Maxim Berman @bermanmaxim

https://t.co/Q41daKf3dN. He/him. Machine learning - physics - music. Research Software Engineer at Google Zürich. All opinions my own. bmax.im Zürich, Switzerland Joined April 2009

Tweets

197
Followers

166
Following

531
Likes

2K

Robby Stein @rmstein

9 months ago

3/ You can also take a picture of a complicated thing you see and the models will be able to annotate them for you in the reply if it's deemed to be helpful. Here's asking what's good for vegetarians on this menu:

1 3 14 917 1

Download Video

Maxim Berman @bermanmaxim

10 months ago

A fresh new start on bsky.app/profile/maximb… where I’ll probably end up posting just as little as here 😄

0 0 1 12 0

Shahram Izadi @izadi_shahram

a year ago

Excited to showcase our early research with @GoogleDeepMind in bringing Gemini to Glasses at #GoogleIO2024

Google DeepMind @GoogleDeepMind

a year ago

Excited to showcase our early research with @GoogleDeepMind in bringing Gemini to Glasses at #GoogleIO2024

216 981 4K 1.3M 735

Download Video

2 34 84 11K 4

Thiemo Alldieck @thiemoall

3 years ago

In the middle of your PhD and interested to work with us at Google Zurich? Consider applying for a Student Researcher position (flexible duration, time commitment, on-site/remote). Apply here and mention your interest in our team: careers.google.com/jobs/results/8… Questions? DMs are open!

4 51 241 39K 108

Davis Blalock @davisblalock

3 years ago

"Breaking the Softmax Bottleneck: A High-Rank RNN Language Model" I don't hear this paper talked about a lot, but a recent large-scale study found that it was one of only three methods that actually improved vanilla transformers. Here's how it works: [1/10]

5 109 689 0 340

Download Image

HeavensLastAngel @HvnsLstAngel

3 years ago

“A still of Kermit The Frog in Blade Runner 2049 (2017)” #dalle

116 2K 8K 0 1K

Download Image

Giannis Daras @giannis_daras

3 years ago

DALLE-2 has a secret language. "Apoploe vesrreaitais" means birds. "Contarra ccetnxniams luryca tanniounons" means bugs or pests. The prompt: "Apoploe vesrreaitais eating Contarra ccetnxniams luryca tanniounons" gives images of birds eating bugs. A thread (1/n)🧵

190 2K 9K 0 1K

Download Image

Maxim Berman @bermanmaxim

4 years ago

Impact of wordle on Google trends 📈 trends.google.com/trends/explore…

0 0 0 0 0

Download Image

Maxim Berman @bermanmaxim

4 years ago

🟨⬛⬛🟨⬛ ⬛🟩🟩⬛🟩 🟨🟩🟩⬛🟩 🟩🟩🟩🟩🟩 Wordle 187 4/6

0 0 0 0 0

Ross Wightman @wightmanr

4 years ago

The timm monster 'pareto curve' (x-axis log scale). 500+ points in one crazy graph. Dug this up today due to recent discussions. This variant is color / shape coded across architecture variants and pretrain dataset - technique. On GPUs, LeViT stands out where most don't.

9 39 192 0 28

Download Image

Lucas Beyer (bl16) @giffmana

4 years ago

1/N The return of patch-based self-supervision! It never worked well and you had to bend over backwards with ResNets (I tried). Now with ViT, very simple patch-based self-supervised pre-training rocks! First BeIT, now Masked AutoEncoders i1k=87.8% arxiv.org/pdf/2111.06377… 🧶

15 217 1K 0 344

Download Image

Andrej Karpathy @karpathy

4 years ago

Errr ok wow, I am shook by the new ConvMixer architecture openreview.net/forum?id=TVHS5… "the first model that achieves the elusive dual goals of 80%+ ImageNet top-1 accuracy while also fitting into a tweet" 😐

24 331 2K 0 384

Download Image

MMitchell @mmitchell_ai

4 years ago

I decided to give the new wave of AI companies a try. Care to join me @huggingface ? =)

👩‍💻 Paige Bailey @DynamicWebPaige

4 years ago

I decided to give the new wave of AI companies a try. Care to join me @huggingface ? =)

20 33 318 0 99

5 10 186 0 17

Gintare Karolina Dziugaite @gkdziugaite

4 years ago

1/ Welcome, Twitter, to my 1st tweet (!) a🧵on new work "Deep Learning on a Data Diet: Finding Important Examples Early in Training" arxiv.org/abs/2107.07075 We find that, at init, you can identify and prune a large % of DATA with NO effect on accuracy. w/@mansiege @SuryaGanguli

10 130 731 0 189

Download Image

Comédie-Française @ComedieFr

5 years ago

@DelmasEric « Les anciens, Monsieur, sont les anciens, et nous sommes les gens de maintenant. » Molière (Le Malade imaginaire, II, 6)

46 896 4K 0 55

Paragios, Nikos @AgoniGrammi

5 years ago

#spotlight for M. Sahasrabudhe @MICCAI2020's paper #attentionnetworks,#digitalpathology,#nucleisegmentation,#selfsupervisedlearning,#cancer Preprint: arxiv.org/abs/2007.08373 @centralesupelec,@Inserm,@Inria,@GustaveRoussy,@UnivParisSaclay,@PeterMacCC,@gzaziekenhuizen,@TheraPanacea

0 3 11 0 0

Download Image

Ross Wightman @wightmanr

5 years ago

Put a plain ol ResNet50 in front of the BiFPN layer from EfficientDet. Not shabby, 41.6 AP 0.50:.95 with the same GPU runtime traits as the EfficientDet-D1 with 39.4 AP. Faster than any of the R50 based R-CNN models w/ better mAP, single scale test, fairly basic train aug.