Dr. Michael Heck

Hello, I am Michael. I am a natural language processing researcher with a background in automatic speech recognition and currently focusing on dialogue systems.

I am a member of the Dialog Systems and Machine Learning Group at Heinrich Heine University Düsseldorf, Germany, where I work with Milica Gašić. My current research interests revolve around representation learning in the context of dialogue systems and dialogue state tracking.

I graduated with a Doctor of Engineering (D.Eng.) from Nara Institute of Science and Technology (NAIST), Japan in 2018. My doctoral course research interests were acoustic modeling for automatic speech processing, unsupervised learning, representation learning, and other machine learning challenges. For my doctoral thesis, I conducted zero resource speech processing research where I used Bayesian nonparametric methods to tackle the task of unsupervised subword modeling and speech representation learning. My supervisors were Satoshi Nakamura and Sakriani Sakti. From 2018 to 2019 I was Technical Staff in the RIKEN Center for Advanced Intelligence Project (AIP), Tourism Information Analytics (TIA) Team, Japan.

Apart from my current research focus and my doctoral course interests I am experienced in building large scale automatic speech recognition systems, automatic language identification, automatic speech segmentation, among other speech processing related topics.

I graduated from Karlsruhe Institute of Technology (KIT) with a diploma in Informatics (Dipl.-Inform.) after writing my thesis during a scholarship-supported internship period at NAIST. At KIT I worked with Alex Waibel and Sebastian Stüker.

Updates

December 2024 - Acceptance of our paper "Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling" at AAAI'25 (pre-print)

Selected Talks

ACL2023_heck_v1.mp4

ACL 2023

ChatGPT for Zero-shot Dialogue State Tracking:
A Solution or an Opportunity?

chatgpt2023.pdf

HeiCAD Special Event

Explaining the Sensation:
An Accessible Introduction to ChatGPT

TowardsAdaptiveTOD.pdf

INVITED TALK 2021

Towards Adaptive and Evolving Task-oriented Dialogue Systems

COLING2020.mp4

COLING 2020

Out-of-Task Training for Dialog State Tracking Models

SIGDIAL2020.mp4

SIGDIAL 2020

TripPy: A Triple Copy Strategy for Value Independent Neural Dialog State Tracking

 

Get in touch with me at michael [dot] heck [at] hhu [dot] de

Universitätsstr. 1 Building 25.12, Düsseldorf, Germany 40225