Eimantas Ledinauskas

Eimantas Ledinauskas, Julius Ruseckas, Julius Marozas et al.

Manual travel pattern inference from visa page stamps is a time consuming activity and constitutes an important bottleneck in the efficiency of traveler inspection at border crossings. Despite efforts to digitize and record the border crossing information into databases, travel pattern inference from stamps will remain a problem until every country in the world is incorporated into such a unified system. This could take decades. We propose an automated document analysis system that processes scanned visa pages and automatically extracts the travel pattern from detected stamps. The system processes the page via the following pipeline: stamp detection in the visa page; general stamp country and entry/exit recognition; Schengen area stamp country and entry/exit recognition; Schengen area stamp date extraction. For each stage of the proposed pipeline we construct neural network models and train then on a mixture of real and synthetic data. We integrated Schengen area stamp detection and date, country, entry/exit recognition models together with a graphical user interface into a prototype of an automatic travel pattern extraction tool. We find that by combining simple neural network models into our proposed pipeline a useful tool can be created which can speed up the travel pattern extraction significantly.

Eimantas Ledinauskas, Julius Ruseckas, Alfonsas Juršėnas et al.

Computation using brain-inspired spiking neural networks (SNNs) with neuromorphic hardware may offer orders of magnitude higher energy efficiency compared to the current analog neural networks (ANNs). Unfortunately, training SNNs with the same number of layers as state of the art ANNs remains a challenge. To our knowledge the only method which is successful in this regard is supervised training of ANN and then converting it to SNN. In this work we directly train deep SNNs using backpropagation with surrogate gradient and find that due to implicitly recurrent nature of feed forward SNN's the exploding or vanishing gradient problem severely hinders their training. We show that this problem can be solved by tuning the surrogate gradient function. We also propose using batch normalization from ANN literature on input currents of SNN neurons. Using these improvements we show that is is possible to train SNN with ResNet50 architecture on CIFAR100 and Imagenette object recognition datasets. The trained SNN falls behind in accuracy compared to analogous ANN but requires several orders of magnitude less inference time steps (as low as 10) to reach good accuracy compared to SNNs obtained by conversion from ANN which require on the order of 1000 time steps.