Machine Teaching: An Inverse Problem to Machine Learning and an Approach Toward Optimal Education

By Xiaojin Zhu. 2015

301 citations recorded by the source corpus.

I draw the reader's attention to machine teaching, the problem of finding an optimal training set given a machine learning algorithm and a target model. In addition to generating fascinating mathematical questions for computer scientists to ponder, machine teaching holds the promise of enhancing education and personnel training. The Socratic dialogue style aims to stimulate critical thinking.

Subfields: Artificial Intelligence, Management Science and Operations Research, Meta & Continual Learning, Human-Computer Interaction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Management Science and Operations Research

In Meta & Continual Learning

In Human-Computer Interaction

Citations in the corpus

Advances and Open Problems in Federated Learning (2019, positive): “ing is provably stronger than data poisoning, we are unaware of any formal statements addressing this. One possible approach would be to use insights and techniques from work on machine teaching (see [438] for reference) to understand “optimal” data poisoning attacks, as in [292]. Any formal stateme”
Counterfactual Visual Explanations (2019, background): “Machine teaching (Zhu, 2015) works have mainly focused on how to show examples to humans so that they can learn a task better.”
MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models (2023, background): “Question bootstrapping also has an intrinsic connection to dataset distillation [73, 82] and machine teaching [40, 41, 58, 85], where the shared target is to construct a training dataset that best facilitates generalization.”
Learning to Teach (2018, background): “The goal of machine teaching (Zhu, 2015; 2013) is to construct a minimal training set for the student model to learn a target model (i.e., an oracle).”
Optimization Models for Machine Learning: A Survey (2019, background): “The computation of the optimal training set remains, in general, an open problem, and is especially challenging in the case where the learning algorithm does not have a closed-form solution with respect to the training set [219].”
Debugging Machine Learning Models (2016, result): “We note its similarity to the machine teaching problem (Zhu, 2015).”
Teaching Multiple Concepts to Forgetful Learners (2018, unknown): “[45] Xiaojin Zhu.”
The Teaching Dimension of Linear Learners (2015, background): “The optimal teaching set is then well-defined, and represents the best personalized lesson for the student [25, 24, 16].”
SOLOIST: Few-shot Task-Oriented Dialog with A Single Pre-trained Auto-regressive Model (2020, methodology): “letion using large heterogeneous dialog corpora. (ii) In the ﬁne-tuning stage, we ﬁne-tune the pre-trained model to complete a new task using a a handful of task-speciﬁc dialogs via machine teaching (Zhu, 2015; Shukla et al., 2020). We show through a comprehensive empirical study that SOLOIST is an ”
Human-in-the-loop machine learning: a state of the art (2022, background): “Instead, he chooses to give a very specific definition to the term Machine Teaching, defining it as “the problem of finding an optimal training set given a machine learning algorithm and a target model” (Zhu 2015).”
On Human Predictions with Explanations and Predictions of Machine Learning Models: A Case Study on Deception Detection (2018, background): “This direction also connects to the area of machine teaching [51, 68, 82].”
Some Submodular Data-Poisoning Attacks on Machine Learners (2015, background): “It is based on machine teaching [17], which creates an optimal training data to make the learner learn the model in the teacher’s mind.”
Learning to Teach with Dynamic Loss Functions (2018, background): “For example, machine teaching [59, 35, 36] aims at identifying the smallest training data that is capable of producing the optimal learner models.”
BeliefBank: Adding Memory to a Pre-Trained Language Model for a Systematic Notion of Belief (2021, background): “Although preliminary, this suggests that our architecture may have value in an interactive “machine teaching” setting, where the user is supervising and correcting the system, and it continually improves as a result (Zhu, 2015).”
Towards Understanding Knowledge Distillation (2019, background): “A more distantly related topic is machine teaching (Zhu, 2015).”