neuroimaging, shifting to building artificial neural networks felt like learning a new language. The parallels were fascinating—both involved understanding how networks of units process information to recognize patterns. But where EEG showed me what the brain does, CNNs let me explore how those processes might work mechanistically.

Designing experiments to evaluate model generalization and learning efficiency taught me to think about reading from a computational perspective. How many examples does a system need to learn a character? Which visual features matter most? These questions became tools for generating hypotheses about human learning.

A New Methodological Toolkit

This experience expanded my research toolkit significantly. I realized that computational modeling wasn’t just a separate field—it was a powerful method for testing theories about cognition. Building models that could perform reading tasks helped clarify what aspects of reading development we understood well and where our theories still had gaps.

The exchange planted the seeds for my eventual transition into computational cognitive neuroscience, showing me that the future of understanding reading development would likely involve integrating behavioral, neural, and computational approaches.

Completing my M.Phil. marked my formal entry into cognitive neuroscience research on reading development. Working under the supervision of Kevin Kien Hoa Chung, Duo Liu, and Savio Wai Ho Wong, I explored how dyslexic adolescents process morphological structures in Chinese, using neurological methods to understand the underlying mechanisms.

This research introduced me to the intersection of linguistics, neuroscience, and education. A space that would define my career trajectory. Learning to work with EEG technology and analyze neural correlates of reading opened up new ways of thinking about how the brain processes language, particularly in the unique context of Chinese orthography.

The experience taught me that understanding reading difficulties requires looking beyond behavioral measures. The neurological perspective revealed patterns and processing differences that weren’t always visible in standard assessments, emphasizing the value of using multiple methodological approaches to study cognition.

This milestone also reinforced my interest in applied research. Working with dyslexic adolescents made it clear that rigorous scientific investigation could have direct implications for educational interventions and supporting struggling readers.