Iterative Teaching by Label Synthesis
arXiv:2110.14432
Abstract
In this paper, we consider the problem of iterative machine teaching, where a teacher provides examples sequentially based on the current iterative learner. In contrast to previous methods that have to scan over the entire pool and select teaching examples from it in each iteration, we propose a label synthesis teaching framework where the teacher randomly selects input teaching examples (e.g., images) and then synthesizes suitable outputs (e.g., labels) for them. We show that this framework can avoid costly example selection while still provably achieving exponential teachability. We propose multiple novel teaching algorithms in this framework. Finally, we empirically demonstrate the value of our framework.
arXiv commentNeurIPS 2021 Spotlight (v5: 28 pages, 20 figures, fixed typos in v4)
References in corpus (23)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Distilling the Knowledge in a Neural Network
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- Learning to Optimize
- Coresets for Scalable Bayesian Logistic Regression
- An Overview of Machine Teaching
- Iterative Machine Teaching
- Learning to Teach with Dynamic Loss Functions
- Near-Optimally Teaching the Crowd to Classify
- Sub-Sampled Newton Methods II: Local Convergence Rates
- Policy Poisoning in Batch Reinforcement Learning and Control
- Understanding the Role of Adaptivity in Machine Teaching: The Case of Version Space Learners
- Teaching Inverse Reinforcement Learners via Features and Demonstrations
- Training Set Debugging Using Trusted Items
- Learner-aware Teaching: Inverse Reinforcement Learning with Preferences and Constraints
- The Teaching Dimension of Linear Learners
- Preference-Based Batch and Sequential Teaching: Towards a Unified View of Models
- Optimal Cooperative Inference
- Neural Similarity Learning
- Machine Teaching of Active Sequential Learners
- Toward a general, scaleable framework for Bayesian teaching with applications to topic models
- The Teaching Dimension of Kernel Perceptron
- Average-case Complexity of Teaching Convex Polytopes via Halfspace Queries