Youjing YU

Youjing YU

Engineering PhD Student at University of Cambridge

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

experiences

SOFO Project

University of Cambridge, Department of Engineering,

  • Developed SOFO, a second-order subspace Gauss-Newton optimiser that uses parallelised forward-mode differentiation instead of backpropagation, achieving constant memory complexity with respect to sequence length.
  • SOFO enables efficient training of recurrent neural networks on long-horizon tasks, outperforming Adam on challenging benchmarks while maintaining per-iteration runtimes comparable to first-order optimisers.
  • Check out our Neurips Spotlight paper here.

Masters Project

University of Cambridge, Department of Engineering, Oct 2022 - Jun 2023

  • Developed a webcam-based eye-tracking system using TensorFlow FaceMesh to estimate on-screen gaze.
  • We extended WebGazer in two ways: first, we implemented an online process that dynamically tracks environment variables such as the user’s viewing distance from the screen with one initial simple calibration task. Second, we apply Gaussian Process regression with a novel kernel to map eye images to eyeball rotation angles.
  • We achieve on average 22% reduction in gaze-prediction error compared to the original WebGazer. Check out the project here.
  • Supervisor: Dr Guillaume Hennequin

NLP Student Researcher

University College London, Department of Computer Science, Dec 2021 - Oct 2022

  • Developed UNSURE, a Spotify Podcast dataset of user queries and human confidence labels crowdsourced with Amazon Mturk.
  • Developed a transformer-based multimodal network combining text and audio to estimate questioners’ topic-specific knowledge confidence.
  • Check out our paper at EANN!
  • Supervisor: Dr Aldo Lipani

NLP Student Researcher

University of Cambridge, Department of Engineering, Jul 2021 - Oct 2021

  • Developed a Flask-based NLP corpus analysis tool featuring two bias-detection algorithms to visualise and explore gender bias interactively.
  • Implemented a novel debiasing feature and natural-language query processing.
  • Supervisor: Dr Marcus Tomalin

Research Intern

University of Cambridge, Department of Engineering, Mar 2021 - Apr 2021

  • Developed an unsupervised machine-learning pipeline for car number plate recognition in videos using computer vision tools, including OpenCV and Pytesseract.
  • Applied maximum likelihood estimation for data cleaning and preprocessing.
  • Supervisor: Professor Adam Boies

Student Research Intern

SpacePT, Oct 2020 - Dec 2020

  • Built a computer vision change detector system for identifying and predicting building damage and forest fires from satellite images.

Student Researcher

University of Singapore, Department of Building, Mar 2017 - Jan 2018

  • Prepared nano encapsulated organic particles in the wet lab.
  • Our Cadmium Orthostannate nanoparticles with 3-(mercaptopropyl)trimethoxysilane suspended in organic solution were shown to have effective thermal insulation abilities.
  • Silver Award at Singapore Science and Engineering Fair.
  • Supervisor: Professor Shah Kwok Wei

Student Assistant Researcher

University of Hong Kong, Department of Electrical and Electronic Engineering, Jul 2020 - Sep 2020

  • Developed deep learning models in TensorFlow for holography image generation and image-to-image translation, implementing and evaluating CNN, ResNet, Wide ResNet, DenseNet, and SqueezeNet architectures.
  • Supervisor: Professor Edmund Lam

projects

work_experiences

Machine Learning Research Intern

  • We introduce a framework that exploits weight-space symmetries in neural network loss landscapes to construct tractable, structured Hessian approximations from a single gradient, with a user-defined symmetry group controlling the trade-off between accuracy and computational cost.
  • The framework unifies existing curvature estimation methods, including Shampoo/Muon-like approaches, and demonstrates strong potential for second-order optimisation and applications such as uncertainty estimation, continual learning, and model compression.
  • Check out our ICML paper here.
  • Supervisor: Alberto Bernacchia