iiinova — where your digital self keeps evolving

Yining Lu

@yininglu.bsky.social

36 followers 187 following 16 posts

Second year CS PhD student @notredame.bsky.social | Intern: Amazon | Prev: @jhuclsp.bsky.social yining610.github.io

Posts Replies Media

Yining Lu @yininglu.bsky.social · 16/09/2025

8/8 [Convergence rate] The gradient-based method consistently has a higher convergence rate, reducing the required steps by 6.1 on average across RL algorithms.

100

Yining Lu @yininglu.bsky.social · 16/09/2025

7/8 [Generalizability] We further extend experiments to different math datasets and model families. Our two methods yield superior Pareto fronts compared to the baseline, with the gradient-based weighting showing the best overall performance.

100

Yining Lu @yininglu.bsky.social · 16/09/2025

6/8 [Gradient-based weight optimization] Our method generates superior Pareto fronts that dominate all baseline approaches under both GRPO and REINFORCE training.

100

Yining Lu @yininglu.bsky.social · 16/09/2025

5/8 [Hypervolume-guided weight adaptation] Across all three online RL algorithms, there is consistently at least one weight configuration our method outperforms the baselines on all objectives.

100

Yining Lu @yininglu.bsky.social · 16/09/2025

Dynamic reward weights show objectives learn differently. For example, accuracy is a more challenging objective that requires continual learning, while conciseness quickly converges to 0.2. 4/8

100

Yining Lu @yininglu.bsky.social · 16/09/2025

3/8 [Preliminary finding] Different objectives vary in learning difficulty. Each objective reaches saturation at different training stages.

100

Yining Lu @yininglu.bsky.social · 16/09/2025

✴️ Pleased to introduce our new paper yining610.github.io/dynamic-rew... - Rebalance multiobjectives during training through dynamic reward weighting - Build Pareto-dominant front over static baselines across online RL algorithms, datasets, and model families - Faster convergence rate 1/8

100

Yining Lu @yininglu.bsky.social · 25/07/2025

Can't make it to #ACL2025 this year, but for people interested in RL for factuality and textual decomposition, please check out our paper! TL;DR: We found a mismatch between the decomposition policy and LLM verifier, and propose a dynamic training paradigm to bridge the gap.

110

Yining Lu @yininglu.bsky.social · 02/05/2025

Quick reminder that our paper, Benchmarking Language Model Creativity: A Case Study on Code Generation, will be presented today! 📅 11AM-12:30PM, Fri, May 2 📍 Hall 3 📝 arxiv.org/abs/2407.09007 🎥 www.youtube.com/watch?v=v1c...

000