Experience

Researcher Engineer @ InstaDeep

Full-time, Jan. 2021 - present.

• Working on a project linked to DeepChain™.

Full-time, Sep. 2019 - Dec. 2020.

• Led the R&D of reinforcement learning solution and simulator for DeepPCB™.

Part-time, Oct. 2018 - Sep. 2019.

• Participated in the R&D of DeepPCB on mainstream RL algorithms comparison.
• Designed and implemented a novel model-free RL algorithm, which outperformed the plain Monte-Carlo tree search (MCTS).
• Extended the bin packing results to solve the glass cutting optimization problem using RL and we ranked 16 out of 60 teams.
• Led the research and prototype development of NLP (question answering, text summarization).

Internship, Feb. 2018 - Sep. 2018.

• Implemented RL algorithms to solve the bin packing problem.
• Developed ranked reward and outperformed the commercial solver (Gurobi), heuristic algorithms in both 2D and 3D bin packing problems.
• Studied deep RL in auto machine learning (neural network architecture search).
• Studied evolutionary algorithms in combinatorial optimization problems (TSP).
• Received the research internship award from École Polytechnique.

Honorary Research Assistant @ Medical Physics & Biomedical Engineering, University College London

Part-time, Nov. 2019 - present.

• Leading the R&D of DeepReg, an open-source project for image registration base on TensorFlow v2.
• Studied neural network calibration on COVID-19 X-Ray images.

Operations Research Analyst @ SNCF Réseau

Internship, Jun. 2017 - Sep. 2017.

• Modelled railway scheduling as a multi-objective optimization problem.
• Solved the scheduling problem with discrete programming (local search, tabu search, and simulated annealing).
• Outperformed human experts and approached the performance of LocalSolver.

Publication

Morphological Change Forecasting for Prostate Glands using Feature-based Registration and Kernel Density Extrapolation

Qianye Yang, Tom Vercauteren, Yunguan Fu, Francesco Giganti, Nooshin Ghavami, Vasilis Stavrinides, Caroline Moore, Matt Clarkson, Dean Barratt, Yipeng Hu

Accepted as a conference paper to ISBI 2021, arXiv.

• Applied kernel density estimation to forecast morphological changes for prostate glands.

DeepReg: a deep learning toolkit for medical image registration

Yunguan Fu, Nina Montaña Brown, Shaheer U. Saeed, Adrià Casamitjana, Zachary M. C. Baum, Rémi Delaunay, Qianye Yang, Alexander Grimwood, Zhe Min, Stefano B. Blumberg, Juan Eugenio Iglesias, Dean C. Barratt, Ester Bonmati, Daniel C. Alexander, Matthew J Clarkson, Tom Vercauteren, and Yipeng Hu

Accepted as a journal paper at the Journal of Open Source Software, GitHub.

• Implemented major unsupervised and weakly-supervised registration algorithms in a unified framework using TensorFlow 2.
• Provided multiple demos using open-accessible data on real-world clinical applications.

Longitudinal Image Registration with Temporal-Order and Subject-Specificity Discrimination

Qianye Yang, Yunguan Fu, Francesco Giganti, Nooshin Ghavami, Qingchao Chen, J Alison Noble, Tom Vercauteren, Dean Barratt, and Yipeng Hu

Accepted as a conference paper to MICCAI 2020, arXiv.

• Applied deep registration for clinical longitudinal data using staged sampling.
• Demonstrated that an MMD loss to penalize the divergence of data representations from different sampling strategies can improve the registration performance.

More unlabelled data or label more data? A study on semi-supervised laparoscopic image segmentation

Yunguan Fu, Maria R. Robu, Bongjin Koo, Crispin Schneider, Stijn van Laarhoven, Danail Stoyanov, Brian Davidson, Matthew J. Clarkson, and Yipeng Hu

Accepted to MICCAI 2019 Medical Image Learning with Less Labels and Imperfect Data Workshop (acceptance rate is 35%), arXiv

• Implemented image segmentation using U-Net and semi-supervised Mean Teacher model for laparoscopic images.
• Demonstrated that the specific training method of Mean Teacher is responsible for the performance improvement besides the additional unlabeled data.
• Demonstrated that adding more unlabeled data potentially could provide similar performance improvement compared to using more labeled data.

Ranked Reward: Enabling Self-Play Reinforcement Learning for Combinatorial Optimization

Alexandre Laterre, Yunguan Fu, Mohamed Khalil Jabri, Alain-Sam Cohen, David Kas, Karl Hajjar, Hui Chen, Torbjorn S. Dahl, Amine Kerkeni, and Karim Beguir.

Accepted to NeurIPS 2018 Deep RL Workshop and AAAI 2019 RL in Games Workshop, arXiv.

• Implemented AlphaZero, R2, and heuristic algorithm to solve bin packing problems.
• Designed neural network architectures to solve both 2D and 3D problems.
• Designed a supervised dataset to evaluate the neural network architectures.
• Designed and implemented linear programming models for bin packing problems using Gurobi solver.
• Conducted experiments to compare with MCTS, heuristic, supervised agent and Gurobi solver.