I am a senior researcher at the Swiss AI Lab IDSIA
(Istituto Dalle Molle di studi sull'intelligenza artificiale), with
Jürgen Schmidhuber. I work in machine learning
(currently mostly neural network-based).
Istituto Dalle Molle di studi sull'intelligenza artificiale
Galleria 2, Via Cantonale 2c
CH-6928 Manno, Switzerland
E-Mail: michael "att" idsia.ch
More access and contact information.
- Since January 2019: Senior Researcher at IDSIA.
- Since June 2014: Researcher at IDSIA.
- January 2014: PhD degree (Dr.-Ing.) at Cognitive Systems Lab,
Karlsruhe Institute of Technology, with a thesis titled "Advancing Electromyographic Continuous Speech
Recognition: Signal Preprocessing and Modeling". An online copy of my PhD thesis is
found at the German National Library (direct PDF link).
- 2008 - 2014: PhD studies in the field of Silent
Speech Interfaces based on Electromyography (EMG) at Cognitive Systems Lab.
- 2001 - 2007: Studies of Mathematics and Computer Science at Karlsruhe University (now Karlsruhe Institute of Technology),
leading towards a diploma degree (roughly equivalent to Master's) in mathematics (thesis field: applied analysis).
- 2006: Visiting student researcher at Carnegie Mellon University,
Pittsburgh, PA, USA,funded by a state scholarship of the state of Baden-Württemberg.
- 2004 - 2005: Exchange student at Queen's University, Kingston, Ontario, Canada, as a member of
the Ontario-Baden-Württemberg exchange program.
- 2000: Finished Gymnasium (German High School), with a school prize for the best graduation result of the year.
See also the Research page for information about my ongoing work.
A large part of my research has been dedicated to the application of speech recognition methods to
Speech, where an acoustic signal is not available (e.g. for speech-impaired persons) or undesired
(e.g. for confidential communication in the public). A method of choice is Electromyography (EMG)
where electrical signals from the facial muscles are captured with surface electrodes. The resulting signal is fed into a suitably adapted speech
recognizer (EMG-to-text), or alternatively, a speech synthesizer (EMG-to-speech). I also work on Lipreading
, where speech is recognized from images
of a person's face. Here the goal is to leverage the power of contemporary image processing towards this exciting task.
I am interested in all kinds of applications of machine learning to analysis of biophysiological data:
A field which is both rewarding, and offers a large variety of challenges.
From 2014 to 2016, I was a Marie Curie fellow in the Prototouch
project, where I worked
on machine learning for diverse kinds of data related to haptics and haptic/tactile devices. I
continue this line of research within the INPUT
project, where an excellent European Consortium
works on making the control of upper limb prostheses versatile and straightforward. The method of choice is the electromyographic signal,
captured on the arm stump, processed by a powerful neural network based machine learning system.
I am interested in the properties of neural networks, particularly with respect to the data representation in trained networks, and with the
training process itself. This bridges the gap between my interests in applied machine learning and the theoretical foundations of our
field. (Photo credit: "A pampered culture" by Parthiv Haldipur on Flickr
CC BY-NC 2.0 license
The IDSIA is located in beautiful Ticino
, located at the southern border of Switzerland and
featuring a unique blend of a moutain region and warm, mediterranean summers (apparently, summer starts in march and end in october). Hiking is
one of my favorite pastimes, and I have collected a few photos in this area