Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi

Hüseyin Ozkan

Ozgur Yilmaz

The present invention relates to a method for incremental learning of a classification model, where pre-defined weak incremental learners are distributed over the distinct regions in a set of partitionings of the input domain. The partitionings and regions are organized via a binary tree and they are allowed to vary in a data-driven way, i.e., in a way to minimize the classification error rate. Moreover, to test a given data point, a mixture of decisions is obtained through the models learned in the regions that this point falls in. Hence, naturally, in the cold start phase of the data stream, the simpler models belonging to the larger regions are favored and as more data get available, the invention automatically puts more weights on the more complex models.

Incremental learner via an adaptive mixture of weak learners distributed on a non-rigid binary tree

Recurrent neural networks are powerful tools for handling incomplete data problems in machine learning thanks to their significant generative capabilities. However, the computational demand for algorithms to work in real time applications requires specialized hardware and software solutions. We disclose a method for adding recurrent processing capabilities into a feedforward network without sacrificing much from computational efficiency. We assume a mixture model and generate samples of the last hidden layer according to the class decisions of the output layer, modify the hidden layer activity using the samples, and propagate to lower layers. For an incomplete data problem, the iterative procedure emulates feedforward-feedback loop, filling-in the missing hidden layer activity with meaningful representations.

Growing community of inventors

Ankara, Turkey

Hüseyin Ozkan