Multitask learning for denoising and analysis of X-ray polymer acquisitions

Emerging Trends in Machine Learning: A Polymer Perspective

“…Multitask learning has also been used to predict properties from images of nanocomposites using a convolutional neural network 112 and to simultaneously denoise and predict sample characteristics from X-ray hyperspectral images using an autoencoder. 113 Related to data fusion, transfer learning is a ML technique where information is transferred between tasks (e.g., predictions of glass transition temperature or melting temperature), domains (e.g., polymer literature or webpages) or both. Since information is often transferred from a data-rich task or domain, known as the source, to a data-poor task or domain, known as the target, it allows for improved predictions for the target for smaller data set sizes.…”

“…A graphical depiction of this scheme as applied to random copolymers is shown in Figure a. Multitask learning has also been used to predict properties from images of nanocomposites using a convolutional neural network and to simultaneously denoise and predict sample characteristics from X-ray hyperspectral images using an autoencoder …”

“…Another trend in ML in polymer science is the use of advanced deep learning techniques. Examples include recurrent neural networks (RNNs), which are designed to handle sequences such as the sequence in a copolymer, ,,,, variational autoencoders (VAEs), which are composed of an encoder and a decoder with a smaller latent space in between, , reinforcement learning (RL), where an agent takes an action and then receives a reward, generative adversarial networks (GAN), composed of a generator and a discriminator, and graph neural networks (GNNs), which are designed to handle graph based data such as a polymer structure. An example of RL is shown in Figure .…”