Hou Yunqing

Model stuff

[Zhanyi Sun @s_zhanyi]

How to prevent behavior cloning policies from drifting OOD on long horizon manipulation tasks?
Check out Latent Policy Barrier (LPB), a plug-and-play test-time optimization method that keeps BC policies in-distribution with no extra demo or fine-tuning: https://t.co/SM21Yd9dtp

[Hou Yunqing @hyqneuron]

@VictorTaelin @janiczek We use then-chaining to replace do notation. Not nearly as terse as do notation but it removes callback hell.

[Kevin Frans @kvfrans]

Flow-matching implementation:

https://t.co/sP5DXLr4jI

Flow-matching is very similar to diffusion, but simplifies things. Noised images are linear interpolations between (data, noise) pairs, and the network predicts *velocity* of this trajectory.

[Hou Yunqing @hyqneuron]

@PredragGruevski @seanbax I support this. Wanna start a discord server? @seanbax . Let's all abandon hope together lol

[Eddy Quan @waronweakness]

It's been said before but it's worth repeating:

You become what you can't stop thinking about.

[Hou Yunqing @hyqneuron]

@DrEigenbastard @bgavran3 Is there a link? Google didn’t seem to return anything useful

[Hou Yunqing @hyqneuron]

@KL_Div @PengShichong Is there a name for the unconditional case?
- each real point has a vote
- generated points (candidates) go close to real points to win votes

x_fake = G(z), so generation is unconditional. Smells similar to generative topographic map.

[Ke Li 🍁 @KL_Div]

Excited to share our latest work, which generates high-res images non-adversarially without the use of a discriminator. Benefits include: (1) no mode collapse and (2) no training instability.

Joint work with @PengShichong
Paper and Code: https://t.co/Np78VTDuRP

Thread

1/n

[Hou Yunqing @hyqneuron]

@seanbax @iquilezles I’m here for the nice graphics 😍

[Hou Yunqing @hyqneuron]

The democratic process would then bear far greater similarity to investing, which really is what voting is all about.

[Hou Yunqing @hyqneuron]

Voting is a special form of universal basic income.

[Hou Yunqing @hyqneuron]

If UBI exists in monetary form it would make the expression and negotiation of preference more efficient.

[Hou Yunqing @hyqneuron]

https://t.co/7MB8lSEDHt

[Hou Yunqing @hyqneuron]

@seanbax Well, put it on YouTube pls!

[William Bialek @wbialek]

Sad news. Horace Barlow, one of the great thinkers about the brain, has passed away. Born 8 Dec 1921, he saw almost a full century, and taught us about vision, perceptual inference, neural coding, learning, and more. #neuroscience #vision #NeuralNetworks

[Hou Yunqing @hyqneuron]

@seanbax It worries me that these custom tools, each hijacking and taking a piece from the AST, will not be composable. Or maybe they will be, but composing them requires so much expertise it will not be a pleasant experience.

[Erik Nijkamp @erik_nijkamp]

Our work "Flow Contrastive Estimation of Energy-Based Models" (@RuiqiGao, @erik_nijkamp, @dpkingma, Z. Xu @andrewdai, YN Wu) at #NeurIPS2019:
https://t.co/KIsc7F1nJd
We show (1) joint learning of EBM & Glow, (2) correction of over-dispersion, (3) competitive semi-super cls.

[Pietro Lesci @pietro_lesci]

I've always been fascinated by Bayesian Nonparametrics. I struggled to grasp those ideas directly from papers. Today, by chance, I found the best (imo) single reference for anyone interested: A gentle introduction by @yeewhye and Michael Jordan

https://t.co/tDlRHc10Md

[Hou Yunqing @hyqneuron]

@YSongStanford This is fantastic work. Has anyone tried using intermediate features of the score network for classification?