Understanding the Power and Limitations of Teaching with Imperfect Knowledge

@inproceedings{Devidze2020UnderstandingTP,
  title={Understanding the Power and Limitations of Teaching with Imperfect Knowledge},
  author={Rati Devidze and Farnam Mansouri and Luis Haug and Yuxin Chen and Adish Kumar Singla},
  booktitle={International Joint Conference on Artificial Intelligence},
  year={2020}
}
Machine teaching studies the interaction between a teacher and a student/learner where the teacher selects training examples for the learner to learn a specific task. The typical assumption is that the teacher has perfect knowledge of the task---this knowledge comprises knowing the desired learning target, having the exact task representation used by the learner, and knowing the parameters capturing the learning dynamics of the learner. Inspired by real-world applications of machine teaching in… 

Figures from this paper

Explanatory machine learning for sequential human teaching

Empirical results show that sequential teaching of concepts with increasing complexity has a beneficial effect on human comprehension and leads to human re-discovery of divide-and-conquer problem-solving strategies, and studying machine-learned explanations allows adaptations of human problem-Solving strategy with better performance.

Teaching Humans When To Defer to a Classifier via Examplars

This work presents a novel parameterization of the human's mental model of the AI that applies a nearest neighbor rule in local regions surrounding the teaching examples to derive a near-optimal strategy for selecting a representative teaching set.

Preference-Based Batch and Sequential Teaching

This work develops a novel framework that captures the teaching process via preference functions $\Sigma$ and identifies preference functions inducing a novel family of sequential models with teaching complexity linear in the VC dimension: this is in contrast to the best-known complexity result for the batch models, which is quadratic in theVC dimension.

Teaching via Best-Case Counterexamples in the Learning-with-Equivalence-Queries Paradigm

This paper considers the optimal teacher who picks bestcase counterexamples to teach the target hypothesis within a hypothesis class, and introduces LwEQ-TD, a notion of TD capturing the teaching complexity (i.e., the number of queries made) in this paradigm.

Time-Constrained Learning

This work proposes TCT, an algorithm for the TCL task designed based that on principles from Machine Teaching, and presents an experimental study involving 5 different Learners and 20 datasets where it is shown that TCT consistently outperforms two other algorithms.

The Teaching Dimension of Kernel Perceptron

The sample complexity of teaching, aka teaching dimension, for kernelized perceptrons for different families of feature maps is established and a rigorous bound on the complexity for approximately teaching a Gaussian kernel perceptron is established.

Policy Teaching in Reinforcement Learning via Environment Poisoning Attacks

The results show that the attacker can easily succeed in teaching any target policy to the victim under mild conditions and highlight a significant security threat to reinforcement learning agents in practice.

Reward Poisoning in Reinforcement Learning: Attacks Against Unknown Learners in Unknown Environments

A novel black- box attack that can provably achieve a near-matching performance to the state-of-the-art white-box attack is designed, demon-strating the feasibility of reward poisoning even in the most challenging black-box setting.

Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning

The results show that the attacker can easily succeed in teaching any target policy to the victim under mild conditions and highlight a significant security threat to reinforcement learning agents in practice.

Fair Clustering Using Antidote Data

This work proposes an alternate approach to group-level fairness in center-based clustering inspired by research on data poisoning attacks, and forms this as a general bi-level optimization problem which can accommodate any center- based clustering algorithms and fairness notions.

References

SHOWING 1-10 OF 36 REFERENCES

Interactive Optimal Teaching with Unknown Learners

This paper introduces a new approach for machine teaching that partly addresses the (unavoidable) mismatch between what the teacher assumes about the learning process of the student and the actual process, and introduces interactivity as a means to mitigate the impact of imperfect knowledge.

Models of Cooperative Teaching and Learning

Two models of learning from a cooperative teacher that selects "helpful" training examples are introduced, and a new notion of "coding trick"/"collusion" is introduced that can be arbitrarily lower than the classic teaching dimension and known variants thereof, without using coding tricks.

Understanding the Role of Adaptivity in Machine Teaching: The Case of Version Space Learners

Inspired by human teaching, a new model where the learner picks hypotheses according to some local preference defined by the current hypothesis is proposed, and it is shown that the model exhibits several desirable properties, e.g., adaptivity plays a key role, and theLearner’s transitions over hypotheses are smooth/interpretable.

Teaching Inverse Reinforcement Learners via Features and Demonstrations

A teaching scheme is suggested in which the expert can decrease the teaching risk by updating the learner's worldview, and thus ultimately enable her to find a near-optimal policy.

Teaching Multiple Concepts to a Forgetful Learner

This paper looks at the problem from the perspective of discrete optimization and introduces a novel algorithmic framework for teaching multiple concepts with strong performance guarantees, both generic and interactive, and allows the teacher to adapt the schedule to the underlying forgetting mechanisms of the learner.

Learner-aware Teaching: Inverse Reinforcement Learning with Preferences and Constraints

This paper considers the setting where the learner has its own preferences that it additionally takes into consideration, and designs learner-aware teaching algorithms, which show that significant performance improvements can be achieved overLearner-agnostic teaching.

Deep Knowledge Tracing

The utility of using Recurrent Neural Networks to model student learning and the learned model can be used for intelligent curriculum design and allows straightforward interpretation and discovery of structure in student tasks are explored.

Towards Black-box Iterative Machine Teaching

This paper proposes an active teacher model that can actively query the learner for estimating theLearner's status, and provide the sample complexity for both teaching and query, respectively.

Preference-Based Batch and Sequential Teaching: Towards a Unified View of Models

A novel framework which captures the teaching process via preference functions $\Sigma$ and identifies preference functions inducing a novel family of sequential models with teaching complexity linear in the VC dimension of the hypothesis class: this is in contrast to the best known complexity result for the batch models which is quadratic in theVC dimension.

Near-Optimally Teaching the Crowd to Classify

This work proposes a natural stochastic model of the learners, modeling them as randomly switching among hypotheses based on observed feedback, and develops STRICT, an efficient algorithm for selecting examples to teach to workers.