Distributed creation of Machine learning agents for Blockchain analysis

Creating efficient deep neural networks involves repetitive manual optimization of the topology and the hyperparameters. This human intervention significantly inhibits the process.Neural Architecture Search (NAS) algorithms can effectively automate this work and achieve results that surpass the best human-designed models.This research proposes a novel blockchain network protocol that incentivises independent computing nodes to run NAS algorithms and compete in finding better neural network models for a particular task. If implemented, such network can be an autonomous and self-improving source of machine learning models, significantly lowering the cost and access to accurate Machine Learning solutions.

Brain Cells Phenotyping Via Unsupervised Machine Learning With Autoencoder and Clustering

We work with an image database of brain cells and aim to develop an unsupervised neural network, consisting of autoencoder and clustering layers, to classify whether the input image contains a cell or not. The cell types which we have tested the algorithm on, are microglia and inhibitory neurons. We feed image patches to an autoencoder with a small latent space and perform clustering on the compressed data representation from the latent space. We expect that the encodings of „positive“ and „negative“ samples would differ from each other and thus form clusters in space. The achieved results show a good seperation of the two data classes, with an F1 score reaching 0.997.

A novel method for skeletal age estimation based on cranial suture analysis

Due to the cranial suture ossification in conjunction with age, the degree of fusion between the neighboring skull plates has been used as a predictor for estimating age, based on skeletonized human remains. So far, this involved the use of subjective ranking techniques, which rely entirely on the human opinion for assessing the degree of fusion. Such methods have shown to require a lot of human resources and not produce sufficiently accurate results in comparison with other age estimation methods. This project manages to entirely automate the process of age estimation based on cranial suture analysis and based on the available sample, suggests a significant improvement in the error of produced age estimates. This is achieved with the help of computer algorithms and deep learning.