Master thesis project at SYNTRONIC
Syntronic AB is a global consultant company approaching 700 employees. The company is based in Gävle, with offices in Göteborg, Kista, Linköping, Lund, Sandviken, Beijing, Chengdu, Penang, Jakarta, Ottawa and Kuala Lumpur. Our main expertise lies within embedded software and hardware, i.e., electronics with intelligent software. In Göteborg we are building a strong group of algorithm developers with focus in machine learning, sensor fusion, computer vision and automatic control. We are currently 9 developers/consultants within algorithms and 3 of us have PhD in mathematics or physics.
We seek a number of talented master students who want to write their master thesis projects at our office in Göteborg. Students should have a strong mathematical background as well as a strong interest in technology. The successful student will have good chances to become an employee at Syntronic after a finished thesis. Therefore, programming skills are highly valued.
Some suggestions to be made precise together with the student(s):
Reinforcement learning for production planning or test. Reinforcement learning has been used recently to play Atari 2600 games, to beat the world champion Lee Sedol in the game of go and learn cars to drive end-to-end based on raw camera data. Here we are interested in reinforcement learning for production planning or automatic software test. Other applications can be discussed.
Deep learning solvers for stochastic automatic control: Recently, deep learning methods for high dimensional PDE has been introduced. These use a well-known relation between the PDE and a related backward stochastic differential equation. The project concerns applying this methodology to the Hamilton-Jacobi-Bellman PDE from stochastic optimal control to generate control signals to some robotics problem.
Speaker identification: This project concerns machine learning algorithms for speech, or more precisely speaker identification. Focus is on real time algorithms and possibly on recurrent neural networks.
Spectrograms for speech and audio processing: In basic audio analysis spectrograms, or short time Fourier transforms are often used. For human speech, a non-linear frequency scale, the mel-scale is better used as the human ear is more sensitive to lower frequencies. Related mel-spectrogram and mel-cepstrograms are therefore often used. These are very simple. More sophisticated alternatives are wavelets and shirplets and this project concerns implementation of the latter and comparisons with wavelets and mel-spectrograms for some application(s) to be decided.
Apply by sending CV/letter/transcripts to Adam Andersson: email@example.com.