Boqian is a third-year Joint Ph.D. student at the University of Twente in the Data Management & Biometrics (DMB) group and the University of Luxembourg in the BlueNN group. Her Ph.D. research focuses on sparse neural networks, particularly their theoretical aspects. Her goal is to uncover the mechanisms underlying the practical effectiveness of sparse neural networks.


News

  • Jan 01 2025. I start work at University of Luxembourg.
  • Nov 01 2024. Grant Accepted at SURF Compute Call. Project: Large Scaled Neural Network Dynamic Sparse Training.  Nov. 2024 ~ Nov. 2025.
  • Oct 05 2024. I was awarded the NeurIPS Travel Grant for NeurIPS 2024.
  • Sep 21, 2024. One papers is accepted by NeurIPS 2024.
  • May 21, 2024. Our “Edge LLMs: Edge-Device Large Language Model Competition” competition has been accepted by NeurIPS 2024. Submission opens Link.


Selected Publications

  •  Qiao Xiao*, Boqian Wu*, Yu Zhang, Shiwei Liu, Mykola Pechenizkiy, Elena Mocanu, Decebal Constantin Mocanu. Dynamic Sparse Network for Time Series Classification: Learning What to “See”. 36th Annual Conference on Neural Information Processing Systems, NeurIPS 2022. (* equal contribution)
  • Qiao Xiao, Boqian Wu, Lu Yin, Maurice van Keulen, Mykola Pechenizkiy. Can Less Yield More? Insights into Truly Sparse Training. ICLR 2023 Workshop on Sparsity in Neural Networks, Kigali, Rwanda.
  • Shiwei Liu, Tianlong Chen, Xiaohan Chen, Xuxi Chen, Qiao Xiao, Boqian Wu, Mykola Pechenizkiy, Decebal Mocanu, Zhangyang Wang. More convnets in the 2020s: Scaling up kernels beyond 51x51 using sparsity. ICLR 2023, Kigali, Rwanda.
  • Boqian Wu, Benmeng Lyv, Jiawen Gu. Weighted Multivariate Mean Reversion for Online Portfolio Selection. European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2023, Torino, Italy.
  • Benmeng Lyv, Boqian Wu, Sini Guo, Jiawen Gu, WaiKi Ching. Online Portfolio Selection With Constant Cash Inflows, 2024. Omega: The International Journal of Management Science.
  • Qiao Xiao, Boqian Wu, Lu Yin, Christopher Gadzinski, Tianjin Huang, Mykola Pechenizkiy, Shiwei Liu, Decebal Constantin Mocanu. Are Sparse Neural Networks Better Hard Sample Learners? British Machine Vision Conference, BMVC 2024.
  • Boqian Wu*, Qiao Xiao*, Shiwei Liu, Lu Yin, Mykola Pechenizkiy, Decebal Constantin Mocanu, Maurice van Keulen, Elena Mocanu. "Dynamic sparse training for accurate and effective 3D medical image segmentation", 38th Annual Conference on Neural Information Processing Systems, NeurIPS 2024. (* equal contribution)
  •  Qiao Xiao, Pingchuan Ma, Adriana Fernandez-Lopez, Boqian Wu, Lu Yin, Stavros Petridis, Mykola Pechenizkiy, Maja Pantic, Decebal Constantin Mocanu, Shiwei Liu. Dynamic Data Pruning for Automatic Speech Recognition, 25th Interspeech 2024, Kos Island, Greece.
  • Boqian Wu,  Maurice van Keulen,  Decebal Constantin Mocanu, Elena Mocanu. Insights into Dynamic Sparse Training: Theory Meets Practice. ECML PKDD 2024, PhD Forum, VILNIUS, LITHUANIA. 


CO-ORGANIZER ACTIVITIES

  • NeurIPS 2024 Challenge (Co-organizer). Shiwei Liu, Kai Han, Adriana Fernandez-Lopez, AJAY KUMAR JAISWAL, Zahra Atashgahi, Boqian Wu, Edoardo Ponti, Callie Hao, Rebekka Burkholz, Olga Saukh, Jared Tanner, Yunhe Wang. Edge-Device Large Language Model Competition. NeurIPS 2024 Competition Track.  https://edge-llms-challenge.github.io/edge-llm-challenge.github.io/
  • IJCAI 2023 Tutorial  (Co-organizer). Elena Mocanu, Zahra Atashgahi, Ghada Sokar, Boqian Wu, Qiao Xiao, Bram Grooten, Shiwei Liu, Decebal Constantin Mocanu. \textit{Sparse Training for Supervised, Unsupervised, Continual, and Deep Reinforcement Learning with Deep Neural Networks. IJCAI 2023. https://ijcai-23.org/tutorials/

Organisations

Dynamic sparse training algorithms have shown promise in achieving high performance while reducing resource costs, making them an attractive option in the field of machine learning. However, despite their potential, the theoretical properties of dynamic sparse training remain largely unexplored. This research seeks to fill this knowledge gap by thoroughly investigating the theoretical properties of sparse training models, such as their representational ability, generalization, and convergence. As a result, guidelines will be developed for applying dynamic sparse training to real-world problems.

Publications

2024

Robust online portfolio optimization with cash flows (2024)Omega, 129. Article 103169. Lyu, B., Wu, B., Guo, S., Gu, J. & Ching, W.-K.https://doi.org/10.1016/j.omega.2024.103169Are Sparse Neural Networks Better Hard Sample Learners? (2024)In British Machine Vision Conference (BMVC 2024). Xiao, Q., Wu, B., Yin, L., Gadzinski, C. N., Huang, T., Pechenizkiy, M. & Mocanu, D. C.Dynamic Sparse Training versus Dense Training: The Unexpected Winner in Image Corruption Robustness (2024)[Working paper › Preprint]. ArXiv.org. Wu, B., Xiao, Q., Wang, S., Strisciuglio, N., Pechenizkiy, M., van Keulen, M., Mocanu, D. C. & Mocanu, E.https://doi.org/10.48550/arXiv.2410.03030Are Sparse Neural Networks Better Hard Sample Learners? (2024)[Working paper › Preprint]. ArXiv.org. Xiao, Q., Wu, B., Yin, L., Gadzinski, C. N., Huang, T., Pechenizkiy, M. & Mocanu, D. C.https://doi.org/10.48550/arXiv.2409.09196Insights into Dynamic Sparse Training: Theory Meets Practice (2024)[Contribution to conference › Poster] European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2024. Wu, B., van Keulen, M., Mocanu, D. C. & Mocanu, E.Dynamic Data Pruning for Automatic Speech Recognition (2024)In Interspeech 2024 (pp. 4488-4492). Xiao, Q., Ma, P., Fernandez-Lopez, A., Wu, B., Yin, L., Petridis, S., Pechenizkiy, M., Pantic, M., Mocanu, D. C. & Liu, S.Dynamic Data Pruning for Automatic Speech Recognition (2024)[Working paper › Preprint]. ArXiv.org. Xiao, Q., Ma, P., Fernandez-Lopez, A., Wu, B., Yin, L., Petridis, S., Pechenizkiy, M., Pantic, M., Mocanu, D. C. & Liu, S.https://doi.org/10.48550/arXiv.2406.18373E2ENet: Dynamic Sparse Feature Fusion for Accurate and Efficient 3D Medical Image Segmentation (2024)In 38th Annual Conference on Neural Information Processing Systems, NeurIPS 2024. MLResearchPress. Wu, B., Xiao, Q., Liu, S., Yin, L., Pechenizkiy, M., Mocanu, D. C., van Keulen, M. & Mocanu, E.https://openreview.net/forum?id=Xp8qhdmeb4

2023

E2ENet: Dynamic Sparse Feature Fusion for Accurate and Efficient 3D Medical Image Segmentation (2023)[Working paper › Preprint]. ArXiv.org. Wu, B., Xiao, Q., Liu, S., Yin, L., Pechenizkiy, M., Mocanu, D. C., van Keulen, M. & Mocanu, E.https://doi.org/10.48550/arXiv.2312.04727Weighted Multivariate Mean Reversion for Online Portfolio Selection (2023)In Machine Learning and Knowledge Discovery in Databases: Research Track: European Conference, ECML PKDD 2023, Turin, Italy, September 18–22, 2023, Proceedings, Part V (pp. 255-270) (Lecture Notes in Computer Science; Vol. 14173). Wu, B., Lyu, B. & Gu, J.https://doi.org/10.1007/978-3-031-43424-2_16

Other contributions

  • NeurIPS 2024 Challenge (Co-organizer). Shiwei Liu, Kai Han, Adriana Fernandez-Lopez, AJAY KUMAR JAISWAL, Zahra Atashgahi, Boqian Wu, Edoardo Ponti, Callie Hao, Rebekka Burkholz, Olga Saukh, Jared Tanner, Yunhe Wang. Edge-Device Large Language Model Competition. NeurIPS 2024 Competition Track.  https://edge-llms-challenge.github.io/edge-llm-challenge.github.io/
  • IJCAI 2023 Tutorial  (Co-organizer). Elena Mocanu, Zahra Atashgahi, Ghada Sokar, Boqian Wu, Qiao Xiao, Bram Grooten, Shiwei Liu, Decebal Constantin Mocanu. \textit{Sparse Training for Supervised, Unsupervised, Continual, and Deep Reinforcement Learning with Deep Neural Networks. IJCAI 2023. https://ijcai-23.org/tutorials/

Research profiles

  • Joint Ph.D. in Artificial Intelligence, University of Twente and University of Luxembourg Oct. 2022 - Dec. 2026.
  • MPhil. in Communication Engineering, Faculty of Electronics and Information Engineering, Harbin Institute of Technology, China Sept. 2015 - Jan. 2018
  • BEng. in Communication Engineering, Faculty of Electronics and Information Engineering, Harbin Institute of Technology, China Sept. 2011 - Jun. 2015
Scan the QR code or
Download vCard