YI REN
≡
profile photo

Yi (Joshua) Ren

Hey, I am a Ph.D. (2020-now) student who is working on machine learning under the supervision of Prof. Danica J. Sutherland at the University of British Columbia (UBC). I am trying to figure out how to train a model that generalizes well systematically. This idea originates from my master's study at the University of Edinburgh (2018-2019) with Prof. Simon Kirby and Prof. Shay Cohen. We find that not only humans, but the neural network might also prefer the highly compositional mapping when trained under different tasks. Such a preference might make the learned representations become more compositional if we train the model iteratively. To further explore this interesting phenomenon, I visited Prof. Aaron Courville's group at Mila, during which we find that "soft discretization" might play an important role. In the remainder of my Ph.D. study, I hope we can build a theoretical model to explain the aforementioned preference during neural network training, and also design more efficient algorithms for real applications. Recently, I surprisingly found that the Bayesian iterated learning framework (a hypothetic framework depicting human cultural evolution used in cognitive science) has the potential to approximate the behavior of the evolution of large language models (LLMs). Hope this theoretical framework can bring valuable insights to the LLM era!

[News] We are going to have an interesting workshop on Language Gamification at NeurIPS 2024 @ Vancouver this winter.

Email  /  Google Scholar  /  GitHub  /  Twitter  /  CV

UBC Machine Learning  /  MILD  /  AML-TN  / 

News

  • 09/2024, one paper accepted by NeurIPS-2024, talks about how LLM's knowledge gradually evolves if we keep conducting self-improving methods.

  • 07/2024, happy to participate in organizing a workshop about Language Gamification in NeurIPS 2024 @ Vancouver.

  • 01/2024, one paper accepted by ICLR-2024, depicting sample difficulty in NTK space.

  • 09/2023, one paper accepted by NeurIPS-2023, iterated learning is helpful in representation learning.

  • 09/2023, start an internship at Borealis AI, working on time series prediction project.

  • 02/2023, one paper accepted by ICLR-2023, leave enough energy for feature adaptation.

    • Highlight Topics

    I believe "clustering" my work by topics can provide a good overview of my research interest. Here are some.

  • Simplicity Bias and Learning Dynamics

  • Neural Iterated Learning on Various Applications

    • Talks

  • 07/2024 Happy to share some of my understandings about the Platonic Representation Hypothesis . (Slides)

  • 03/2024 Happy to give a talk at Chalmers University of Technology about the application and understanding of neural iterated learning. (Slides)

    • Notes and TA

  • Here are links for TA sessions of CPSC 340 (Machine Learning and Data Mining - Fall 2024):

    Week 1: basic knowledge review

    Week 2: Variance-bias, KNN, Naive Bayes

    Week 3: Ensemble, K-means, Recap of Supervised Learning

    Week 4: Linear Regression

    Week 5: Gradient Descent and Midterm

    Week 6: Feature Selection and Regularization

    • Publications

    Preprints:

    1. Understanding Simplicity Bias towards Compositional Mappings via Learning Dynamics
      Yi Ren, Danica J. Sutherland
      Preprint 2024 | pdf | code

    2. Learning Dynamics of LLM Finetuning
      Yi Ren, Danica J. Sutherland
      Preprint 2024 | pdf | code

    3. Economics arena for large language models
      Shangmin Guo, Haoran Bu, Haochuan Wang, Yi Ren, Dianbo Sui, Yuming Shang, Siting Lu
      Preprint 2024 | pdf

    Journal and Low-Acceptance-Rate Conference Papers:

    1. Bias Amplification in Language Model Evolution: An Iterated Learning Perspective
      Yi Ren, Shangmin Guo, Linlu Qiu, Bailin Wang, Danica J. Sutherland
      NeurIPS 2024 | pdf | code

    2. AdaFlood: Adaptive Flood Regularization
      Wonho Bae, Yi Ren, Mohamad Osama Ahmed, Frederick Tung, Danica J Sutherland, Gabriel L Oliveira
      Transactions on Machine Learning Research (TMLR) 2024 | pdf

    3. lpNTK: Better Generalisation with Less Data via Sample Interaction During Learning
      Shangmin Guo, Yi Ren, Stefano V. Albrecht, Kenny Smith
      ICLR 2024 | pdf

    4. Improving Compositional Generalization using Iterated Learning and Simplicial Embeddings
      Yi Ren, Samuel Lavoie, Mikhail Galkin, Danica J. Sutherland, Aaron Courville
      NeurIPS 2023 | pdf | code

    5. How to prepare your task head for finetuning
      Yi Ren, Shangmin Guo, Wonho Bae, Danica J. Sutherland
      ICLR 2023 | pdf | code

    6. Better Supervisory Signals by Observing Learning Paths
      Yi Ren, Shangmin Guo, Danica J. Sutherland
      ICLR 2022 | pdf | code

    7. Expressivity of Emergent Language is a Trade-off between Contextual Complexity and Unpredictability
      Shangmin Guo, Yi Ren, Kory Mathewson, Simon Kirby, Stefano V. Albrecht, Kenny Smith
      ICLR 2022 | pdf | code | workshop-version

    8. Compositional languages emerge in a neural iterated learning model
      Yi Ren, Shangmin Guo, Matthieu Labeau, Shay B. Cohen, Simon Kirby
      ICLR 2020 | pdf | code | workshop-version

    9. The Emergence of Compositional Languages for Numeric Concepts Through Iterated Learning in Neural Agents
      Shangmin Guo, Yi Ren, Serhii Havrylov, Stella Frank, Ivan Titov, Kenny Smith
      EmeCom@NeurIPS 2019 | pdf

    Welcome to my home page *^_^*
    No. Visitor Since Jan 2022. Powered by w3.css