One of the most crucial challenges faced by the Li-ion battery community concerns the search for the minimum time charging without damaging the cells. This goal can be achieved by solving a large-scale constrained optimal control problem, which relies on accurate electrochemical models. However, these models are limited by their high computational cost, as well as identifiability and observability issues. As an alternative, simple output-feedback algorithms can be employed, but their performance strictly depends on trial and error tuning. Moreover, particular techniques have to be adopted to handle safety constraints. With the aim of overcoming these limitations, we propose an optimal-charging procedure based on deep reinforcement learning. In particular, we focus on a policy gradient method to cope with continuous sets of states and actions. First, we assume full state measurements from the Doyle–Fuller–Newman (DFN) model, which is projected to a lower dimensional feature space via the principal component analysis. Subsequently, this assumption is removed, and only output measurements are considered as the agent observations. Finally, we show the adaptability of the proposed policy to changes in the environment’s parameters. The results are compared with other methodologies presented in the literature, such as the reference governor and the proportional–integral–derivative approach.

Typically, animal mitochondria have very compact genomes, with few short intergenic regions, and no introns. Hence, it may seem that there is little space for unknown functions in mitochondrial DNA (mtDNA). However, mtDNA can also operate through RNA interference, as small non coding RNAs (sncRNAs) produced by mtDNA have already been proposed for humans. We sequenced sncRNA libraries from isolated mitochondria of Ruditapes philippinarum (Mollusca Bivalvia) gonads, a species with doubly uniparental inheritance of mitochondria, and identified several putative sncRNAs of mitochondrial origin. Some sncRNAs are transcribed by intergenic regions that form stable stem-hairpin structures, which makes them good miRNA-like candidates. We decided to name them small mitochondrial highly-transcribed RNAs (smithRNAs). Many concurrent data support that we have recovered sncRNAs of mitochondrial origin that might be involved in gonad formation and able to affect nuclear gene expression. This possibility has been never suggested before. If mtDNA can affect nuclear gene expression through RNA interference, this opens a plethora of new possibilities for it to interact with the nucleus, and makes metazoan mtDNA a much more complex genome than previously thought.

Charge unbalancing in series-connected cells can lead to lower storage capacity and shorter battery life. Model-based optimization strategies have proven to be very effective in addressing this problem. In this article, we propose a general nonlinear model predictive control (NMPC) scheme for obtaining a balancing-aware optimal charging. The presented method relies on an electrochemical model, tailored for control purposes. In view of the possibility of practical implementation, the concepts are subsequently specialized for an easily implementable power supply scheme. Finally, the NMPC approach is validated on commercial cells using a detailed battery simulator, with sound evidence of its effectiveness both under the assumption of full state availability and in the presence of an observer scheme.

Several studies have linked mitochondrial genetic variation to phenotypic modifications; albeit the identity of the mitochondrial polymorphisms involved remains elusive. The search for these polymorphisms led to the discovery of small noncoding RNAs, which appear to be transcribed by the mitochondrial DNA (“small mitochondrial RNAs”). This contention is, however, controversial because the nuclear genome of most animals harbors mitochondrial pseudogenes (NUMTs) of identical sequence to regions of mtDNA, which could alternatively represent the source of these RNAs. To discern the likely contributions of the mitochondrial and nuclear genome to transcribing these small mitochondrial RNAs, we leverage data from six vertebrate species exhibiting markedly different levels of NUMT sequence. We explore whether abundances of small mitochondrial RNAs are associated with levels of NUMT sequence across species, or differences in tissue-specific mtDNA content within species. Evidence for the former would support the hypothesis these RNAs are primarily transcribed by NUMT sequence, whereas evidence for the latter would provide strong evidence for the counter hypothesis that these RNAs are transcribed directly by the mtDNA. No association exists between the abundance of small mitochondrial RNAs and NUMT levels across species. Moreover, a sizable proportion of transcripts map exclusively to the mtDNA sequence, even in species with highest NUMT levels. Conversely, tissue-specific abundances of small mitochondrial RNAs are strongly associated with the mtDNA content. These results support the hypothesis that small mitochondrial RNAs are primarily transcribed by the mitochondrial genome and that this capacity is conserved across Amniota and, most likely, across most metazoan lineages.

Knowledge distillation is recognized as a valuable model compression strategy that alleviates the computational burden of large language models while preserving performance. This strategy involves training a smaller model utilizing both real data and predictions from a more cumbersome model. Traditional distillation methods, however, are often compromised by exposure bias, which results from reliance on next-step prediction training loss. This bias emerges when models are tested in free-running mode, differing from their training regime and leading to a progressive drift in input distributions between testing and training phases. An analogous issue, known as ‘distributional shift’, has been effectively addressed in imitation learning through various methodologies. Therefore, this paper specifically tailors an imitation learning-based solution to a traditional knowledge distillation framework which inherently considers both real data and the teacher’s predictions as dual sources of expert demonstrations. The effectiveness of this approach is demonstrated over five different test datasets, where it outperforms traditional benchmarks across all evaluation metrics. Specifically, it achieves superior results in perplexity, multi-token generation, and G-Eval score, indicating improvements in both predictive accuracy and alignment with human judgment in text quality. These results underscore the potential of this approach to effectively address exposure bias in large language model distillation.