Machine Teaching: A New Paradigm for Building Machine Learning Systems (1707.06742v3)

Abstract: The current processes for building machine learning systems require practitioners with deep knowledge of machine learning. This significantly limits the number of machine learning systems that can be created and has led to a mismatch between the demand for machine learning systems and the ability for organizations to build them. We believe that in order to meet this growing demand for machine learning systems we must significantly increase the number of individuals that can teach machines. We postulate that we can achieve this goal by making the process of teaching machines easy, fast and above all, universally accessible. While machine learning focuses on creating new algorithms and improving the accuracy of "learners", the machine teaching discipline focuses on the efficacy of the "teachers". Machine teaching as a discipline is a paradigm shift that follows and extends principles of software engineering and programming languages. We put a strong emphasis on the teacher and the teacher's interaction with data, as well as crucial components such as techniques and design principles of interaction and visualization. In this paper, we present our position regarding the discipline of machine teaching and articulate fundamental machine teaching principles. We also describe how, by decoupling knowledge about machine learning algorithms from the process of teaching, we can accelerate innovation and empower millions of new uses for machine learning models.

• The paper demonstrates that machine teaching decouples ML model construction from algorithmic expertise, enabling broader participation by domain experts.
• It leverages software engineering principles like abstraction and decomposition to streamline the teaching process and cut retraining costs.
• The research suggests future development of intuitive teaching languages and IDEs, potentially democratizing the creation of bespoke ML models.

An Exploration of Machine Teaching: A Paradigm for Expanding Machine Learning Accessibility

The paper "Machine Teaching: A New Paradigm for Building Machine Learning Systems" presents a novel perspective that shifts focus from creating new ML algorithms to optimizing the teacher's role in the process of building ML systems. This shift proposes an expansion of participation in ML development to include domain experts without requiring them to have deep ML expertise. By emphasizing the role of teachers and the interaction with data, this discipline of machine teaching aspires to democratize access to ML model creation.

Central Arguments and Technical Propositions

The authors introduce machine teaching as a complementary discipline focused on the productivity and efficacy of teachers rather than on the performance of learners. The proposed discipline extends principles from software engineering and programming languages, emphasizing design principles, abstraction, and decomposition. The paper delineates machine teaching from traditional ML by highlighting that while the latter focuses on algorithmic improvements, the former concentrates on making the teaching process more accessible and efficient.

A central argument is the decoupling of knowledge of ML algorithms from the teaching process. This decoupling is posited to lower the expertise barrier, encouraging domain experts to participate directly in model building tasks. This approach echoes principles seen in programming languages where high-level languages are designed to be agnostic to the underlying machine code—“write once, run anywhere.” By doing so, it aims to universally enable the creation and maintenance of ML models.

Numerical Results and Claims

The paper supports its propositions with qualitative illustrations and historical analogies rather than quantitative datasets, which are typical in machine learning. It draws parallels to the evolution of programming and anticipates a similar trajectory for machine teaching. The narrative holds that scaling the number of potential machine teachers to match the scale of programmers (several millions) will subsequently lead to vast expansions in applicable use cases for ML models.

This approach claims that the emphasis on machine teaching can address inefficiencies inherent in current ML model-building processes, particularly the issues surrounding model reproduction, the handling of concept evolution, and the costs related to re-training and re-labeling data. The paper suggests that by empowering domain experts who thoroughly understand the problem semantics, there can be a significant reduction in the human cost and time required to develop ML models.

Implications and Speculation on Future Developments

Practical implications of this research are profound. By making machine teaching processes universally accessible, the paradigm could accelerate innovation in specialized applications by empowering a wider range of users to harness ML. Theoretically, it posits that bridging the gap between human conceptual understanding and machine execution is achievable via teaching languages and processes that are intuitive to non-experts.

Future developments could entail creating a standardized, intuitive teaching language and robust interactive development environments analogous to IDEs in software engineering. The emergence of such tools could lower the barriers to entry, making it feasible for vast numbers of domain experts to create bespoke ML models without needing deep learning expertise.

In conclusion, the paper represents a paradigm shift emphasizing teacher involvement in the construction of machine learning models. This shift holds potential to increase the scalability of ML applications substantially. The discipline of machine teaching proposes a systemic transformation aimed at making machine learning capabilities universally accessible, thus opening a multitude of opportunities for innovation across domains that require expert knowledge for precise model development.