Kimia Hamidieh

@kimiahmdh

PhD student at @MIT_CSAIL, previously @UofT/@VectorInst, @GoogleDeepMind, @MSRNE

Cambridge, MA

kimiah.github.io

Joined September 2020

Replies Reposts Media

Pinned
Kimia Hamidieh
@kimiahmdh
Jul 16
Can we tell whether data domains cooperate or compete during pretraining? Adding code to the mix makes models better at math, while some other combinations hurt each other. We call this data synergy. Turns out you can incorporate data synergy into scaling laws and estimate it 🧵
4
Kimia Hamidieh
@kimiahmdh
Aug 10
How much is your synthetic data in pretraining actually worth? And when we run out of high-quality data, how far can more compute take us? Our new work quantifies how much fresh data different data expansion strategies are worth as we scale compute, and predicts how effectively
Sunny Qin
@sunnytqin
Aug 10
(1/N) 🧵 Chinchilla assumes you'll never run out of fresh data. That era is ending! Compute keeps growing exponentially, but high-quality tokens don't. So what's the exchange rate between extra compute and fresh, high-quality data? We propose Compute-Data (CD) scaling laws,
2
Kimia Hamidieh
@kimiahmdh
Dec 1, 2025
I'll be at #NeurIPS2025 next week! Looking forward to catching up with friends and meeting new people. Feel free to DM me to chat about data mixtures and scaling laws, AI reliability/safety, or anything related!
2
Kimia Hamidieh
@kimiahmdh
May 6, 2024
Our work "Views Can Be Deceiving: Improved SSL Through Feature Space Augmentation" has been accepted as a spotlight💡 at ICLR 2024. We uncover how self-supervised representations can inadvertently rely on spurious features and propose a method to mitigate this issue. A thread🧵
Marzyeh and Swami Sankaranarayanan
1