Abstract-An efficient algorithm for recurrent neural network training is presented. The approach increases the training speed for tasks where a length of the input sequence may vary significantly. The proposed approach is based on the optimal batch bucketing by input sequence length and data parallelization on multiple graphical processing units. The baseline training performance without sequence bucketing is compared with the proposed solution for a different number of buckets. An example is given for the online handwriting recognition task using an LSTM recurrent neural network. The evaluation is performed in terms of the wall clock time, number of epochs, and validation loss value.

No abstract

No abstract

International audienceHandshaking is an important component of social interaction between people in many cultures. Thus, for further applications in human/humanoid-robot interaction it is important to understand and measure the characteristics of a handshake during interaction between humans. In this paper, a new wearable sensor network to measure a handshake is described. It consists of a set of several sensors (accelerometers, gyroscopes and force sensors) attached to the glove, and of a microcontroller for signal acquisition and conditioning. The paper focuses on the applicability and qualitative analysis of the proposed architecture of sensors through several experiments of handshaking between two human subjects. The results show that the proposed system allows reproducible experiments to quantify handshake characteristics such as duration and strength of the grip, vigor and rhythmicity of a handshake

The authors analyze excitability of polyharmonic vibrations in a single-body vibration machine. The developed mathematical model of the vibration system accounts for an elastic component element included in the design of the unbalance vibration exciter drive. The operating limits, frequency content and effect of the main design factors on the flow data of the vibration machine are examined. It is found that superharmonic vibration largely contributes to the polyharmonic spectrum at certain frequencies, depending on the stiffness of the elastic component element of the clutch connecting the vibration exciter and the rotary drive.