HN user

circuithunter

45 karma
Posts22
Comments3
View on HN
arxiv.org 6mo ago

DatBench: Discriminative, faithful, and efficient VLM evaluations

circuithunter
18pts0
arxiv.org 11mo ago

BeyondWeb: Lessons from Scaling Synthetic Data for Trillion-Scale Pretraining

circuithunter
4pts0
www.datologyai.com 1y ago

Models are what they eat: automatic data curation for LLMs

circuithunter
2pts0
arxiv.org 5y ago

Are all negatives created equal in contrastive instance discrimination?

circuithunter
1pts0
arxiv.org 6y ago

Training BatchNorm and Only BatchNorm: On the Power of Random Features in CNNs

circuithunter
4pts0
arxiv.org 6y ago

The Early Phase of Neural Network Training

circuithunter
3pts0
newsroom.fb.com 7y ago

Remove, Reduce, Inform: New Steps to Manage Problematic Content

circuithunter
2pts0
www.buzzfeednews.com 7y ago

Google Founders Have Skipped All of the Company's 2019 Town Hall Meetings

circuithunter
6pts0
www.vanityfair.com 7y ago

“Men Are Scum”: Inside Facebook's War on Hate Speech

circuithunter
13pts2
medium.com 7y ago

Why Fears of Fake News Are Overhyped

circuithunter
1pts0
newsroom.fb.com 7y ago

What Is Facebook Doing to Address the Challenges It Faces?

circuithunter
1pts0
venturebeat.com 7y ago

Facebook hires prominent privacy managers ahead of end-to-end encryption rollout

circuithunter
2pts0
www.wired.com 7y ago

AI Has Started Cleaning Up Facebook, but Can It Finish?

circuithunter
1pts0
www.nbcnews.com 7y ago

Facebook tries to explain why companies could erase your messages

circuithunter
3pts0
arstechnica.com 7y ago

Facebook “partner” arrangements: Are they as bad as they look?

circuithunter
1pts0
www.wired.com 7y ago

Leaked Audio Reveals Google's Efforts to Woo Conservatives

circuithunter
3pts0
ai.googleblog.com 8y ago

How Can DNN Similarity Help Us Understand Training and Generalization?

circuithunter
2pts0
arxiv.org 8y ago

DeepMind – On the importance of single directions for generalization

circuithunter
8pts0
deepmind.com 8y ago

Understanding Deep Learning Through Neuron Deletion – DeepMind

circuithunter
14pts0
deepmind.com 9y ago

DeepMind expands to Canada with new research office in Edmonton, Alberta

circuithunter
7pts1
www.arimorcos.com 11y ago

Clustering subreddits by common word usage

circuithunter
2pts0
www.arimorcos.com 11y ago

Creating a Reddit dataset

circuithunter
1pts0

This is a really nice paper which asks some critical questions for the future of NAS research.

However, it's important to note that this paper doesn't show that NAS algorithms as originally designed, with completely independent training of each proposed architecture, are equivalent to random search. Rather, it shows that weight sharing, a technique introduced by ENAS [1] which tries to minimize necessary compute by training multiple models simultaneously with shared weights, doesn't outperform random baselines. Intuitively, this makes sense: weight sharing dramatically reduces the number of independent evaluations, and thereby leads to far less signal for the controller, which proposes architectures.

The paper itself makes this fairly clear, but I think it's easy to misinterpret this distinction from the abstract.

[1] ] H. Pham, M. Y. Guan, B. Zoph, Q. V. Le, and J. Dean. Efficient neural architecture search via parameter sharing. ICML, 2018.

There's actually quite a bit of evidence suggesting that brains, both behaviorally and mechanistically, are Bayesian [0].

As for your second point, assuming that humans are Bayesian, there are many reasons why people would have variability in their mathematical ability, including different priors and differences in the ability to estimate posteriors.

[0] https://scholar.google.com/scholar?q=brain+bayesian&hl=en&bt...