Edward Laurence

I'm an Ai research scientist at Hectiq.ai. Located in Québec.

A recent addition to PRX has surfaced! A nice collaboration with J.G. Young, L.H. Dufresne, and the Dynamica group on inferring how networks emerge. Check out the publication tab!


Netsci is a large conference with over 500 participants. A former member of my research group has been appointed the co-chair role and asked me to conceive the name tags.

Instead of wasting this opportunity in doing something pretty but harmless, we decided to design it to increase interactions between participants. To do so, I've been doing some NLP on the submitted abstracts, and scrapping some of arxiv publications to extract participants' research interests into latent vectors. Then, I've extracted the dominant clusters and computed the participants' affinity with each cluster.

Finally, I've designed a badge that shares the design direction adopted by the conference and that communicates the inferred research interests. The idea is each cluster is represented by a color and the top 5 clusters for a participant are indicated by the color of the five nodes, in descending order of affinity. For example, two participants with the first node of the same color have high probability of sharing scientific interest.

This new paper proposes a formalism to reduce large systems of differential equations coupled by graphs. We expose how the spectral properties of the graph can be leveraged to obtain an effective low-dimensional representation. Our work leads to a highly interpretable analysis of resilience of complex systems. PRX (abbreviation for Physical Review X) is an open-source journal from the APS society.


How can we detect an attack on a complex system by measuring the activity of its components? The Dynamica group has been retreated in Charlevoix to provide an answer to this prime importance question. The endeavor has been an indubitable success as we developed machine learning and Bayesian methods to pinpoint the disrupted components. A publication is soon to come!

I’m happy to announce that I have accepted a position as data scientist at Hectiq.ai. This company is based in Quebec city and is specialized in solving artificial intelligence problems. The exciting aspect of the job is that by working as a consultant, we have the change to explore many branches of machine learning. Depeding on the client needs, I'll be expecting to drill into Bayesian inference, graph theory, image segmentation, classification, just to name a few!

From the 8th to the 14th of July, I’ll be attending SIAM18 conference, in Portland (Oregon, USA) and present my latest project named A New Dimension-reduction Method for Complex Dynamical Networks. SIAM’s Annual Meeting provides a broad view of the state of the art in applied mathematics, computational science, and their applications.

During the semester of winter 2018, I’ve taken a deep learning course that covers fully connected and convolutional networks, state-of-the-art architectures and regulations mechanisms, recurrent networks, natural language processing, GAN, and miscellaneous topics. For the final assignment, we have chosen to construct our own dataset by scraping Unsplash website and to infer the EXIF metadata of the images (final paper and image below). We have learned so much... particularly that the learning can only be as good as the dataset.


We used deep convolutional networks to estimate the camera settings, lens aperture, ISO sensibility and exposure time, known as the EXIF metadata solely based on the pixels of the photos. The training has been performed on a novel dataset of 19 000 high quality photos labeled with the camera settings. The results indicated that deep convolutionnal networks have trouble solving the task and achieving high accuracy. We find that the general low performances we obtained are due to the dataset corruption, some class imbalances and possibly the lack of information in the picture alone.

Link to paper

I’ve created a Github repo to host the code for the fast networks generator. It basicaly reads real images (portrait, animal) and converts it to networks.