Sangheum retweeted
Apr 25
contrasting the # of authors/participants vs. # of sponsors, i think we see the directionality of brain drain and also the AI market size.
one notable entry here is south korea: pretty much brain drain only.
5
17
98
49,635
Sangheum retweeted
21 Jun 2025
i was finally convinced by @_sungmin_cha and @beopst to work on unlearning. the first we did (or i learned) together with sungjin and dason was to learn how people evaluate unlearning, and horror š±
here is our short writeup and report on what we found and what we propose as a remedy.
3
16
151
18,975
Most companies donāt need well-published academics. They need good engineers who make things work. Nice discussion: reddit.com/r/MachineLearningā¦
19
314
901
Sangheum retweeted
18 Oct 2017
Thanks to @Uber for open-sourcing Horovod! A library to speed-up distributed training with #TensorFlow
Learn hereā goo.gl/UXzLrT
3
225
468
Sangheum retweeted
18 Oct 2017
Newly proposed activation "Swish" outperforms ReLU and many other activation functions on multiple models/tasks. arxiv.org/abs/1710.05941
5
32
90
Sangheum retweeted
21 Sep 2017
Want to visualize what your ConvNet is learning? Blogpost summarizing a couple of approaches! reddit.com/r/MachineLearningā¦
1
22
65
Sangheum retweeted
16 Jun 2017
Variational Approaches for Auto-Encoding Generative Adversarial Networks. (arXiv:1706.04987v1 [stat.ML]) ift.tt/2sgM2hb
1
13
68
Sangheum retweeted
19 Jun 2017
L2 Regularization versus Batch and Weight Normalization. (arXiv:1706.05350v1 [cs.LG]) ift.tt/2spX4AM
7
49
Sangheum retweeted
29 May 2017
Bayesian GANs
arxiv.org/abs/1705.09558
Our new paper about posterior inference with GANs!
3
90
222
Sangheum retweeted
25 May 2017
Everything that Works Works Because it's Bayesian: Why Deep Nets Generalize? inference.vc/everything-thatā¦
4
11
Sangheum retweeted
22 May 2017
Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics arxiv.org/abs/1705.07115 #computervision #deeplearning
47
101
Sangheum retweeted
19 May 2017
TensorFlow 1.2rc0 is now out with the tf.contrib.data module for simpler and more powerful input pipelines! github.com/tensorflow/tensorā¦
1
70
135
Sangheum retweeted
26 Apr 2017
Deep Over-sampling Framework for Classifying Imbalanced Data. (arXiv:1704.07515v1 [cs.LG]) ift.tt/2qccF4j
3
8
Sangheum retweeted
17 Apr 2017
"Stochastic Gradient Descent as Approximate Bayesian Inference", S. Mandt et al. arxiv.org/abs/1704.04289 Unifying views should be tried more
2
5
Sangheum retweeted
4 Apr 2017
Notes on catastrophic forgetting and recent combating techniques.
theneuralperspective.com/201ā¦
24
70
Sangheum retweeted
3 Apr 2017
Super impressive image quality - "BEGAN: Boundary Equilibrium Generative Adversarial Networks," Berthelot et al: arxiv.org/abs/1703.10717
3
42
91
Deep Hybrid Nets. Fixing 1st convnet layer w/ predefined wavelet filters to increase efficiency generalize better. arxiv.org/abs/1703.08961
4
25
88
Sangheum retweeted
29 Mar 2017
Early Stopping without a Validation Set. (arXiv:1703.09580v1 [cs.LG]) ift.tt/2o4khbV
7
12