Optimal Online Learning Procedures for Model-Free Policy Evaluation

A semiparametric statistical approach to model-free policy evaluation

Proceedings of the 25th international conference on Machine learning - ICML '08 ◽

10.1145/1390156.1390291 ◽

2008 ◽

Cited By ~ 4

Author(s):

Tsuyoshi Ueno ◽

Motoaki Kawanabe ◽

Takeshi Mori ◽

Shin-ichi Maeda ◽

Shin Ishii

Keyword(s):

Policy Evaluation ◽

Statistical Approach ◽

Model Free

Download Full-text

E-learning rhetoric, creativity and innovation

Research in Learning Technology ◽

10.3402/rlt.v13i2.10978 ◽

2005 ◽

Vol 13 (2) ◽

Author(s):

Jane Seale

Keyword(s):

Online Learning ◽

Policy Evaluation ◽

Learning Technology ◽

Current Position ◽

The Past ◽

E Learning ◽

Creativity And Innovation

In this issue of ALT-J we have five papers that cover a range of policy, evaluation and development issues. The first paper, by Smith, sets the scene for the remaining papers with its focus on policy and how this may be influenced by rhetoric, and in turn may influence creativity and innovation. In ‘From flowers to palms: 40 years of policy for online learning’, Smith presents a review of learning technology-related policy over the past 40 years. The purpose of the review is to make sense of the current position in which the field finds itself, and to highlight lessons that can be learned from the implementation of previous policies.DOI: 10.1080/09687760500104039

Download Full-text

Truly Batch Apprenticeship Learning with Deep Successor Features

Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence ◽

10.24963/ijcai.2019/819 ◽

2019 ◽

Cited By ~ 2

Author(s):

Donghun Lee ◽

Srivatsan Srinivasan ◽

Finale Doshi-Velez

Keyword(s):

Data Acquisition ◽

Policy Evaluation ◽

Learning Algorithm ◽

Industrial Process ◽

Feature Representation ◽

Sepsis Management ◽

Policy Model ◽

Model Free ◽

Apprenticeship Learning ◽

Batch Data

We introduce a novel apprenticeship learning algorithm to learn an expert's underlying reward structure in off-policy model-free batch settings. Unlike existing methods that require hand-crafted features, on-policy evaluation, further data acquisition for evaluation policies or the knowledge of model dynamics, our algorithm requires only batch data (demonstrations) of the observed expert behavior. Such settings are common in many real-world tasks---health care, finance, or industrial process control---where accurate simulators do not exist and additional data acquisition is costly. We develop a transition-regularized imitation learning model to learn a rich feature representation and a near-expert initial policy that makes the subsequent batch inverse reinforcement learning process viable. We also introduce deep successor feature networks that perform off-policy evaluation to estimate feature expectations of candidate policies. Under the batch setting, our method achieves superior results on control benchmarks as well as a real clinical task of sepsis management in the Intensive Care Unit.

Download Full-text

Towards Automatic Evaluation of Dialog Systems: A Model-Free Off-Policy Evaluation Approach

10.18653/v1/2021.emnlp-main.589 ◽

2021 ◽

Author(s):

Haoming Jiang ◽

Bo Dai ◽

Mengjiao Yang ◽

Tuo Zhao ◽

Wei Wei

Keyword(s):

Policy Evaluation ◽

Automatic Evaluation ◽

Dialog Systems ◽

Model Free ◽

Evaluation Approach

Download Full-text

Representation, abstraction, and simple-minded sophisticates

Behavioral and Brain Sciences ◽

10.1017/s0140525x19002942 ◽

2020 ◽

Vol 43 ◽

Author(s):

Peter Dayan

Keyword(s):

Decision Theory ◽

Predictive Coding ◽

Bayesian Decision Theory ◽

Bayesian Decision ◽

Model Based ◽

Model Free

Abstract Bayesian decision theory provides a simple formal elucidation of some of the ways that representation and representational abstraction are involved with, and exploit, both prediction and its rather distant cousin, predictive coding. Both model-free and model-based methods are involved.

Download Full-text

Election Poster Persuasion

Social Psychology ◽

10.1027/1864-9335/a000323 ◽

2018 ◽

Vol 49 (1) ◽

pp. 3-15 ◽

Cited By ~ 1

Author(s):

Malte Schott ◽

Jule Wolf

Keyword(s):

Political Parties ◽

Policy Evaluation ◽

Political Party ◽

Need For Cognition ◽

Evaluation Study ◽

Additional Measure ◽

Political Preferences ◽

Policy Statements ◽

Voting Intentions

Abstract. We examined the effect of presenting unknown policy statements on German parties’ election posters. Study 1 showed that participants inferred the quality of a presented policy from knowledge about the respective political party. Study 2 showed that participants’ own political preferences influenced valence estimates: policy statements presented on campaign posters of liked political parties were rated significantly more positive than those presented on posters of disliked political parties. Study 3 replicated the findings of Study 2 with an additional measure of participants’ need for cognition. Need for cognition scores were unrelated to the valence transfer from political parties to policy evaluation. Study 4 replicated the findings of Studies 2 and 3 with an additional measure of participants’ voting intentions. Voting intentions were a significant predictor for valence transfer. Participants credited both their individually liked and disliked political parties for supporting the two unknown policies. However, the credit attributed to the liked party was significantly higher than to the disliked one. Study 5 replicated the findings of Studies 2, 3, and 4. Additionally, participants evaluated political clubs that were associated with the same policies previously presented on election posters. Here, a second-degree transfer emerged: from party valence to policy evaluation and from policy evaluation to club evaluation. Implications of the presented studies for policy communications and election campaigning are discussed.

Download Full-text