Haochen Zhang (@jhaochenz) / X

Haochen Zhang

33 posts

Haochen Zhang

@jhaochenz

Member of Technical Staff @AnthropicAI | prev. CS PhD @StanfordAILab

San Francisco

cs.stanford.edu/~jhaochen/

Joined August 2013

Haochen Zhang
@jhaochenz
Jun 20, 2024
I joined the amazing team at @AnthropicAI last month. Excited to share what we’ve been working on recently!
Anthropic
@AnthropicAI
Jun 20, 2024
Introducing Claude 3.5 Sonnet—our most intelligent model yet. This is the first release in our 3.5 model family. Sonnet now outperforms competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost. Try it for free: claude.ai
26K
Haochen Zhang
@jhaochenz
Nov 28, 2023
🔥🔥🔥 Congrats on the launch! @demi_guo_ @chenlin_meng
Pika
@pika_labs
Nov 28, 2023
Introducing Pika 1.0, the idea-to-video platform that brings your creativity to life. Create and edit your videos with AI. Rolling out to new users on web and discord, starting today. Sign up at pika.art
00:00
2.3K
Haochen Zhang
@jhaochenz
Jun 17, 2020
Glad to share this new work! We theoretically show that parameter-dependent noises provide a bias toward parameters where noise itself is small. This is a potentially stronger effect than simply escaping sharp local minima. Code for our project is publicly available!
Tengyu Ma
@tengyuma
Jun 17, 2020
We theoretically prove that not only the scale of the noise in SGD has an implicit regularization, but its covariance also matters. For a simple model, SGD with label noise biases towards sparsity and leaves the NTK regime, whereas Gaussian noise does not. arxiv.org/abs/2006.08680
Haochen Zhang
@jhaochenz
May 5, 2023
How does model architecture influence the contrastive representations? Check out our paper "A theoretical study of inductive biases in contrastive learning" at #ICLR2023 . Virtual poster: iclr.cc/virtual/2023/p…. Joint work with @tengyuma.
1.1K
Haochen Zhang
@jhaochenz
Apr 18, 2022
Replying to @Josh_d_robinson @ananyaku and @tengyuma
My understanding is that in our spectral contrastive paper, the inductive bias of the architecture plays a role in the "empirical loss -> population loss" part: implicitly, we are assuming that there exists a model with small weight that can minimize the *population* loss.
Haochen Zhang
@jhaochenz
Apr 18, 2022
Replying to @jhaochenz @Josh_d_robinson and 2 others
I like this approach because with this relatively simple assumption, we can avoid getting into the complication of inductive biases of specific architectures (which is not well understood even for supervised learning), and instead focus on properties of the contrastive loss.
Haochen Zhang
@jhaochenz
May 8, 2024
Replying to @niladrichat
Congrats!
75
Haochen Zhang
@jhaochenz
Sep 3, 2024
Replying to @yanda_chen_ and @AnthropicAI
Welcome!
290
Haochen Zhang
@jhaochenz
Jul 18, 2024
Replying to @ren_hongyu
Congrats!
289