Review
Psychology, Mathematical
Leendert van Maanen, Steven Miletic
Summary: The rise of computational modeling in the past decade has led to an increase in papers reporting parameter estimates. Model identification requires specific parameter settings, and caution is advised against overinterpreting the associative relations found.
PSYCHONOMIC BULLETIN & REVIEW
(2021)
Article
Biology
Olof Leimar, Sasha R. X. Dall, Alasdair I. Houston, John M. McNamara
Summary: Interactions in social groups can promote behavioral specialization, and individuals achieve specialization by learning to choose specific actions. Specialization develops more rapidly when there are few neighbors in a network and when learning rates are high. Frequency-dependent competition for resources is the main driver of specialization.
PROCEEDINGS OF THE ROYAL SOCIETY B-BIOLOGICAL SCIENCES
(2022)
Article
Psychology, Biological
Jasmin A. Strickland, Joseph M. Austen, Rolf Sprengel, David J. Sanderson
Summary: The study suggests that the GluA1 subunit plays a role in consumption behavior, but its involvement in encoding hedonic value may not be significant. Testing sensitivity to sucrose rewarding properties under negative/positive contrast effects showed that both Grial knockout mice and wild-type controls preferred the CS+ flavor over the CS- flavor.
PHYSIOLOGY & BEHAVIOR
(2021)
Article
Computer Science, Artificial Intelligence
Souhir Khessiba, Ahmed Ghazi Blaiech, Khaled Ben Khalifa, Asma Ben Abdallah, Mohamed Hedi Bedoui
Summary: Electroencephalography (EEG) is commonly used for studying brain electrical activity. Deep learning networks can accurately predict individuals' states of vigilance based on EEG signals. Experimental results demonstrate that the proposed 1D-UNet and 1D-UNet-LSTM models perform well in stabilizing training and recognizing vigilance states, indicating the effectiveness of the proposed methods.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Multidisciplinary Sciences
Thomas N. Sherratt, Erica O'Neill
Summary: Signal detection theory (SDT) is widely used for optimal decision-making under uncertainty, but it assumes decision-makers immediately adopt the appropriate acceptance threshold, which may not be the case in real-world situations that require learning. This study recasts the traditional SDT model into a contextual multi-armed bandit (CMAB), where decision-makers must infer the relationship between a continuous cue and the desirability of a signal while seeking to exploit the acquired information. Various CMAB heuristics are discussed to address the trade-off between estimating the underlying relationship and exploiting it. The results suggest that CMABs provide principled parametric solutions to SDT problems when decision-makers have incomplete information.
ROYAL SOCIETY OPEN SCIENCE
(2023)
Article
Agriculture, Dairy & Animal Science
Sabrina Brando, Lillian Basom, Meredith Bashaw, Caitlin Druyor, Ellen Fonte, Roger Thompson
Summary: Coercion and non-voluntary procedures can lead to fear and maladaptive behaviors in captive animals, affecting animal welfare. Positive reinforcement training has been shown to reduce fear responses and encourage voluntary cooperation in various species. Individualized target training can facilitate voluntary movement of captive animals, benefiting both animals and caretakers.
Review
Chemistry, Multidisciplinary
Abdikarim Mohamed Ibrahim, Kok-Lim Alvin Yau, Yung-Wey Chong, Celimuge Wu
Summary: Recent advancements in deep reinforcement learning have enabled its application in multi-agent scenarios, with MADRL allowing multiple agents to interact and learn from each other. Currently, MADRL shows significant performance improvements in various multi-agent domains.
APPLIED SCIENCES-BASEL
(2021)
Review
Computer Science, Information Systems
Liying Yang, Sheng-Feng Qin
Summary: Emotion is defined as a subject's response to external or internal stimulus events, reflected in changes of facial expression, gesture, gait, eye-movement, etc., which has significant impact on both physical and mental health and work performance. The primary challenge in emotion recognition is to easily and accurately identify emotional states.
Article
Computer Science, Artificial Intelligence
Liwen Zhu, Peixi Peng, Zongqing Lu, Yonghong Tian
Summary: Traffic signal control aims to improve traffic efficiency by coordinating signals across intersections. However, challenges such as neighbor influence and poor policy generalizability exist. To address these issues, a novel Meta Variationally Intrinsic Motivated RL method is proposed, which learns decentralized policies considering neighbor information and introduces intrinsic rewards for stable policy learning.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Carla Fernandez, Martin Gonzalez-Rodriguez, Daniel Fernandez-Lanvin, Javier De Andres, Miguel Labrador
Summary: This paper presents a supervised classifier based on machine learning to identify the operative hand as left or right, without relying on gyroscopes or accelerometers, making it applicable to any touchscreen device.
PEERJ COMPUTER SCIENCE
(2021)
Article
Computer Science, Information Systems
Kok-Lim Alvin Yau, Yung-Wey Chong, Xiumei Fan, Celimuge Wu, Yasir Saleem, Phei-Ching Lim
Summary: This paper presents the application of various variants of reinforcement learning (RL) in diabetes management. It focuses on improving blood glucose levels and the similarity between RL and physician's policies. The paper discusses the attributes of RL, essential training elements, representation of diabetes attributes, and different types of RL algorithms. It also explores open issues and potential future developments in using RL for diabetes management.
Article
Biology
Sergey Levine, Dhruv Shah
Summary: Navigation is a complex problem in robotics, traditionally approached through geometric mapping and planning. However, machine learning offers a new approach by allowing robots to make decisions based on prior experiences, considering physical outcomes and patterns in real-world environments. This article presents a toolkit for experiential learning of robotic navigation skills, unifying recent approaches, discussing design principles, experimental results, and directions for future work.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY B-BIOLOGICAL SCIENCES
(2023)
Article
Automation & Control Systems
Alekh Agarwal, Sham M. Kakade, Jason D. Lee, Gaurav Mahajan
Summary: This work provides provable characterizations of policy gradient methods in the context of discounted Markov Decision Processes, focusing on different policy parameterizations and providing approximation guarantees that avoid explicit worst-case dependencies on the size of state space.
JOURNAL OF MACHINE LEARNING RESEARCH
(2021)
Article
Computer Science, Artificial Intelligence
Zixuan Deng, Yanping Xiang, Zhongfeng Kang
Summary: This paper discusses the incomplete state representation in simulation as one cause of errors and proposes a supervised learning approach to correct human-unacceptable policies calculated by simulators using human feedback. The approach involves detecting blind spots, training classifiers on noisy human feedback, and correcting policies through a complementary model based on linear function approximation and a policy iteration algorithm that uses radial basis functions. Experiments show the approach's higher accuracy compared to baselines in terms of human suboptimality, human errors, and human feedback types, as well as its scalability.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Physics, Multidisciplinary
Jia-Hao Cao, Feng Chen, Qi Liu, Tian-Wei Mao, Wen-Xin Xu, Ling-Na Wu, Li You
Summary: Discrimination of entangled states is crucial in quantum-enhanced metrology, often requiring low-noise detection technology. This challenge can be overcome by introducing a nonlinear readout process. In this study, reinforcement learning is used to manipulate the spin-mixing dynamics in a spin-1 atomic condensate to achieve nonlinear readout of highly entangled states.
PHYSICAL REVIEW LETTERS
(2023)