💡 Wu Shin-Ting's Research Projects

I believe that to solve complex problems of our era, the best approach is a collaboration between human intelligence and the processing power of machines. To address this approach, I focus on building systems that combine human intelligence with the computational power of machines. My research centers on creating synergy through intuitive visual interfaces that connect robust, scalable data processing with human reasoning, involving developing data structures, geometric modeling techniques, interactive visualization tools, and domain-specific algorithms for manipulating, simulating, and analyzing complex 3D data, particularly in fields such as energy, healthcare, and data scineces.

In digital systems education, where the focus is often on accessibility and high-level abstractions, I work to bridge the gap between abstract concepts and the foundational principles of computing. My approach connects the logical behavior of systems, specifically microcontrollers and GPUs, to the underlying physical behavior of electronic circuits. I believe that exposing these inner workings not only strengthens student understanding but also significantly boosts their engagement. To this end, I develop educational materials and methods that bring learners closer to the physical architecture of computing.

I address challenges in geometric modeling and data management, recognizing that scalable and robust digital systems, including "distant", highly interactive 3D graphics interfaces, critically depend on how data is structured, accessed, and transformed. The projects below illustrate how I approach these issues in practice:

• TDM (in Portuguese): A library designed to manage and ensure the topological consistency of geometric models. It extends the Boundary Representation (B-Rep) framework to preserve the integrity of surfaces and solids, supporting reliable and consistent geometric operations.
• Differential Geometry: Investigating its potential to enable precise, local interactions with visible geometry. This research led to three distinct projects:
  • InterSurf: Leverages differential geometry principles to compute the complete intersection loci of surfaces, robustly handling all associated singularities.
  • DesMo: Simulates cloth deformation using the Cosserat shell model, which is a continuum mechanics framework that incorporates microstructural effects and differential-geometric quantities.
  • RFM: Reconstructs a mesh from a point set by iteratively deforming a sphere to fit the data, guided by principles of the Balloon model.

Despite the emergence of advanced 3D interaction paradigms such as augmented and virtual reality, and the development of specialized input devices, 2D WIMP (Windows, Icons, Menus, Pointer) interfaces remain the dominant model for human-computer interaction. This persistence is not due to technical superiority, but rather to fundamental advantages in economic accessibility and deep-rooted user familiarity. WIMP’s reliance on widely available, low-cost hardware, and its decades-long presence has established a strong foundation of learned user behavior, securing its place as the most viable and globally adopted interaction model.

In the 1990s, the prevailing approach to interacting with 3D geometric data was to integrate complete geometric models into traditional WIMP interfaces. In contrast, I proposed a paradigm shift: leveraging differential geometry to enable precise, intuitive interactions based solely on visible geometry, without requiring full access to the underlying model data. This significantly reduces computational demands and enables more flexible interaction modalities. Crucially, the physical realization of these ideas has been made possible by the evolution of programmable GPU hardware. The ability to perform custom computations directly on the GPU enables real-time generation and manipulation of geometry-aware data, bridging the gap between theory and interactive experience. The following project serves as a concrete proof of concept for this integration:

• MTK: Reconstructs 3D point coordinates from enhanced GPU-generated depth maps, forming a practical foundation for interaction techniques based on any visible geometry.

Drawing on a foundation in data management and user-centered design, I have developed systems that transform complex datasets into actionable insights through responsive and intuitive visual interfaces. The following two projects highlight this approach:

• VDX: Designed for energy pre-dispatch in a geo-referenced power network. To support efficient, interactive exploration of large-scale grid data at multiple resolution levels, we developed and implemented a scalable map simplification algorithm.
• VMTK-Neuro: Focused on the direct exploration of complex medical head imaging data, with targeted applications in teaching, neuronavigation, and neurosurgical planning. Our work represents a paradigm shift by leveraging the visual characteristics that clinicians already trust. Instead of relying on intermediate models like meshes or idealized reconstructions, our innovative approach gives clinicians the confidence to make decisions based on the original magnetic resonance imaging (MRI) and computed tomography (CT) images they interpret daily. To deliver responsive performance across use cases, we adapted and optimized visualization algorithms for real-time rendering and low-latency interaction. Our work also includes methods for processing raw diffusion-weighted imaging (DWI) data into diffusion tensor imaging (DTI) or high angular resolution diffusion imaging (HARDI) representations, along with custom algorithms for visualizing discrete diffusion data and tractography results. Together, they provide a comprehensive and intuitive view for clinical use.

Motivated by the growing demand for high-quality training data in AI, I have recently expanded this line of work to include visual systems for data preprocessing, aiming at improving the clarity, structure, and usability of complex datasets in machine learning workflows.

---

🧩 Teaching 🧑

My challenge as an educator is helping students demystify complex technologies by opening their "black boxes” and exploring their internal workings. In undergraduate courses, I introduce techniques that bridge the gap between electronic circuits and programming, using tools such as JTAG (IEEE 1149.1), Serial Wire Debug (SWD), and measurement instruments. These tools enable precise measurements and protocol-level signal analysis, allowing students to establish direct connections between physical circuitry and programming logic. At the graduate level, I extend this methodology to GPUs, linking low-level APIs to the performance demands of interactive graphics systems. To foster engagement, I incorporate formative challenges that encourage students to propose their own solutions before introducing established strategies. This active-learning approach is concretely reflected in the handouts we developed for the following courses:

• Undergraduate
• EA773 – Second semester of 2009 (in Portuguese): We introduced the practical development of logic circuits using Max+Plus II software with FPGA boards. This setup allowed students to connect schematic capture with breadboard-based circuit construction, helping them relate physical implementation to abstract circuit representations.
• EA773 – First semester of 2010 (in Portuguese): FPGA technology enabled students to synthesize physical logic circuits directly from schematics, bridging theory and practice. This provided a natural entry point for introducing hardware description languages (HDLs), allowing students to model, simulate, and implement complex digital systems with formal, structured tools. This abstraction enhances, rather than replaces, their understanding of hardware, promoting scalable, modular, and rigorous design thinking.
• EA871 – Second semester of 2020 (in Portuguese): In response to the pandemic, I designed a remote laboratory where students could access and control physical lab boards from a distance. The goal was to preserve the essential link between theoretical concepts and real-world signals. A particular challenge involved modifying the lab's translucent plastic enclosure, which distorted LED colors on camera. Ensuring accurate color representation became a critical part of the engineering process.
• EA701 – Second semester of 2024 (in Portuguese): In the first edition of EA701, my colleague Prof. Quevedo secured a donation of Nucleo-144 boards with integrated STM32H7A3 microcontrollers. This provided a valuable opportunity for low-level, hands-on exploration. Using STM32CubeIDE and STM32MAX, students engaged directly with the system’s control and data registers, observing physical system behavior in response to bit-level manipulations within the clock tree and peripheral configurations.
• EA701 – First semester of 2025 (in Portuguese): In this second edition, we expanded our focus on the direct relationship between digital logic and the system's physical responses, deepening the students’ understanding of the microcontroller’s behavior at the hardware-software interface.
• Graduate
• IA725 (in Portuguese): In our recorded lectures from 1997, we already introduced Mesa 3D as a practical complement to theoretical instruction. Mesa 3D was selected for its clear mapping between API functions and core computer graphics concepts, offering students the flexibility to compose these functions into complex rendering tasks. It was likely the first graduate course in Brazil to introduce Mesa 3D, an early precursor to the OpenGL graphics API.
• IA376M – First semester of 2025 (in Portuguese): As data preprocessing for AI training becomes more demanding, and visual tools play an increasingly central role in building intuitive interfaces, we transitioned from traditional computer graphics instruction to an innovative course that integrates computer graphics concepts with statistical and inferential tools for data preprocessing in data science. By merging these traditionally separate fields, the course fosters intuitive, human-centered understanding of data and prepares students to tackle emerging interdisciplinary challenges in AI and data-driven applications.

These materials reflect a pedagogical commitment to anchoring abstract theoretical concepts in hands-on engineering practice, empowering students to trace system behavior from observable outcomes back to the digital logic that drives them.