Pre-trained models are commonly used in Continual Learning to initialize the model before training on the stream of non-stationary data. However, pre-training is rarely applied during Continual Learning. We investigate the characteristics of the Continual Pre-Training scenario, where a model is continually pre-trained on a stream of incoming data and only later fine-tuned to different downstream tasks. We introduce an evaluation protocol for Continual Pre-Training which monitors forgetting against a Forgetting Control dataset not present in the continual stream. We disentangle the impact on forgetting of 3 main factors: the input modality (NLP, Vision), the architecture type (Transformer, ResNet) and the pre-training protocol (supervised, self-supervised). Moreover, we propose a Sample-Efficient Pre-training method (SEP) that speeds up the pre-training phase. We show that the pre-training protocol is the most important factor accounting for forgetting. Surprisingly, we discovered that self-supervised continual pre-training in both NLP and Vision is sufficient to mitigate forgetting without the use of any Continual Learning strategy. Other factors, like model depth, input modality and architecture type are not as crucial.

In this paper, we present an introduction and critical experimental evaluation of a reservoir computing (RC) approach for ambient assisted living (AAL) applications. Such an empirical analysis jointly addresses the issues of efficiency, by analyzing different system configurations toward the embedding into computationally constrained wireless sensor devices, and of efficacy, by analyzing the predictive performance on real-world applications. First, the approach is assessed on a validation scheme where training, validation and test data are sampled in homogeneous ambient conditions, i.e., from the same set of rooms. Then, it is introduced an external test set involving a new setting, i.e., a novel ambient, which was not available in the first phase of model training and validation. The specific test-bed considered in the paper allows us to investigate the capability of the RC approach to discriminate among user movement trajectories from received signal strength indicator sensor signals. This capability can be exploited in various AAL applications targeted at learning user indoor habits, such as in the proposed indoor movement forecasting task. Such a joint analysis of the efficiency/efficacy trade-off provides novel insight in the concrete successful exploitation of RC for AAL tasks and for their distributed implementation into wireless sensor networks.

Estimation of mortality risk of very preterm neonates is carried out in clinical and research settings. We aimed at elaborating a prediction tool using machine learning methods. We developed models on a cohort of 23747 neonates <30 weeks gestational age, or <1501 g birth weight, enrolled in the Italian Neonatal Network in 2008–2014 (development set), using 12 easily collected perinatal variables. We used a cohort from 2015–2016 (N = 5810) as a test set. Among several machine learning methods we chose artificial Neural Networks (NN). The resulting predictor was compared with logistic regression models. In the test cohort, NN had a slightly better discrimination than logistic regression (P < 0.002). The differences were greater in subgroups of neonates (at various gestational age or birth weight intervals, singletons). Using a cutoff of death probability of 0.5, logistic regression misclassified 67/5810 neonates (1.2 percent) more than NN. In conclusion our study – the largest published so far – shows that even in this very simplified scenario, using only limited information available up to 5 minutes after birth, a NN approach had a small but significant advantage over current approaches. The software implementing the predictor is made freely available to the community.