A team led by University of California Santa Barbara and UMBC deploys convolutional neural networks on one-second segments of pupil diameter and gaze data to accurately detect stimulus onsets, revealing generalization and task-specific patterns in cognitive event recognition with Matthews correlation coefficients up to 0.75.

Key points

• Five CNN models—including four task-specific and one generalized—process 1 s of 250 Hz pupil diameter and gaze data to detect stimulus onsets.
• SMOTE oversampling rebalances training data for unbiased binary classification, achieving MCC scores from 0.43 to 0.75 across tasks.
• Permutation feature importance shows task-specific models focus on gaze and pupillary light reflex, while the generalized model balances pupil dilation and gaze contributions.

Why it matters: This method enables rapid, individualized detection of cognitive events via ML-driven pupillometry for real-time attention and workload monitoring.

Scientists from Nature’s Scientific Reports harness 1D-CNN, ANN, RF, AdaBoost, EL and LSSVM—optimized via Coupled Simulated Annealing—to model pH shifts in LB and M63 media across three bacterial strains, delivering precise data-driven forecasts and reducing experimental overhead.

Key points

• 1D-CNN delivers R² = 0.9983, RMSE = 0.0519, MAPE = 0.37 across 379 data points.
• Coupled Simulated Annealing hyperparameter tuning optimizes all AI models for maximal predictive performance.
• Monte Carlo sensitivity highlights OD600 and incubation time as primary drivers of pH variations.

Why it matters: AI-driven pH prediction accelerates microbiological process control by reducing reliance on laborious experimental assays.