Research @ HEC Liege

Research @ HEC Liege

Business Analytics and Supply Chain Management



Groundbreaking NLP Results by J. Poumay & A. Ittoo

Judicael Poumay's and Prof. Ashwin Ittoo's latest research has yielded groundbreaking results in the representation of meaning as vectors and algebraic operations performed over these vectors (word embeddings).

Their study has been accepted for publication by “Empirical Methods in Natural Language Processing (EMNLP)”, a high impact outlet for machine/deep learning and NLP research. EMNLP is ranked 2nd among all publication outlets in that field according to Google Scholar, with a h5-index of 132. It has an acceptance rate of 10-15%, and competition was intense (against the likes of Stanford U, Facebook AI Research,...)

In their work, J. Poumay and A. Ittoo investigated the performance of different embeddings on the task of event coreference resolution (i.e. linking similar events together).  They discovered high degree of diminishing returns -  the performance of the smallest model (in terms of parameters) was on par with that of larger ones. They also found model size and run-time to be weakly correlated, with larger models converging faster than smaller ones.  These results are expected to have significant applications in the theoretical study of embeddings as well as in their deployment in practice. A pre-print is available via this link. It will also be available via ORBI soon.

Judicael Poumay is a PhD candidate and researcher at HEC Liège working on weak signal detection, under the supervision of Professor Ashwin Ittoo. The authors acknowledge the funding received from KPMG via the HEC Digital Lab for their research.

If you are interested in learning more about this research, please contact judicael.poumay@uliege /


Using VR audiences to improve presentation skills

Speaking out in public or during meetings is a challenge for many of us. It is nonetheless a competency required in many activities: the sales representative who presents a product to customers, the tourist guide visiting a city with a group, the manager who defends his project in front of stakeholders, the candidate during a job interview, the professor in front of students… Unfortunately, many firms complain about the very low level of this skill within their staff. To fight public speaking anxiety, which may impede oral presentation performances, speakers need to be prepared and trained. Our PhD candidate Elodie Etienne and Prof. Schyns decided to explore how they can optimize business processes thanks to machine learning and virtual reality in a 3-step work.

First, along with Dr. Angélique Remacle, and Prof. Leclercq, they tried to validate the emotional valence and level of arousal of the attitudes of a virtual public through a statistical analysis. Based on these results, a library of public attitudes corresponding to different levels of arousal and valence has been created. Their experiment also investigates the benefits of using low-end and high-end VR headsets, as well as the use of photo-realistic audiences compared to sketched avatars.  The next step is the analysis of voice and speech (voices indexes, detection of jokes or feelings, etc) using Statistics, NLP, ML, and DP approaches. And finally, they will use the library during the training of the speakers in front of an interactive audience reacting in an autonomous way (machine learning, deep learning, ...) to the speaker's presentation.

These are some of the many applications of machine learning and virtual reality for the optimization of business processes that Elodie and Prof. Schyns are exploring in their research.

If you are interested in learning more about this research, please contact Elodie Etienne.