In 1969 Charles Kelley described adaptive training as ‘training in which the problem, the stimulus, or the task is varied as a function of how well the trainee performs’, since this seminal work by Kelley adaptive training has become a concept of interest to researchers, educators and learners.
Adaptive Training (AT) is a personalised approach to learning and development that uses data and technology to tailor educational experiences to the needs, abilities, and preferences of individual learners. AT is proposed to improve the speed and robustness of learning and enable trainees to achieve expertise quickly. But what is the evidence for adaptive training? Is it effective? And how does it work? In this short article, I want to introduce the concepts behind AT and explain ways it might be realised using technology such as eye tracking and virtual reality.
During adaptive training, a demanding or complex task is first simplified according to the trainee’s current performance level, so that the difficulty of the task can be gradually increased as learning occurs. Early forms of AT were manual in nature and brought to life by skilled and experienced instructors working on a one-to-one basis with learners or trainees. Instructors had to perceive and interpret verbal and non-verbal cues from the trainee and make adjustments and adaptations to their style and approach of teaching. For example, if a learner has body language or facial expressions which suggest they aren’t understanding, or they are explicitly getting the wrong answers or showing poorer performance, the instructor could adjust training appropriately to elicit a change. In an early piece of research on this topic, Bloom (1984) found that students who received this kind of one-on-one and bespoke tutoring performed two standard deviations higher than students in a typical classroom setting. Super!
Computers to the Rescue
But rather than celebrate this finding, Bloom viewed this as a problem… the so-called ‘2 sigma problem’. Bloom recognised that while one-to-one adaptive tutoring was clearly more effective (making you 2 standard deviations, or 2 sigma, better), it was also time-consuming, costly, and unlikely to be achievable at any scale. The value of AT was not going to be impactful if the time, money, and resources needed to make it happen were too high. Thankfully, the human factors research community had recognised the value of computer-enabled training systems, which allowed algorithms to acquire and process information about the trainee or operators’ behaviour. Computers were able to fulfil the role of the skilled trainer and monitor the trainee. The computer delivering the training stimulus (the questions or the tasks) could then also adapt to this measurement of trainee performance. Thus, the first real-time and automated adaptive training devices were made. These promised to be cheaper, less onerous, and thus scalable.
The Evolution of AT
From this initial work has grown a whole body of AT literature, and some distinctive approaches to achieve the same general aim. Specifically, three types of AT have emerged within the scientific literature over the past 30+ years. One approach, called macro adaptation, allows instructors to choose either alternate instructional goals, depth of content, or delivery prior to training based on a student’s general ability or achievement level. A second type called the Aptitude Treatment Interaction (ATI) approach, adapts to a student’s aptitudes or abilities prior to training and provides content that is matched to that ability. A third approach is the micro-adaptive approach. In this approach, training is adapted to a student’s performance during training in real time. Finally, in the hybrid approach, the ATI and micro approaches are combined. In training when there is a paucity of performance data, adaptations are first based on a student’s basic aptitudes, but later in training when more performance data is available, training starts to adapt to these as well (e.g. performance on a task, for example, Park & Lee, 2003). Typically, in all applications of AT Adaptive training algorithms alter subtask difficulty as a function of subject performance across a range of related subtasks, a process also known as dynamic difficulty adjustment (Hunicke, 2005; Zohaib, 2018; Moon and Seo, 2020).
AT in Action
AT has been operationalised in many different ways in recent years, using a range of different computerised trainee assessment techniques (performance, psychophysiology, brain activity etc) and generating a number of different adaptation approaches. It is beyond the scope of this article to review all of these applications but below are a range of different literature which report on the approach and effectiveness of some of the AT devices designed and developed.
| – Van Buskirk, W. L., Steinhauser, N. B., Mercado, A. D., Landsberg, C. R., & Astwood, R. S. (2014). A Comparison of the Micro-Adaptive and Hybrid Approaches to Adaptive Training. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 58(1), 1159-1163. https://doi.org/10.1177/1541931214581242 |
| – A. Dey, A. Chatburn and M. Billinghurst, “Exploration of an EEG-Based Cognitively Adaptive Training System in Virtual Reality,” 2019 IEEE Conference on Virtual Reality and 3D User Interfaces (VR), Osaka, Japan, 2019, pp. 220-226, doi: 10.1109/VR.2019.8797840. |
| – Hale, K.S., Del Giudice, K., Flint, J., Wilson, D.P., Muse, K., Kudrick, B. (2015). Designing, Developing, and Validating an Adaptive Visual Search Training Platform. In: Schmorrow, D.D., Fidopiastis, C.M. (eds) Foundations of Augmented Cognition. AC 2015. Lecture Notes in Computer Science(), vol 9183. Springer, Cham. https://doi.org/10.1007/978-3-319-20816-9_70 |
| – Aguilar Reyes, C.I., Wozniak, D., Ham, A. et al. Design and evaluation of an adaptive virtual reality training system. Virtual Reality 27, 2509–2528 (2023). https://doi.org/10.1007/s10055-023-00827-7 |
| Finseth, T., Dorneich, M. C., Keren, N., Franke, W. D., & Vardeman, S. (2024). Virtual Reality Adaptive Training for Personalized Stress Inoculation. Human Factors, 0(0). https://doi.org/10.1177/00187208241241968 |
A study by Finseth et al (2024) is my favourite. They deliver Stress Exposure Training (SET) using AT principles. They were able to demonstrate that adaptive stress exposure training was better than non-adaptive.
The Cineon Approach
At Cineon we have taken a unique approach to the development of adaptive training. Harnessing the power of virtual reality and eye-tracking technology, we have developed an API which allows any training to become seamlessly adaptive. Our adaptation algorithms are based not just on conventional performance data, but also data about how the trainee is behaving and how they might be thinking, that we garner from eye tracking technology the trainee is wearing.
Eye Tracking
The main challenge of adaptive training is the ability to monitor and understand the trainee in real-time. Sensors are needed to monitor movement, performance, emotions and cognitions but this is not easy. The hardware can be cumbersome and unreliable and can disrupt the training experience. As such we believe that eye-tracking technology, which is lightweight, robust, and unobtrusive is the best sensor to focus on for enabling real-time assessment of trainee performance.
Data
But while eye tracking data is reasonably easy to collect (particularly in VR training, for which eye trackers are commonly built into the heads-up display), it’s often quite tricky to make sense of these data in a way that is useful for training. Sensor data from eye trackers need to be understood and interpreted for it to be useful in AT applications. We have developed a pipeline of data capture and analysis to make sense of the data we receive and be able to initiate adaptations based on it. We call this PAVE.
So, at Cineon we harness the powers of Virtual reality, eye tracking and data science to develop new forms of adaptive training applicable to a range of different high-performance or safety-critical industries, including aviation, heavy industry, and the military. To learn more, see some of our existing products – TACET, iSAVE, RCAT, and INCORA, which include these adaptive features.
References
Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 13(6), 4-16.
Park, O., & Lee, J. (2003). Adaptive Instructional Systems. Educational Technology Research and Development.
Linters, G. (1978). Adaptive training of perceptual-motor skills: issues, results, and future directions. International Journal of Man-Machine Studies.
Finseth, T., Dorneich, M. C., Keren, N., Franke, W., Vardeman, S., Segal, J., Deick, A., Cavanah, E., & Thompson, K. (2021). The effectiveness of adaptive training for stress inoculation in a simulated astronaut task. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 65(1), 1541-1545. https://doi.org/10.1177/1071181321651241
Finseth, T., Dorneich, M. C., Keren, N., Franke, W. D., & Vardeman, S. (2024). Virtual Reality Adaptive Training for Personalized Stress Inoculation. Human Factors, 0(0). https://doi.org/10.1177/00187208241241968
Hunicke, R. (2005). The Case for Dynamic Difficulty Adjustment in Games. ACM SIGCHI. Zohaib, M. (2018). Dynamic Difficulty Adjustment (DDA) in Computer Games: A Review. Advances in Human-Computer Interaction.
Moon, J., & Seo, J. (2020). Dynamic Difficulty Adjustment in Digital Games: A Review. Entertainment Computing.
