Underage Detection through a Multi-Task and MultiAge Approach for Screening Minors in Unconstrained Imagery

Accurate automatic screening of minors in unconstrained images demands models that are robust to distribution shift and resilient to the children under-representation in publicly available data. To overcome these issues, we propose a multi-task architecture with dedicated under/over-age discrimination tasks based on a frozen FaRL vision-language backbone joined with a compact two-layer MLP that shares features across one age-regression head and four binary under-age heads for age thresholds of 12, 15, 18, and 21 years, focusing on the legally critical age range. To address the severe class imbalance, we introduce an $\alpha$-reweighted focal-style loss and age-balanced mini-batch sampling, which equalizes twelve age bins during stochastic optimization. Further improvement is achieved with an age gap that removes edge cases from the loss. Moreover, we set a rigorous evaluation by proposing the Overall Under-Age Benchmark, with 303k cleaned training images and 110k test images, defining both the "ASORES-39k" restricted overall test, which removes the noisiest domains, and the age estimation wild shifts test "ASWIFT-20k" of 20k-images, stressing extreme pose ($>$45{\deg}), expression, and low image quality to emulate real-world shifts. Trained on the cleaned overall set with resampling and age gap, our multiage model "F" lowers the root-mean-square-error on the ASORES-39k restricted test from 5.733 (age-only baseline) to 5.656 years and lifts under-18 detection from F2 score of 0.801 to 0.857 at 1% false-adult rate. Under the domain shift to the wild data of ASWIFT-20k, the same configuration nearly sustains 0.99 recall while boosting F2 from 0.742 to 0.833 with respect to the age-only baseline, demonstrating strong generalization under distribution shift. For the under-12 and under-15 tasks, the respective boosts in F2 are from 0.666 to 0.955 and from 0.689 to 0.916, respectively.

Machine Learning for Screening Large Organic Molecules

Organic semiconductors are promising materials for cheap, scalable and sustainable electronics, light-emitting diodes and photovoltaics. For organic photovoltaic cells, it is a challenge to find compounds with suitable properties in the vast chemical compound space. For example, the ionization energy should fit to the optical spectrum of sun light, and the energy levels must allow efficient charge transport. Here, a machine-learning model is developed for rapidly and accurately estimating the HOMO and LUMO energies of a given molecular structure. It is build upon the SchNet model (Sch\"utt et al. (2018)) and augmented with a `Set2Set' readout module (Vinyals et al. (2016)). The Set2Set module has more expressive power than sum and average aggregation and is more suitable for the complex quantities under consideration. Most previous models have been trained and evaluated on rather small molecules. Therefore, the second contribution is extending the scope of machine-learning methods by adding also larger molecules from other sources and establishing a consistent train/validation/test split. As a third contribution, we make a multitask ansatz to resolve the problem of different sources coming at different levels of theory. All three contributions in conjunction bring the accuracy of the model close to chemical accuracy.