Interactive Learning

# click the upper-left icon to select videos from the playlist

source: Simons Institute    2017年2月13日
Interactive learning is a modern machine learning paradigm of significant practical and theoretical interest, where the algorithm and the domain expert engage in a two-way dialog to facilitate more accurate learning from less data compared to the classical approach of passively observing labeled data. This workshop will explore several topics related to interactive learning broadly defined, including active learning, in which the learner chooses which examples it wants labeled; explanation-based learning, in which the human doesn't merely tell the machine whether its predictions are right or wrong, but provides reasons in a form that is meaningful to both parties; crowdsourcing, in which labels and other information are solicited from a gallery of amateurs; teaching and learning from demonstrations, in which a party that knows the concept being learned provides helpful examples or demonstrations; and connections and applications to recommender systems, automated tutoring and robotics. Key questions we will explore include what are the right learning models in each case, what are the demands on the learner and the human interlocutor, and what kinds of concepts and other structures can be learned. A main goal of the workshop is to foster connections between theory/algorithms and practice/applications. For more information, please visit https://simons.berkeley.edu/workshops/machinelearning2017-1
These presentations were supported in part by an award from the Simons Foundation.

Machine Learning from Verbal User Instruction 1:05:29 Tom Mitchell, Carnegie Mellon University
Machine Teaching in Interactive Learning 1:04:53
Interactively Learning Robot Objective Functions 48:05
Words, Pictures, and Common Sense 48:26
Stochastic Variance Reduction Methods for Policy Evaluation 47:16
Interactive Clustering 45:26
Crowdsourcing and Machine Learning 44:33
Active Learning Beyond Label Feedback 47:33
Active Learning for Multidimensional Experimental Spaces of Biological Responses 44:56
Interactive Language Learning from the Extremes 1:04:43
Robot Learning from Motor-Impaired Teachers and Task Partners 55:29
Sample-Efficient Reinforcement Learning with Rich Observations 58:48
Robots Learning from Human Interactions 45:00
Robot Learning, Interaction and Reliable Autonomy 52:59
Interactive Learning of Parsers from Weak Supervision 1:11:57
Power of Active Sampling for Unsupervised Learning 51:35
Leveraging Union of Subspace Structure to Improve Constrained Clustering 1:00:42
Hierarchical Learning for Human-Robot Collaboration 43:39
Active Nearest-Neighbor Learning in Metric Spaces 49:37
Systems Presentation 58:44
TicToc: A General Technique for Near-Optimal Active Learning with Noise 53:59
Interactive Learning of Mixtures of Submodular Functions 1:03:53

No comments: