Yifang Chen
@yiiiifang.bsky.social
PhD student @uw. Undergrad @usc. Data curation, active learning, ml theory.
2. The poster paper arxiv.org/pdf/2406.18853 introduces a super simple but effective decoding-time TLHF algorithm that can align with multiple objectives, inspired by theory from f-divergence.
arxiv.org
December 6, 2024 at 10:12 PM
2. The poster paper arxiv.org/pdf/2406.18853 introduces a super simple but effective decoding-time TLHF algorithm that can align with multiple objectives, inspired by theory from f-divergence.
Checkout my two papers here: 🧐🧐
1. The spotlight paper arxiv.org/pdf/2405.19547 introduces negCLIPLoss & NormSim, two universally applicable data selection strategies that lead to SOTA on #DataComp-medium, a popular testbed for dataset design.
1. The spotlight paper arxiv.org/pdf/2405.19547 introduces negCLIPLoss & NormSim, two universally applicable data selection strategies that lead to SOTA on #DataComp-medium, a popular testbed for dataset design.
arxiv.org
December 6, 2024 at 10:12 PM
Checkout my two papers here: 🧐🧐
1. The spotlight paper arxiv.org/pdf/2405.19547 introduces negCLIPLoss & NormSim, two universally applicable data selection strategies that lead to SOTA on #DataComp-medium, a popular testbed for dataset design.
1. The spotlight paper arxiv.org/pdf/2405.19547 introduces negCLIPLoss & NormSim, two universally applicable data selection strategies that lead to SOTA on #DataComp-medium, a popular testbed for dataset design.