Individual differences in the learning potential of human beings

On the promise of personalized learning for educational equity

Students enter school with a vast range of individual differences, resulting from the complex interplay between genetic dispositions and unequal environmental conditions. Schools thus face the challenge of organizing instruction and providing equal opportunities for students with diverse needs. Schools have traditionally managed student heterogeneity by sorting students both within and between schools according to their academic ability. However, empirical evidence suggests that such tracking approaches increase inequalities. In more recent years, driven largely by technological advances, there have been calls to embrace students' individual differences in the classroom and to personalize students' learning experiences. A central justification for personalized learning is its potential to improve educational equity. In this paper, we discuss whether and under which conditions personalized learning can indeed increase equity in K-12 education by bringing together empirical and theoretical insights from different fields, including the learning sciences, philosophy, psychology, and sociology. We distinguish between different conceptions of equity and argue that personalized learning is unlikely to result in "equality of outcomes" and, by definition, does not provide "equality of inputs". However, if implemented in a high-quality way, personalized learning is in line with "adequacy" notions of equity, which aim to equip all students with the basic competencies to participate in society as active members and to live meaningful lives.

Differential Biases and Variabilities of Deep Learning-Based Artificial Intelligence and Human Experts in Clinical Diagnosis: Retrospective Cohort and Survey Study

Background

Deep learning (DL)–based artificial intelligence may have different diagnostic characteristics than human experts in medical diagnosis. As a data-driven knowledge system, heterogeneous population incidence in the clinical world is considered to cause more bias to DL than clinicians. Conversely, by experiencing limited numbers of cases, human experts may exhibit large interindividual variability. Thus, understanding how the 2 groups classify given data differently is an essential step for the cooperative usage of DL in clinical application.

Objective

This study aimed to evaluate and compare the differential effects of clinical experience in otoendoscopic image diagnosis in both computers and physicians exemplified by the class imbalance problem and guide clinicians when utilizing decision support systems.

Methods

We used digital otoendoscopic images of patients who visited the outpatient clinic in the Department of Otorhinolaryngology at Severance Hospital, Seoul, South Korea, from January 2013 to June 2019, for a total of 22,707 otoendoscopic images. We excluded similar images, and 7500 otoendoscopic images were selected for labeling. We built a DL-based image classification model to classify the given image into 6 disease categories. Two test sets of 300 images were populated: balanced and imbalanced test sets. We included 14 clinicians (otolaryngologists and nonotolaryngology specialists including general practitioners) and 13 DL-based models. We used accuracy (overall and per-class) and kappa statistics to compare the results of individual physicians and the ML models.

Results

Our ML models had consistently high accuracies (balanced test set: mean 77.14%, SD 1.83%; imbalanced test set: mean 82.03%, SD 3.06%), equivalent to those of otolaryngologists (balanced: mean 71.17%, SD 3.37%; imbalanced: mean 72.84%, SD 6.41%) and far better than those of nonotolaryngologists (balanced: mean 45.63%, SD 7.89%; imbalanced: mean 44.08%, SD 15.83%). However, ML models suffered from class imbalance problems (balanced test set: mean 77.14%, SD 1.83%; imbalanced test set: mean 82.03%, SD 3.06%). This was mitigated by data augmentation, particularly for low incidence classes, but rare disease classes still had low per-class accuracies. Human physicians, despite being less affected by prevalence, showed high interphysician variability (ML models: kappa=0.83, SD 0.02; otolaryngologists: kappa=0.60, SD 0.07).

Conclusions

Even though ML models deliver excellent performance in classifying ear disease, physicians and ML models have their own strengths. ML models have consistent and high accuracy while considering only the given image and show bias toward prevalence, whereas human physicians have varying performance but do not show bias toward prevalence and may also consider extra information that is not images. To deliver the best patient care in the shortage of otolaryngologists, our ML model can serve a cooperative role for clinicians with diverse expertise, as long as it is kept in mind that models consider only images and could be biased toward prevalent diseases even after data augmentation.

The Potential of Naturalistic Eye Movement Tasks in the Diagnosis of Alzheimer's Disease: A Review

Extensive research has demonstrated that eye-tracking tasks can effectively indicate cognitive impairment. For example, lab-based eye-tracking tasks, such as the antisaccade task, have robustly distinguished between people with Alzheimer’s disease (AD) and healthy older adults. Due to the neurodegeneration associated with AD, people with AD often display extended saccade latencies and increased error rates on eye-tracking tasks. Although the effectiveness of using eye tracking to identify cognitive impairment appears promising, research considering the utility of eye tracking during naturalistic tasks, such as reading, in identifying cognitive impairment is limited. The current review identified 39 articles assessing eye-tracking distinctions between people with AD, mild cognitive impairment (MCI), and healthy controls when completing naturalistic task (reading, real-life simulations, static image search) or a goal-directed task involving naturalistic stimuli. The results revealed that naturalistic tasks show promising biomarkers and distinctions between healthy older adults and AD participants, and therefore show potential to be used for diagnostic and monitoring purposes. However, only twelve articles included MCI participants and assessed the sensitivity of measures to detect cognitive impairment in preclinical stages. In addition, the review revealed inconsistencies within the literature, particularly when assessing reading tasks. We urge researchers to expand on the current literature in this area and strive to assess the robustness and sensitivity of eye-tracking measures in both AD and MCI populations on naturalistic tasks.

Learning With a Double-Edged Sword? Beneficial and Detrimental Effects of Learning Tests-Taking a First Look at Linkages Among Tests, Later Learning Outcomes, Stress Perceptions, and Intelligence

It has often been shown that tests as intentionally hindered and difficult learning tasks increase long-term learning compared to easier tasks. Previous work additionally indicated that higher intelligence might serve as a prerequisite for such beneficial effects of tests. Nevertheless, despite their long-term learning effects, tests were also found to be evaluated as more negative and to lead to more stress and anxiety compared to easier control tasks. Stress and anxiety, in turn, often yield detrimental effects on learning outcomes. Hence, we hypothesized that tests increase later learning outcomes but simultaneously also lead to more stress perceptions. Such increased stress was, in turn, hypothesized to reduce later learning outcomes (thus, stress might serve as a mediator of the beneficial effects of tests on learning). All these assumed effects should further be moderated by intelligence, insofar as that higher intelligence should increase beneficial effects of tests on learning, should decrease stress perceptions caused by tests, and should reduce detrimental effects of stress on learning outcomes. Higher intelligence was also assumed to be generally associated with higher learning. We conducted a laboratory study (N=89) to test these hypotheses: Participants underwent an intelligence screening, then worked on either a test or a re-reading control task, and reported their immediate stress perceptions. Later learning outcomes were assessed after 1week. The results supported all assumed main effects but none of the assumed interactions. Thus, participants using tests had higher long-term learning outcomes compared to participants using re-reading tasks. However, participants using tests also perceived more immediate stress compared to participants that only re-read the materials. These stress perceptions in turn diminished the beneficial effects of tests. Stress was also generally related to lower learning, whereas higher intelligence was linked to higher learning and also to lower stress. Hence, our findings again support the often assumed benefits of tests-even when simultaneously considering learners' intelligence and and when considering the by tests caused stress perceptions. Notably, controlling for stress further increases these long-term learning benefits. We then discuss some limitations and boundaries of our work as well as ideas for future studies.

Information needs for recovery after colorectal surgery: a patient focus group study

AIM

The provision of information to patients is an important part of recovery after colorectal surgery. This study aimed to define patient information needs, barriers to effective understanding and insights into how information provision may be improved.

METHOD

A patient focus group was convened. This comprised a broad, convenience sample of 11 participants from across the United Kingdom with experience of major colorectal surgery. A semistructured topic guide was used to facilitate discussion about previous experiences of information provision and how this may be improved. Data were analysed thematically and are presented as major themes.

RESULTS

Overall, participants felt that their information needs are poorly prioritized by healthcare professionals. Barriers to understanding and retaining information include highly emotional situations (such as receiving bad news) and inappropriate information design (such as the use of inaccessible language). Participants expressed how information resources should: (a) address patients' individual information needs; (b) empower patients to take an active role in their recovery; (c) support patients with meaningful education and sign-posted resources; and (d) recognize patients' heightened need for information during recovery at home.

CONCLUSION

This study provides key insights into the information needs of patients undergoing colorectal surgery. These should inform the development of future information resources, whose format, timing and design are currently supported by low-quality evidence.

Assessing High Performers in the Life Sciences: Characteristics of Exams Used at the International Biology Olympiad (IBO) and Their Implications for Life Science Education

For decades, studies have revealed students' decreasing interest in science. Extracurricular learning opportunities-the Science Olympiads being a publicly well-known example-are an important means identified to tackle this challenge and help students further differentiate their interests. Better understanding the underlying constructs and characteristics of Science Olympiad exams can provide several implications not just for Science Olympiads, but also science education more broadly, for example, with regard to how the competitions' international juries defines expectations for high performance in the life sciences. This study analyzes exams set by the International Biology Olympiad (IBO) as an example for a top-tier international competition in the life sciences. The findings extend previous works on test item characteristics toward student competitions and high-performer education. We conducted a systematic analysis of N = 703 closed-ended and laboratory test items from six IBO assessment years across the competition's history. A categorical framework was developed to analyze items according to four areas: formal characteristics, content and practices, cognitive aspects, and the use of representations. Our findings highlight assessment characteristics used to challenge high-performing students. We derive implications for general life sciences education, as well as for further developing the assessments of Science Olympiads.

A Fundamental Cognitive Taxonomy for Cognition Aids

OBJECTIVE

This study aimed to organize the literature on cognitive aids to allow comparison of findings across studies and link the applied work of aid development to psychological constructs and theories of cognition.

BACKGROUND

Numerous taxonomies have been developed, all of which label cognitive aids via their surface characteristics. This complicates integration of the literature, as a type of aid, such as a checklist, can provide many different forms of support (cf. prospective memory for steps and decision support for alternative diagnoses).

METHOD

In this synthesis of the literature, we address the disparate findings and organize them at their most basic level: Which cognitive processes does the aid need to support? Which processes do they support? Such processes include attention, perception, decision making, memory, and declarative knowledge.

RESULTS

Cognitive aids can be classified into the processes they support. Some studies focused on how an aid supports the cognitive processes demanded by the task (aid function). Other studies focused on supporting the processes needed to utilize the aid (aid usability).

CONCLUSION

Classifying cognitive aids according to the processes they support allows comparison across studies in the literature and a formalized way of planning the design of new cognitive aids. Once the literature is organized, theory-based guidelines and applied examples can be used by cognitive aid researchers and designers.

APPLICATION

Aids can be designed according to the cognitive processes they need to support. Designers can be clear about their focus, either examining how to support specific cognitive processes or improving the usability of the aid.

Classification of Visual and Non-visual Learners Using Electroencephalographic Alpha and Gamma Activities

This study analyzes the learning styles of subjects based on their electroencephalo-graphy (EEG) signals. The goal is to identify how the EEG features of a visual learner differ from those of a non-visual learner. The idea is to measure the students' EEGs during the resting states (eyes open and eyes closed conditions) and when performing learning tasks. For this purpose, 34 healthy subjects are recruited. The subjects have no background knowledge of the animated learning content. The subjects are shown the animated learning content in a video format. The experiment consists of two sessions and each session comprises two parts: (1) Learning task: the subjects are shown the animated learning content for an 8-10 min duration. (2) Memory retrieval task The EEG signals are measured during the leaning task and memory retrieval task in two sessions. The retention time for the first session was 30 min, and 2 months for the second session. The analysis is performed for the EEG measured during the memory retrieval tasks. The study characterizes and differentiates the visual learners from the non-visual learners considering the extracted EEG features, such as the power spectral density (PSD), power spectral entropy (PSE), and discrete wavelet transform (DWT). The PSD and DWT features are analyzed. The EEG PSD and DWT features are computed for the recorded EEG in the alpha and gamma frequency bands over 128 scalp sites. The alpha and gamma frequency band for frontal, occipital, and parietal regions are analyzed as these regions are activated during learning. The extracted PSD and DWT features are then reduced to 8 and 15 optimum features using principal component analysis (PCA). The optimum features are then used as an input to the k-nearest neighbor (k-NN) classifier using the Mahalanobis distance metric, with 10-fold cross validation and support vector machine (SVM) classifier using linear kernel, with 10-fold cross validation. The classification results showed 97% and 94% accuracies rate for the first session and 96% and 93% accuracies for the second session in the alpha and gamma bands for the visual learners and non-visual learners, respectively, for k-NN classifier for PSD features and 68% and 100% accuracies rate for first session and 100% accuracies rate for second session for DWT features using k-NN classifier for the second session in the alpha and gamma band. For PSD features 97% and 96% accuracies rate for the first session, 100% and 95% accuracies rate for second session using SVM classifier and 79% and 82% accuracy for first session and 56% and 74% accuracy for second session for DWT features using SVM classifier. The results showed that the PSDs in the alpha and gamma bands represent distinct and stable EEG signatures for visual learners and non-visual learners during the retrieval of the learned contents.