Machine learning projects

Deep generative modeling and NLP for finance @ Sony Computer Science Labs (Tokyo, 2022)

I developed a method to compute time-dependent loadings in security factor models based on the optical-flow algorithm and trained a VAE to learn and visualize low-dimensional fund-month exposure embeddings, to support GPIF activities. I also researched NLP algorithms (including approaches using transformer-based language models such as BERT and MPNet) for various problems, such as fake news detection and greenwashing evaluation, to help Nomura Asset management assess corporate sustainability.

Cine-MR image forecasting with RNNs and transformers for latency compensation in radiotherapy @ The University of Tokyo (2017-2021, independent work until 2026)

I designed an algorithm to forecast future frames in dynamic chest and liver MR images, based on motion modeling with PCA and temporal dynamics prediction with RNNs trained online and transformer encoders, for safe MR-guided radiotherapy. The proposed approach is modular, interpretable, and can help overcome data scarcity, motion irregularity, and distributional shift between the training and test datasets for population models.

External marker position forecasting using RNNs trained online for safer radiotherapy @ The University of Tokyo (2020-2021, independent work until 2025)

I proposed implementations of online training algorithms for RNNs (UORO, SnAp-1, and DNI) with closed-form simplifications as well as improved accuracy and reduced time/memory complexity. I assessed their ability to forecast the positions of markers on the chest and abdomen for latency compensation and robotic control in radiotherapy for the first time. In our experiments, modern online learning algorithms enabled RNNs to adapt to irregular breathing patterns with limited data and achieve competitive accuracy compared to prior deep learning approaches trained on larger datasets, while keeping inference time relatively low.

UORO article:
Follow-up article (covering also SnAp-1 and DNI):

Simultaneous segmentation and recognition of characters @ Idemia (Île-de-France, 2016)

I prototyped an OCR pipeline in C for degraded text images, consisting of three steps: over-segmentation, character recognition, and shortest-path graph decoding via the Viterbi algorithm to select the most likely string (cf the image above). It achieved an accuracy 13.4% higher than Tesseract 3.04, an open-source software. The first step of the pipeline (candidate cut generation) relied on the Dijkstra algorithm, as each pixel was assigned to a node in the graph representation of the image and weights were based on pixel intensities; I optimized speed using a binary-search-style approach. Regarding the second step, I implemented from scratch a character-level neural network classifier, trained it on a database of ~20k character images, and improved recognition accuracy by ~5% by including statistical information on aspect ratio.

Graph matching for computer-aided neurological disease diagnosis with PET imaging @ Fresnel Institute (France, Marseille / 2016)

I applied spectral graph theory to the classification of PET images to assess whether a subject's brain is healthy or not. The F1 classification score achieved was 0.78/1.0

Other projects

Participation in the French robotics cup (2015)

We built and programmed autonomous robots that execute actions on a table to score as many points as possible during a 90 seconds match. I actively took part in the following tasks (among others):

  • micro-controller programming in C/C++ (obstacle avoidance, motor control using a proportional controller, programming the sequence of actions that the robot executes). Tools: VisualMicro / Arduino
  • selection/purchase of the electronic components and connection/interfacing with the micro-controller

Mechanical modelling of a periodic bundle of fibers @ IEMN (France, Nord / 2015)

I described the mechanical response of the mechanical structure mentioned above under various conditions using systems of differential equations based on physical laws and linear algebra. This work enhances our understanding of DNA degradation due to X-rays in radiotherapy, and also represents a step forward for the development of new bioinspired materials.

Development of new fluid dynamics models to assess the feasibility of non-contact sensors measuring borewell vibrations @ Schlumberger (Japan, Kanagawa / 2014)

I developed a fluid flow mathematical model to assess the feasibility of a non-contact sensor measuring borehole vibrations and requiring less time and power compared with usual devices. I derived and interpreted analytical pressure and velocity expressions as a function of the wellbore vibration parameters in different scenarios using partial differential equations, Fourier analysis, and asymptotic expansions to back up the sensor design recommendations.

Recommendations

Ritu Bhusal Chhatkuli, Researcher at the National Institute of Radiological Sciences and Assistant Professor at Chiba University

I had the pleasure to advise Michel for almost five years on his research project about tumor position prediction for robotic control in radiotherapy, which helped him gain a thorough understanding of machine learning and computer vision. He developed the ability to identify relevant problems quickly, find innovative solutions, prioritize tasks, communicate results effectively, and handle uncertainty. When talking with him, one can appreciate his sense of humor and modesty. He is a great engineer and computer scientist.

Stefano Giordano, Senior researcher at IEMN CNRS

As a supervisor of one of Michel's projects during his course of study, I should admit that he stands among the best young researcher I have ever worked with. Michel proved to have very good skills as an engineer, and also as a physicist: he is brilliant, he is able to developing at the same time both original physical ideas and code implementations. Besides, he is a very committed and concerned collaborator. In addition to the above professional skills, I like to add that Michel is a very nice and pro-active person. It is a real pleasure to collaborate with him, since he is able to offer the best combination of original thought and attitude of a hard team work.

Mouloud Adel, Professor at Aix-Marseille University

I met Michel Pohl when he was a master's student in 2016. He attended my lectures on digital image processing and he was always asking me very interesting questions on the mathematical aspects of image processing. During the same year, I supervised his master's internship at Fresnel Institute on brain graph representation from positron emission tomography for computer-aided diagnosis purposes. Michel obtained very interesting results but unfortunately I was not able to obtain a scholarship to keep him working with me on a PhD subject. Michel is a very good scientist and he has all the skills to work either in a university or in a company.

Oussama Zouaghia, Quantitative Analyst at Gunvor Group Ltd

I had the pleasure of working with Michel for two years on several projects at the Ecole Centrale de Lille. And, I have to say that I was impressed by Michel’s analytical skills. He was an adept at assimilating a lot of new information and he had good collaboration skills within the other members of the project team.