Journal: Sensors (Basel, Switzerland)

Article Title: Application of a Laser Profile Sensor for the Full-Field Measurement of the Continuous Icing Process of Rotating Blades

doi: 10.3390/s24144480

Figure Lengend Snippet: Bottom view of the rotor blades mid-part alpha shape, illustrating the effect of an offset reference profile: The facets on the concave leading edge, as well as the edges themselves, are much more detailed, when applying an appropriate offset ( b ).

Article Snippet: Matlab provides an in-built function to compute an alpha shape object from the point cloud data.

Techniques:

Journal: Nature

Article Title: Basis functions for complex social decisions in dorsomedial frontal cortex

doi: 10.1038/s41586-025-08705-9

Figure Lengend Snippet: (a,b) Primary basis functions exert their influence on choice earlier than secondary basis functions . Using a model-free approach, we replicated our finding that primary basis functions exert their influence on choice earlier in the decision process compared to secondary basis functions, for both Self and Partner decisions. See the drift diffusion modelling results in the main text for model-based evidence for the same point (see also ). Again, the analyses are based on the two types of dyadic decisions, as only here decisions can be made as a combination of primary and secondary basis function (both projections already “inverted” to agent-centric, social space for the analysis here). We tested the following prediction: If primary basis functions exert their influence earlier in the decision process compared to secondary basis functions, then faster decisions should be relatively more driven by primary compared to secondary basis functions. By contrast, slower decisions allow the secondary basis that contribute to decisions later in time to increase in influence on choice. This meant we predicted an interaction effect with the predominance of primary over secondary basis function effects being more pronounced in trials with fast reaction times compared to trials with slow reaction times. We found precisely such an basis function x reaction time interaction in a 2 x 2 [basis function projection: primary/secondary; reaction time: fast/slow] within-participant analysis of variance for both Self decisions (F 1,55 = 16.991; p < 0.001) and Partner decisions (F 1,55 = 24.363; p < 0.001). The GLM underlying this analysis was constructed as follows. First, we split decisions by their median reaction time in trials with fast and slow reaction times. Then, we predicted choice as a linear combination of the primary and the secondary basis projection (translated to agent-centric space), and the bonus. To prevent overfitting given the reduced trial number for each GLM, we used L2 regularization. To do so, we used Matlab’s lassoglm function (setting Alpha to Matlab’s smallest possible value) and used a lambda of 0.015 , thereby ensuring constrained decision weights. (n = 56; repeated measures ANOVA was used; boxplot middle line shows median, box is -+1.58 interquartile range [Tukey boxplot] and points are outliers relative to that). (c) Model comparison of behavioural basis function and agent model . Our results showed that irrelevant players in dyadic decisions impact choice in a manner consistent with their group membership. For example, during Self decisions, P and Oi are irrelevant. Nevertheless, our GLM analyses showed that P-performance had a significantly positive effect during Self decisions, and Oi-performance had a significantly negative one. This pattern of effects is predicted by our basis function model which suggests that participants first consider a primary basis function and then a secondary one. In support of our approach, we conducted an additional model comparison. We compared two models. One model was the “basis function model” which predicted choices as a function of the performances of S, P, Or, Oi and the non-social bonus (same as in main manuscript). The second one was an “agent model” that predicted choices only as a function of the relevant players and the bonus (e.g., S, Or and bonus for Self decisions). The agent model captured the intuition that decisions are made only based on the relevant players in a decision. The basis function model does not negate this. However, it suggests that there is a necessary addition to the agent model that is needed to explain choice better. This addition is reflected in the weights of influence of the irrelevant players, which are predicted by the basis function model. We compared the two logistic GLM models using a cross-validation approach. We collapsed the data over both Self and Partner decisions. Then we used a leave-one-participant-out cross-validation. Cross-validation has the advantage that model performance can be directly compared across models without setting somewhat arbitrary penalisation criteria for the number of parameters in the model (like AIC or BIC). Overfitting is directly punished by worse generalisation performance in the held-out sample. We implemented the cross-validation in the following way. We gathered the data from all participants except a left-out participant. We ran the model GLM on the concatenated data of the participants and then applied the resulting decision weights to the left-out participants data. We predicted the choices of the left-out participants on the basis of these weights that had been determined without the influence of the left-out participant’s data. We then calculated the model deviance and model accuracy for the left-out participant . The model deviance is a simple function of the difference between actual choices and the models’ choice probabilities. To derive an estimate of the model accuracy, we sampled 10.000 times from the predicted choice probability distribution to get model simulated binary choices. We calculated model accuracy by comparing these simulated choices with the actual choices and averaged the result to a single accuracy value. We repeated this procedure for all participants and aggregated the model deviances and model accuracies for both models. A smaller model deviance indicates a better model fit and we found that the basis function model indeed had significantly smaller deviance (paired t-test; t 55 = −2.981, p = 0.004). The result from the estimated model accuracies also indicated that the basis function model significantly predicted a higher percentage of participants’ decisions (paired t-test; t 55 = 8.111, p < 0.001, Cohen’s d = 1.084). A histogram of the difference in estimated model accuracies is shown in this figure. Note that the distribution is significantly shifted to the right, indicating better model fit for the basis function compared to the agent model. (n = 56; *, ***, p < 0.001 error bars are S.E.M; tests are ANOVAs and two-sided t-tests).

Article Snippet: To do so, we used Matlab’s lassoglm function (setting Alpha to Matlab’s smallest possible value) and used a lambda of 0.015 , thereby ensuring constrained decision weights. (n = 56; repeated measures ANOVA was used; boxplot middle line shows median, box is -+1.58 interquartile range [Tukey boxplot] and points are outliers relative to that). (c) Model comparison of behavioural basis function and agent model . Our results showed that irrelevant players in dyadic decisions impact choice in a manner consistent with their group membership.

Techniques: Diffusion-based Assay, Construct, Comparison, Biomarker Discovery