groar

9 ms·

31.
Bitcoin crossed the $10k mark again(cnbc.com)
4 points by groar 6y ago|0 comments
32.
YOLOv4: Optimal Speed and Accuracy of Object Detection(arxiv.org)
234 points by groar 6y ago|54 comments
33.
The Cost of Training NLP Models: A Concise Overview(arxiv.org)
2 points by groar 6y ago|0 comments
34.
by groar 6y ago
Agreed, my initial title is not accurate. It should say "finds errors in 20% of annotations".
35.
by groar 6y ago
That is true up to a certain point (for instance, in my experience, having bounding boxes that are not pixel-perfect acts as a regularizer), but there is also a good chance that you are mislabelling edge cases, situations that happen rarely
36.
by groar 6y ago
Yes agreed with that ! I can't change the title unfortunately
37.
by groar 6y ago
If I understand correctly they actually did not change the test set.
38.
by groar 6y ago
Using simple techniques, they found out that popular open source datasets like VOC or COCO contain up to 20% annotation errors in. By manually correcting those errors, they got an average error reduction of 5% for state-of-the-art computer
39.
Cleaning algorithm finds 20% of errors in major image recognition datasets(deepomatic.com)
221 points by groar 6y ago|72 comments
40.
Deep Molecular Programming(arxiv.org)
130 points by groar 7y ago|11 comments
41.
A chrome extension that quantifies how much you touch your face(chrome.google.com)
1 points by groar 7y ago|0 comments
42.
Neural Networks Are Surprisingly Modular(arxiv.org)
30 points by groar 7y ago|0 comments
43.
by groar 7y ago
Sure, in terms of expressivity, you can obtain much better results with a CNN. But very often, it is done at the cost of computational efficiency: SIFT descriptors are "easy" to compute.
44.
Generating MIDI melody from lyrics using LSTM-GANs(github.com)
120 points by groar 7y ago|25 comments
45.
The patent on SIFT expired yesterday(patents.google.com)
212 points by groar 7y ago|67 comments
46.
Google patent on Deep Neural Networks now applicable(patents.google.com)
3 points by groar 7y ago|1 comments
47.
Towards Brain-Computer Interfaces for Drone Swarm Control(arxiv.org)
1 points by groar 7y ago|0 comments
48.
Abstract Interpretation for Provably Robust Neural Networks [pdf](proceedings.mlr.press)
2 points by groar 7y ago|0 comments
49.
StarGAN v2: Image Synthesis for Multiple Domains(github.com)
14 points by groar 7y ago|2 comments
50.
Multicellular Biological Deep Neural Nets(arxiv.org)
1 points by groar 7y ago|0 comments
51.
Microsoft Research’s Lin Xiao Earns Test of Time Award at NeurIPS(venturebeat.com)
2 points by groar 7y ago|0 comments
52.
Handwriting-Based Gender Classification Using Deep Neural Networks(arxiv.org)
1 points by groar 7y ago|0 comments
53.
Variational Physics-Informed Neural Nets: Solving Partial Differential Equations(arxiv.org)
1 points by groar 7y ago|0 comments
54.
Mesh R-CNN: predicting 3D shapes(arxiv.org)
2 points by groar 7y ago|0 comments
55.
SinGAN: Learning a Generative Model from a Single Natural Image(arxiv.org)
77 points by groar 7y ago|9 comments
56.
Logic-Inspired Deep Neural Networks(arxiv.org)
4 points by groar 7y ago|0 comments
57.
POPL 2020 Accepted Papers(popl20.sigplan.org)
1 points by groar 7y ago|0 comments
58.
by groar 7y ago
This paper titled "Backpropagation in the Simply Typed Lambda-calculus with Linear Negation" describes how to generalize the backpropagation algorithm used for computational graphs (like in PyTorch or TF) to the simply-typed lambd
59.
Extending Backpropagation to Functional Programs(arxiv.org)
2 points by groar 7y ago|1 comments
60.
The unreasonable effectiveness of the forget gate(arxiv.org)
3 points by groar 8y ago|0 comments

More ›