Types of Engineering Research: A dependency graph model

Eddie Zhang
3 min readDec 17, 2023

tl;dr: Formalizing the different kinds of engineering research

Image taken from here

Model:

Principles (axioms) -> Theory -> Upstream Vertex (Building blocks, like SGD and neural nets. Telltale sign is that they test on multiple domains like supervised learning, RL, etc.) -> Application (benchmark, simulator, evaluation). Datasets sit somewhere between building blocks and application. Depends on support and distribution of data.

Types of Research:

1. NEW UPSTREAM VERTEX: Find new building block (vertex) and demonstrate superior effectiveness on downstream applications
2. NEW EDGE: New method for old application
3. NEW DOWNSTREAM VERTEX: Find new application of old method
4. IMPROVE UPSTREAM VERTEX: Improve building blocks by upstream information (application-driven building block), or by finding new insight/principle (theory-driven building block)

Research in engineering, particularly in fields such as computer science and machine learning, often involves identifying, exploring, and developing new concepts or improving existing ones. Using a dependency graph model, this approach can be visualized as a network where vertices represent concepts or methodologies (building blocks and applications), and edges represent the relationships or dependencies between them (methods, insights, principles). Below is a formal write-up for a medium article that flushes out the six types of research based on the dependency graph model.

1. Discovery of New Building Blocks and Demonstrating Effectiveness: Central to progress, researchers are often in pursuit of uncovering new building blocks — the fundamental concepts or tools across various domains. Discovering such a vertex within the graph implies not only finding a new entity but also demonstrating its superiority in downstream applications. This involves experimentally or theoretically proving that the new vertex can lead to more efficient, accurate, or practical solutions compared to its predecessors across multiple use cases.

2. Introduction of New Edges (Methods for Established Applications): Often evolution in engineering isn’t about creating something entirely new, but connecting the old in novel ways. By finding new edges in our graph, researchers develop novel methods or algorithms that enhance or redefine the relationship between established building blocks and applications. This type of research prioritizes the optimization and re-contextualization of existing connections to breathe new life into established technologies.

3. Identification of New Downstream Vertices (Applications): The downstream vertices on our graph characterize practical applications of theoretical concepts. Innovative research involves discovering entirely new applications — untapped markets or unforeseen utilities for existing building blocks. This type of discovery often leads to transformative shifts and expansion in the impact of engineering solutions.

4. Improvements Informed by Downstream Applications (Application-Driven Enhancements): In a feedback-driven mechanism, the efficacy of a building block can often be refined based on performance in applications. This approach uses downstream information to iterate upon and hone the building blocks at play. It involves adjusting methodologies to specific use-cases, thus enhancing their versatility and effectiveness in applied scenarios.

5. Enhancement Via New Insights or Principles (Theory-Driven Developments): In other instances, improvements in building blocks stem from deep theoretical insights or emerging principles. This research ventures deep into the theoretical underpinnings of existing technologies, garnering understanding that refines or completely alters the foundational layers of our dependency graph. The effort yields theory-driven building blocks, which can then be applied widely.

6. From Principles to Applications: Tracing the Research Lifecycle: The crux of engineering research can often be charted from principles (axioms), to theory, building blocks, and finally, applications. This tiered, hierarchical approach sees principles serving as the starting point — the axioms on which theories are built. Theories then lead to the development/building blocks, which are essential for the creation and improvement of various applications such as supervised learning or reinforcement learning simulations. Under this “lifecycle”, datasets could be placed between building blocks and applications as they support the empirical data needs of varying methods.

Conclusion: Using the dependency graph model to frame engineering research uncovers the multifaceted nature of innovation and discovery. From finding new building blocks to refining applications driven by deep insights, each type represents a piece of a larger whole. Recognizing and expanding these types facilitate a more structured and systematic approach to research that can foster significant advancements and create robust and effective technological solutions.

--

--