Embedded Files
In recent years, BlueStem has further developed foundational formulations in the fields of computational methods and artificial intelligence that were developed by our founder and Managing Director. These formulations have lead to the development of advanced artificial intelligence frameworks, sophisticated data analytics engines and high-accuracy time-series forecasting programs.
Artificial Intelligence
We have developed a novel paradigmatic revision of traditional neural networks using network theoretic methods and conformal geometric algebra. This world-first theoretical framework is called the ‘hyperfield cognition network’ (HCN). The HCN expands upon the mathematical foundations of neural networks in five dimensional conformal geometric algebra (visual abstract representations of HCN networks are presented in the figures to the right). This framework allows one to construct a novel theoretical computational engine which is similar to a standard artificial neural network but admits numerous added benefits:
Permits multiple training programs simultaneously.
Affords computational multiplicity in a single engine.
Reduces sensitivity to adversarial perturbations in training sets.
Affords broader capabilities and plasticity in the training of networks
Produces robust and streamlined ‘neural networks’.
Furthermore, in comparison to state of the art machine learning techniques the HCN:
Is comparable in terms of performance
Requires a training domain that is about 7.5 times smaller
Is consistent over a prediction domain that is 6 times larger than the training domain.
Uses fewer neurons by numerous factors.
Trains faster.
Some recently published research work is shown here: https://doi.org/10.1016/j.neucom.2019.12.116
Data Analysis
In 2019, BlueStem developed a stand-alone provenancing and classification software package (called the 'BlueStem Provenancing Classifier') which is composed of a novel featurisation technique, coupled with an advanced ensemble neural network-based classification program.
The BlueStem Provenancing Classifier, applied to numerous application theatres in the agriculture and aquaculture sectors, has been shown to be:
Highly effective at trace-back and provenancing exercises.
Outperforms state-of-the-art machine learning methods, such as XGBoost, nearest neighbour, support vector machines, random forest algorithms and multi-level perceptrons.
Is less sensitive to sparse and fuzzy datasets when compared to state of the art techniques.
In 2019, we developed formulations for the prediction of qualitative and quantitative data from spare and fuzzy geochemical datasets. This formulation is called 'Breadcrumbs' and the Breadcrumbs formulation was fortified with an AI-based formulation that allows us to generate predictive spatial maps of geological systems.
These predictive spatial maps:
Can allow for buried deposits to be discovered using very small and fuzzy data sets.
Can be used to accurately model and predict the concentration of a metal of interest within a potential deposit (see figure to the left).
Can greatly assist exploration and mine geologists in identifying the locations of hidden orebodies and predicting an orebody's morphology and metal concentration.
The Breadcrumbs formulation and associated spatial maps can be used by geochemical exploration and mining companies to discover currently undiscovered mineral deposits, allowing for greater profits associated with mining and potentially reducing the scarcity of critical minerals.
Some recently published research works are shown
here: https://doi.org/10.1007/s11004-020-09856-3
Advanced Computational Methods in Forecasting Multi-Variate Systems and Time Series
We have developed a highly accurate quantitative prediction engine which is capable of accurately predicting future states of multi-variate semi-stochastic systems that manifest as temporally evolving data streams (time series) in the physical and mathematical sciences. This engine has been applied to numerous industrial sectors for clients requiring accurate systems modelling.
Some recently published research work is shown here: https://doi.org/10.1007/s00006-020-01061-z
Page updated
Report abuse