At Brainalyzed we have the vision to open up the possibilities of AI to every organization, by significantly lowering the barriers of entry for this fascinating technology. Selection of training data from heterogenous sources, feature engineering, hyperparameter-tuning, model deployment and maintenance have long been a major inhibitor for AI adoption.
With Brainalyzed INSIGHT we introduce the first distributed learning software for Artificial Swarm Intelligence (ASI) that gives banks an all-in-one solution for AI development to reduce cost and increase efficiency of portfolio decisions. The software is a self-serving platform that is driven by the combined mighty forces of multi-objective optimization and state-of-art deep learning technology to strike an unparalleled balance of AI fidelity and AI creation simplicity.
Optimization of inputs and network architecture
Traditional learning approaches require that model inputs for the training are fixed in advance. This method requires a hypothesis that a given data parameter has added value for the prediction. However, for complex systems, human intuition is a poor guide, resulting in reduced AI performance due to missing inputs or increased numerical noise from unnecessary data points. Brainalyzed Insight ensures that only the relevant inputs are used in the resulting AI swarm. Furthermore, within the training process, the network architecture is adapted to best fit the prediction problem as well as the data available. This effectively prevents the overfitting of the AI models in cases where only a small data set exists.
Artificial Swarm Intelligence
The actual prediction is based on a whole swarm of AIs that have proven to be optimal in terms of the selected performance parameters during optimization. A performance parameter is a metric that the user can define to distinguish a good model from a bad model. One has the option to define a variety of metrics to describe the prediction problem. For example, the minimization of false positive and false negative rates can be chosen as independent performance parameters for a classification. Each member of the AI Swarm is therefore unique in its combination of architecture, input and, of course, neuron weights. This improves prediction stability and diversification.
AI Swarm adjustment in real time
The operational AI Swarm initially consists of all AI models that are chosen by the user or best represent the selected performance parameters. However, the real time performance is not only monitored for these initial models, but also for the near-optimal models of the AI Academy. If the prediction performance of the AI Swarm deteriorates, poorly performing models of the initial swarm will be exchanged with those of the AI Academy.
Pre-built use cases for the financial industry
Brainalyzed Insight includes pre-built use cases for the finance industry, developed in collaboration with industry experts. In this way, we ensure that customers can immediately use the best practice for standard applications and start model training. The final AI Swarms enable the customer to reduce the time for manual data analysis, take decisions based on data, or run processes completely automated.