China
Africa
Explore chapters and articles related to this topic
Risk and social interaction (samhandling) to meet the unforeseen
Published in Stein Haugen, Anne Barros, Coen van Gulijk, Trond Kongsvik, Jan Erik Vinnem, Safety and Reliability – Safe Societies in a Changing World, 2018
G.E. Torgersen, T.J. Steiro, L.I. Magnussen
We have seen from paragraph 4, that social interaction plays an important role. In another study, by Nyhus, Steiro & Torgersen (2018), mentoring and coaching was studied for a joint operation course held at the Norwegian Defence College. Personnel form all weapon branches participated (Air, Army and Sea). The purpose of the course was to increase understanding of joint operations (Andersen, 2016; FHS, 2015). Since conflicts and wars rarely follow a familiar pattern, unpredictable factors are tried to be put into the education (Heier, 2015). The course is executed by more experienced officers, guiding less experienced officers and soldiers with given military cases and scenarios that are often based on experiences from actual events. The theoretical framework applied was therefore “apprenticeship learning”. Apprenticeship is rooted in sociocultural learning theory (Saljø), which emphasizes that knowledge is constructed through social interaction (social interaction) in a context and not primarily through individual processes (Lave & Wenger, 1991, Maguire, 1999, Dysthe, 2001). 10 hours of observation was conducted in the sessions. In addition five group interviews were conducted with a total of 23 informants (out of a number of total 100 students). After a sufficient number of data was obtained (saturation), the interviews were stopped. A thematic analysis was adopted when interpretating the interview material. This is a suitable method for identifying, analysing, and reporting patterns within the data analysis (Braun & Clarke, 2006). The analysis reveals that some supervisors emphasize mainly on the product, and others more on how the group reached the result, i.e. the process. The product said something about the specific deliveries that the group arrived at. Furthermore, the product, among other things, referred to the supervisors who made sure that the group followed certain structured patterns of action and worked towards a goal that was embodied in doctrines and drills. Nyhus et al. (2018) found further that mentors who focused a lot on the product in the preparation phase did not take the unforeseen into account. A process-oriented supervisor with authority enough to create a shared commitment and understanding of situations. By the virtues of experience and competence, the supervisor can pave the way in preparation for meeting the unforeseen.
A knowledge-driven layered inverse reinforcement learning approach for recognizing human intents
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2020
R. Bhattacharyya, S. M. Hazarika
Sequential windowed inverse reinforcement learning (SWIRL) has been proposed by Krishnan et al. (2019) which falls within the sequential IRL approach. SWIRL learns a sequence of local reward functions from multi-step tasks. It computes a policy over the sequence of local rewards using RL. The task of learning from an expert is referred to as apprenticeship learning (Abbeel & Ng, 2004); hierarchical apprenticeship learning has been proposed by Kolter, Abbeel, and Ng (2007) which allows to extend the apprenticeship learning paradigm to the complex domain. The algorithm is furnished with a hierarchy of sub-tasks to learn bipedal locomotion. The RL community is working towards eliminating the problems of applying the existing techniques in complex tasks involving delayed rewards. It has been observed that reward functions which capture the structure of these complex tasks are beneficial. Thus, the decomposition of the reward function is carried out in their work, instead of considering a single monolithic reward function. The work reported in Köpf, Inga, Rothfuß, Flad, and Hohmann (2017) is an extension to maximum entropy IRL (MaxEnt-IRL) (Ziebart et al., 2008). Their proposed game theoretic IRL framework has been described and analysed using a two-player control architecture.
Detecting Physiological Needs Using Deep Inverse Reinforcement Learning
Published in Applied Artificial Intelligence, 2022
Khaoula Hantous, Lilia Rejeb, Rahma Hellali
(3) Infinite-state MDP with unknown optimal policy also called Apprenticeship Learning with Sample trajectories. It is designed to handle complex applications in which the desired trajectory is hard to describe e.g., the helicopter aerobatic maneuver trajectory that depends on the helicopter dynamics (Abbeel, Coates, and Ng, 2010). The demonstrations provided by the expert can be used to extract the desired trajectory through apprenticeship learning.