Yongwei Che (@CheYongwei) profile picture

Currently @ Radix Trading, Princeton 2024

Joined August 2024
Photo by Yongwei Che (@CheYongwei)Photo by Yongwei Che (@CheYongwei)Photo by Yongwei Che (@CheYongwei)
Yongwei Che retweeted
It鈥檚 actually incredible just how fast Indians burned all the social capital they had in the west. 10 years ago when you asked a young American about Indians the response was something positive about some YouTube video teaching calc 101 better than their high school teachers. Now nobody can stand you. That鈥檚 how damaging you guys not assimilating to America was. Nobody wants to see you in stores, they hate how you crowd around free samples in Costco, everyone cringes when we hear your flip flops clap around the gas station. The near universal heel turn from ambivalence/positive views to disgust and contempt of indians has to be studied. Truly remarkable.
The hate on Indians is actually fkn absurd.. 馃拃 It's so funny when people ask me where I'm from and I say India and their entire mood towards me switches, but when I say I grew up in Dubai it switches back and they start saying shit like "woah that's so cool"
135
553
11
9,565
206,383
I'll be presenting this work at ICLR 2025 in Singapore! Stop by to learn more :) - Poster: Fri 3pm-5:30pm Hall 3 + Hall 2B #328 #ICLR2025
1/ When using multimodal models, we often want to infer over modalities unseen together in training. Prior works typically address this by learning a shared contrastive embedding space between modality pairs, but is this approach principled? We prove that it is! (under certain assumptions) Paper+Code: arxiv.org/html/2501.11326v1
1
2
139
1/ When using multimodal models, we often want to infer over modalities unseen together in training. Prior works typically address this by learning a shared contrastive embedding space between modality pairs, but is this approach principled? We prove that it is! (under certain assumptions) Paper+Code: arxiv.org/html/2501.11326v1
1
1
1
2
288
2/ We also provide a Monte Carlo algorithm (coined LogSumExp) for when these assumptions aren't met, allowing us to infer across multimodal pre-trained models that weren't trained together (such as CLIP and CLAP)
1
33
3/ We apply our results to language-conditioned reinforcement learning, showing better goal-reaching when prompted with ambiguous language goals! Direct evaluation navigates to the "mean embedding", while our LogSumExp takes the shortest path.
26