No abstract

The field of chiral plasmonics has registered considerable progress with machine-learning (ML)-mediated metamaterial prototyping, drawing from the success of ML frameworks in other applications such as pattern and image recognition. Here, we present an end-to-end functional bidirectional deep-learning (DL) model for three-dimensional chiral metamaterial design and optimization. This ML model utilizes multitask joint learning features to recognize, generalize, and explore in detail the nontrivial relationship between the metamaterials’ geometry and their chiroptical response, eliminating the need for auxiliary networks or equivalent approaches to stabilize the physically relevant output. Our model efficiently realizes both forward and inverse retrieval tasks with great precision, offering a promising tool for iterative computational design tasks in complex physical systems. Finally, we explore the behavior of a sample ML-optimized structure in a practical application, assisting the sensing of biomolecular enantiomers. Other potential applications of our metastructure include photodetectors, polarization-resolved imaging, and circular dichroism (CD) spectroscopy, with our ML framework being applicable to a wider range of physical problems.

Renewable energy, as an environmentally friendly and sustainable source of energy, is key to realizing the nationally determined contributions of the United States (US) to the December 2015 Paris agreement. Policymakers in the US rely on energy forecasts to draft and implement cost-minimizing, efficient and realistic renewable and sustainable energy policies but the inaccuracies in past projections are considerably high. The inaccuracies and inconsistencies in forecasts are due to the numerous factors considered, massive assumptions and modeling flaws in the underlying model. Here, we propose and apply a machine learning forecasting algorithm devoid of massive independent variables and assumptions to model and forecast renewable energy consumption (REC) in the US. We employ the forecasting technique to make projections on REC from biomass (REC-BMs) and hydroelectric (HE-EC) sources for the 2009-2016 period. We find that, relative to reference case projections in Energy Information Administration's Annual Energy Outlook 2008, projections based on our proposed technique present an enormous improvement up to~138.26-fold on REC-BMs and 24.67-fold on HE-EC; and that applying our technique saves the US~2692.62 PJ petajoules (PJ) on HE-EC and~9695.09 PJ on REC-BMs for the 8-year forecast period. The achieved high-accuracy is also replicable to other regions.

No abstract

Emotion recognition is a topic of significant interest in assistive robotics due to the need to equip robots with the ability to comprehend human behavior, facilitating their effective interaction in our society. Consequently, efficient and dependable emotion recognition systems supporting optimal human-machine communication are required. Multi-modality (including speech, audio, text, images, and videos) is typically exploited in emotion recognition tasks. Much relevant research is based on merging multiple data modalities and training deep learning models utilizing low-level data representations. However, most existing emotion databases are not large (or complex) enough to allow machine learning approaches to learn detailed representations. This paper explores modalityspecific pre-trained transformer frameworks for self-supervised learning of speech and text representations for data-efficient emotion recognition while achieving state-of-the-art performance in recognizing emotions. This model applies feature-level fusion using nonverbal cue data points from motion capture to provide multimodal speech emotion recognition. The model was trained using the publicly available IEMOCAP dataset, achieving an overall accuracy of 77.58% for four emotions, outperforming state-of-the-art approaches