An artificial neural network using multi-head intermolecular attention for predicting chemical reactivity of organic materials

Abstract

This is an Accepted Manuscript, which has been through the RSC Publishing peer review process and has been accepted for publication. Accepted manuscripts are published online shortly after acceptance. This version of the article will be replaced by the fully edited, formatted and proof read Advance Article as soon as this is available.

Attention towards chemistry agnostic and explainable battery lifetime prediction

“…The model then integrates its own previous predictions and known future conditions, such as the expected discharge current and the cycle number. These two inputs are temporally encoded to capture their positional relevance 65 , ensuring that the decoder is informed of the predefined condition and the timing of each data point within the life cycle. The decoder employs an attention mechanism that can dynamically adjust sequence weights, identifying critical information at each prediction step.…”

“…2b. Here, the model processes the temporal data using a sliding window approach that enhances the ability to discern local patterns within long input sequences 65 .…”

“…6, the softmax function is applied to these scaled attention scores, which originate from the interactions between the query and key matrices. This process results in the production of a context vector, which integrates information from different representational subspaces and allows the model to consider multiple aspects of historical data 65,60 .…”