Yi Hung Lim

Yi Hung Lim

I'm a undergraduate student with a strong interest at the intersection of AI and other fields.

Posts by Collection

portfolio

Human level Atari 200x faster Reimplementation

  • Implemented a deep reinforcement-learning agent based on DQN that surpasses human benchmark on the atari suite.
  • Implemented algorithmic optimizations to reduce compute requirements by 200x compared to its predecessor Agent57.
  • Achieved state-of-the-art performance across all 57 Atari games, demonstrating robust generalization and sample efficiency.

Transformer Gallery

  • Architected and implemented core Transformer architectures—including Transformer, Transformer-XL, Longformer, and Block-Recurrent Transformer—mirroring foundations of large-scale language models.
  • Bootstrapped the codebase and personally wrote 80 % of the implementation, ensuring modularity for easy extension and experimentation.
  • Validated model correctness through benchmarked language modeling tasks and attention‐visualization tools.

Recent AI Papers Website

  • Built a web platform that automatically aggregates and ranks the week’s newest AI research papers by author prominence and citation count.
  • Implemented real-time features: “Like” button, personalized saving of papers, and dynamic category filtering via interactive JavaScript charts.
  • Integrated Google OAuth for secure login and user session management, facilitating personalized reading lists and alerts.

Navy Battle (Clash Royale style 2D Game)

  • Designed and developed a 2D multiplayer strategy game in Unity inspired by Clash Royale, featuring three unique character classes.
  • Programmed core gameplay mechanics (unit spawning, resource management, combat resolution) in C#, ensuring smooth networked play.
  • Implemented UI/UX elements (health bars, cooldown timers) and balanced character abilities through iterative playtesting.

publications

Estimating Uncertainty with Implicit Quantile Network

preprint, 2023

Inspired by Implicit Quantile Network from the reinforcement learning literature, this work aims to repurpose it for uncertainty estimation in supervised learning settings by modeling the entire distribution of the error.

Latent Diffusion with LLMs for Reasoning

preprint, 2024

This work augments encoder-decoder language models with latent diffusion models in an attempt to enhance language reasoning by diffusing the intermediate representations. By reasoning over many iterative forward steps, it has the ability to allocate more compute to hard reasoning tasks.

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.