Article
Computer Science, Artificial Intelligence
Eleftherios Triantafyllidis, Fernando Acero, Zhaocheng Liu, Zhibin Li
Summary: This paper presents a hybrid hierarchical learning framework called ROMAN, which integrates behavioural cloning, imitation learning, and reinforcement learning to solve multiple complex tasks over long time horizons in robotic manipulation. By coordinating an ensemble of specialized neural networks, ROMAN generates correct sequential activations to accomplish long sequences of sophisticated manipulation tasks and exhibits robustness to sensory noises.
NATURE MACHINE INTELLIGENCE
(2023)
Article
Biology
Anthony Formaux, Dany Paleressompoulle, Joel Fagot, Nicolas Claidiere
Summary: Conventions are an important aspect of human social and cultural behavior, and they may also play a crucial role in animal societies. However, our understanding of non-human conventions is limited. Through experimentation, it has been found that conventions can readily emerge in non-human primates and exhibit similar fundamental properties as human conventions.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY B-BIOLOGICAL SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Zhihao Cheng, Li Shen, Miaoxi Zhu, Jiaxian Guo, Meng Fang, Liu Liu, Bo Du, Dacheng Tao
Summary: This paper proposes an algorithm that can adaptively learn safe policies from a single expert dataset under diverse safety constraints. It introduces the use of a Lagrange multiplier to balance imitation and safety performance, and employs a two-stage optimization framework to solve the problem. Experimental results demonstrate the effectiveness of the approach.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Multidisciplinary Sciences
Krisztina Liszkai-Peres, Dora Kampis, Ildiko Kiraly
Summary: The study found that 3-4-year-old children are able to adapt to an altered context after a 1 week delay and remember and apply previously irrelevant actions. This demonstrates the flexibility of preschooler's memory, but it is also influenced by previous demonstrated strategies.
Article
Humanities, Multidisciplinary
Amber Chen
Summary: The study found that in mainland China, there is a high proportion of mixed-aged classrooms among in-service Montessori teachers and administrators. However, there are also instances of co-teaching, lower student-teacher ratios, and shortened work cycles, indicating a departure from high-fidelity Montessori implementation. This shows the need for Chinese Montessori educators to balance localization with fidelity to achieve better student outcomes.
HUMANITIES & SOCIAL SCIENCES COMMUNICATIONS
(2021)
Article
Physics, Applied
Xu Han, Yimeng Xu, Bowen Sun, Ruixue Xu, Jing Xu, Wang Hong, Zhiwei Fu, He Zhu, Xin Sun, Jingjing Chang, Kai Qian
Summary: This study demonstrates a flexible transparent artificial nociceptor device that can simulate key features of biological nociceptors, including threshold, relaxation, hyperalgesia, and allodynia in response to external stimuli. Additionally, an alarm system was built to showcase the simplicity and feasibility of this artificial nociceptor for future neuromorphic systems.
APPLIED PHYSICS LETTERS
(2022)
Article
Business
Beata Glinka, Przemyslaw G. Hensel
Summary: This paper introduces a theoretical framework for studying imitation decisions in immigrant entrepreneurship, validates it through literature review and qualitative studies, discusses the specific character of immigrant entrepreneurs' imitation decisions, and highlights differences in that regard between immigrant entrepreneurs and established local businesses.
MANAGEMENT DECISION
(2021)
Article
Computer Science, Artificial Intelligence
Chao-Fan Pan, Xue-Yang Min, Heng-Ru Zhang, Guojie Song, Fan Min
Summary: This paper presents a method for modeling player behavior using imitation learning under the framework of meta-learning. By learning a generic behavior model from historical records and personalizing the policy by imitating individual player behavior, the method achieves good action similarity.
APPLIED INTELLIGENCE
(2023)
Article
Chemistry, Analytical
Tho Nguyen Duc, Chanh Minh Tran, Phan Xuan Tan, Eiji Kamioka
Summary: Imitation learning is effective for an autonomous agent to learn control policies using demonstrations from an expert without an explicit reward function. Domain adaptive imitation learning addresses the challenge of applying learned policies in distinct domains by learning how to perform a task optimally in a learner domain from demonstrations in an expert domain. A model based on Generative Adversarial Network is proposed to learn domain-shared and domain-specific features to find an optimal policy across domains, showing effectiveness in various tasks.
Article
Computer Science, Artificial Intelligence
Peng Ju, Yi Zhang
Summary: This paper proposes a knowledge distillation scheme for object detection, which utilizes inconsistency-based feature imitation and global relation imitation to improve performance without additional computational load.
Article
Automation & Control Systems
Jongcheon Park, Seungyong Han, S. M. Lee
Summary: This paper proposes a new imitation learning algorithm based on observation, where the action of a robot manipulator is trained to mimic a demonstrator's behavior using restored actions. The trajectory is generated through a recurrent generative adversarial network, and the action is restored from the output of a tracking controller. The algorithm eliminates the need for accessing the demonstrator's action and achieves better learning performances.
Article
Chemistry, Multidisciplinary
Junfeng Jiang, Yikang Rui, Bin Ran, Peng Luo
Summary: With the development of AI, the intelligence level of vehicles is increasing. The driving behavior decision-making of intelligent vehicles has always been a controversial and difficult research topic. This thesis proposes an intelligent vehicle driving behavior decision method based on DQN generative adversarial imitation learning (DGAIL) in the structured road traffic environment. The experimental results show that the DGAIL method can achieve safe and efficient driving on structured roads.
APPLIED SCIENCES-BASEL
(2023)
Article
Robotics
Yuzhe Qin, Hao Su, Xiaolong Wang
Summary: This paper proposes an imitation learning method for dexterous manipulation with multi-finger robot hand, where the policy is learned from human demonstrations and transferred to a real robot hand. The authors introduce a novel single-camera teleoperation system to efficiently collect 3D demonstrations, and construct a customized robot hand for each user to ensure stable data collection. The results show significant improvement over baselines in multiple complex manipulation tasks, and the learned policy is robust when transferred to the real robot.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Multidisciplinary Sciences
Anne Sibilsky, Heidi Colleran, Dominik Deffner, Daniel B. M. Haun
Summary: Observational learning plays a key role in cultural transmission, as shown in previous transmission chain experiments. This study examines children's copying fidelity in observational learning across different communities, finding that functional features are transmitted more faithfully than non-functional features, and the accuracy of transmission increases with age. Furthermore, being observed has varying effects on transmission across communities. Overall, the study shows that children have a high propensity and developing abilities for observational learning, allowing for effective cultural transmission.
Article
Multidisciplinary Sciences
James W. A. Strachan, Arianna Curioni, Merryn D. Constable, Gunther Knoblich, Mathieu Charbonneau
Summary: The ability to transmit information through social learning is fundamental to cultural evolution. There are two main theories on how this transmission occurs: copying behavior or reconstructing information. It is difficult to distinguish between these theories empirically, but manipulating the task context between model and learner can help differentiate their predictions.