Article
Multidisciplinary Sciences
Raphael Le Bouc, Mathias Pessiglione
Summary: Despite being aware of potential adverse consequences, humans still procrastinate. This study uses fMRI to understand the neuro-computational mechanisms underlying procrastination. The findings suggest that procrastination is influenced by a cognitive bias that attenuates the expected effort cost with delay, making tasks appear less effortful when postponed.
NATURE COMMUNICATIONS
(2022)
Article
Psychology, Developmental
Ruth Speidel, Laura Zimmermann, Lawrie Green, Natalie H. Brito, Francys Subiaul, Rachel Barr
Summary: Older preschoolers tend to overimitate more than younger ones, and rates of overimitation increase with age. Cognitive load and processing costs impact imitation and overimitation, with older children outperforming younger ones.
JOURNAL OF EXPERIMENTAL CHILD PSYCHOLOGY
(2021)
Review
Physics, Multidisciplinary
Kana Moriwaki, Takahiro Nishimichi, Naoki Yoshida
Summary: In the next decade, a series of large observational programs using ground-based and space-borne telescopes are planned. These programs are expected to generate an unprecedented amount of data, exceeding an exabyte. Processing such massive astronomical data poses technical challenges and calls for urgently needed fully automated technologies based on machine learning and artificial intelligence. Maximizing the scientific potential of big data requires collaborative efforts from the scientific community. In this article, we summarize the recent progress in the application of machine learning in observational cosmology, as well as address crucial issues in high-performance computing for data processing and statistical analysis.
REPORTS ON PROGRESS IN PHYSICS
(2023)
Article
Computer Science, Artificial Intelligence
Yao Li, Yuhui Wang, Xiaoyang Tan
Summary: Goal-conditioned reinforcement learning aims to control agents to reach desired goals, but current methods have limitations in sample efficiency and training data. This paper proposes integrating self-imitation learning with goal-conditioned RL to create a compatible framework. The introduced target action value function effectively combines these two training mechanisms and can learn a superior policy compared to existing approaches.
PATTERN RECOGNITION
(2023)
Article
Automation & Control Systems
Nicolas Bougie, Ryutaro Ichise
Summary: This paper proposes a novel goal-conditioned method that leverages very small sets of goal-driven demonstrations to massively accelerate the learning process. By introducing the concept of active goal-driven demonstrations and a strategy for prioritizing sampling of goals where the disagreement between the expert and the policy is maximized, the method outperforms prior imitation learning approaches in terms of exploration efficiency and average scores in most tasks, as demonstrated by experimental results in various benchmark environments from the Mujoco domain.
AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS
(2021)
Article
Multidisciplinary Sciences
Bjorn Lindstrom, Martin Bellander, David T. Schultner, Allen Chang, Philippe N. Tobler, David M. Amodio
Summary: The popularity of social media may be driven by reward learning mechanisms, where users adjust their behavior in posting based on the type of rewards, with individual differences in social reward learning. Computational modeling analyses show that human behavior on social media conforms to the principles of reward learning.
NATURE COMMUNICATIONS
(2021)
Article
Economics
Harry Pei
Summary: Under limited memory and observational learning, consumers are concerned that the seller may not take the Stackelberg action when his reputation is good, but will take it after losing his reputation. Such concern results in low payoff for the seller in reputation building. The reputation failure is influenced by consumers' observational learning.
REVIEW OF ECONOMIC STUDIES
(2023)
Article
Automation & Control Systems
Yoshihisa Tsurumine, Takamitsu Matsubara
Summary: Generative Adversarial Imitation Learning (GAIL) is a method that can learn policies from demonstrations without explicitly defining the reward function. This paper proposes Goal-Aware Generative Adversarial Imitation Learning (GA-GAIL), which addresses the issue of imperfect demonstration data by introducing a second discriminator to distinguish the goal state. GA-GAIL extends the standard GAIL framework to robustly learn desirable policies even from imperfect demonstrations.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2022)
Article
Psychology, Developmental
Francys Subiaul
Summary: This article explores different forms of social learning in preschoolers through two studies. The results show that familiar imitation and goal emulation develop early, while novel imitation and affordance learning develop late. In addition, the study finds that imitation is domain-specific, while emulation has multidimensional characteristics in spatial tasks and unidimensional characteristics in cognitive tasks. These results reveal the mosaic nature of children's social learning development.
COGNITIVE DEVELOPMENT
(2023)
Correction
Multidisciplinary Sciences
Bjoern Lindstroem, Martin Bellander, David T. Schultner, Allen Chang, Philippe N. Tobler, David M. Amodio
Summary: This paper has been corrected and the updated version is available at https://doi.org/10.1038/s41467-021-22067-6.
NATURE COMMUNICATIONS
(2021)
Article
Behavioral Sciences
Suwandschieff Elisabeth, Wein Amelia, Folkertsma Remco, Bugnyar Thomas, Huber Ludwig, Schwing Raoul
Summary: Social learning is an adaptive strategy that reduces trial-and-error learning risks. Different mechanisms of social learning are distinguished based on the type of information acquired and associations formed. Imitation, considered cognitively demanding, is one such process associated with high-fidelity response matching. A replication study was conducted on kea birds, observing their behavior after witnessing a trained demonstrator. Although motor imitation was not found, the study revealed strong social effects on exploration rates, suggesting possible emulation or selective imitation tendencies.
Article
Computer Science, Artificial Intelligence
Yirui Zhou, Mengxiao Lu, Xiaowei Liu, Zhengping Che, Zhiyuan Xu, Jian Tang, Yangchun Zhang, Yan Peng, Yaxin Peng
Summary: Generative adversarial imitation learning (GAIL) treats imitation learning (IL) as a distribution matching problem between expert policy and learned policy. This paper focuses on the generalization and computational properties of policy classes. It is proven that GAIL can ensure generalization when policies are well controlled. By incorporating distributional reinforcement learning (RL) into GAIL, the greedy distributional soft gradient (GDSG) algorithm is proposed to solve GAIL. GDSG has advantages including alleviating Q-value overestimation problem and improving policy performance through sufficient exploration, as well as attaining a sublinear convergence rate to a stationary solution. Comprehensive experimental verification in MuJoCo environments demonstrates that GDSG outperforms previous GAIL variants in mimicking expert demonstrations.
Article
Computer Science, Hardware & Architecture
Hannes Holm
Summary: This article introduces the red team emulation tool Lore, which uses boolean logic and trained models to automatically select and execute red team actions. Lore improves the current state of red team automation and provides a more fun and educational experience compared to manual red teaming in cyber defence exercises. Empirical tests show that Lore's trained models result in double the compromised machines compared to expert-defined models and five times more compromised machines compared to random action selection.
IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING
(2023)
Review
Biochemistry & Molecular Biology
Andra V. Krauze, Kevin Camphausen
Summary: Computational approaches, including machine learning and artificial intelligence, are increasingly important in all medical specialties as large data repositories are being optimized. Radiation oncology, as a discipline, is at the forefront of large-scale data acquisition and analysis, with the potential to improve patient outcomes.
INTERNATIONAL JOURNAL OF MOLECULAR SCIENCES
(2021)
Review
Neurosciences
Weixi Kang, Sonia Pineda Hernandez, Jie Mei
Summary: Observational learning is an important way for humans and animals to learn stimulus-response associations by observing others' behavior. It relies on individuals' ability to map others' actions into their own behaviors and process outcomes to achieve goals.
FRONTIERS IN HUMAN NEUROSCIENCE
(2021)