On 25th of April the first in this year Open Scientific Seminar of the Big Data Analytics and Artificial Intelligence Master Program was conducted. Students presented preliminary results of their master's and term papers. The seminar was broadcasted online and open for the Open Data Science Siberia community.
The first thesis in the domain of Physiology was presented under the topic “Separability of silent speech phonemes for English language”. The idea of Andrey Zubkov’s work is to research if the silent speech recognition could be improved from distinguishing limited set of commands to speech recognition similar to modern audible solutions. By the Andrey’s work it was made possible to create a corpora of silent phonemes for the future use in transforming signals from EEG sensors taken from deaf and dumb people into text.
Next two presentations were devoted to the geological aspect of Oil Production. The topics are “Automatic facies classification from well logs data using machine learnings methods” from Alexandra Luchkina and “Automatic correlation of well logs” from Evgeniy Kurochkin, both devoted to the application of machine learning technique in well logs recognition. The reporters showed us the general possibility to successfully application of pattern recognition to well log data.
After that, three thesis on different aspects of Natural Language Processing were presented. Leyuan Sheng described his speech enhancement technique under the topic “Text to speech synthesis using Generative Adversarial networks for speech enhancement”, which was verified by more than 200 people and obtained valuable improvement in voice generation from spectrograms. The next speaker, Olga Yakovenko, presented her topic “Convolutional Variational Autoencoders for Audio Feature Representation in Speech Recognition”. She found an approach that allows an audio information processing to consume 3 times less hard drive space without loss of quality for analysis. Juan Fernando Pinzon Correa reported on the research of “Sentiment Analysis in Social Media Texts (Spanish language)”, wh ere he collected a representable corpora of Spanish short texts and successfully applied semi-supervised approach for sentiment prediction.
Malysheva Anastasia’s research was in the field of fundamental science: “The development and research of the prediction methods for time series obtained by the combination of different patterns”. She described experiments with implementation of efficient Krichevsky algorithm for times series analysis.
All the topics were discussed with great interest and active questions from the online and offline audience.
The recorded video stream of Open Scientific Seminar you can found under the link: https://www.youtube.com/watch?v=T7Oa4hBR3g4.