Module data
 Share
The version of the browser you are using is no longer supported. Please upgrade to a supported browser.Dismiss

Comment only
 
 
ABCDEFGHIJKLMNOPQRSTUVWXYZAAABAC
1
ArchitectureCategoryTotal # modules
# gradient descent modules
# meta-modulesModule detailsUncertaintiesSource
2
An end-to-end neural network that takes in all sense data and outputs motor plans11
3
AlphaGoGame32Policy net; value net; MCTSDo two distinct neural nets count as two modules?
4
AlphaGo ZeroGame21Neural net; MCTS
5
AlphaZeroGame21Neural net; MCTS
6
AlphaStarGame1 to 31Big net (transformer input + LSTM core + policy head), and unclear whether the “centralised value baseline” is another moduleAre the modules only referring to the final architecture, without taking the training process into account?
https://www.youtube.com/watch?v=3N9phq_yZP0&t=1478s
7
AlphaFoldScience4/5 or 8/9[Distance net; angle net; distribution score; aggregate-distance-scoring net] + [generative net XOR gradient descent]How to interpret the fact that they used multiple methods? And should the act of choosing between methods be seen as yet another module?
8
Neural Architecture SearchArchitecture search11Controller RNNShould the child networks also be counted?
9
AlexNetImage recognition11Deep conv net
10
Visualizing and Understanding Conv netsImage recognition11(Just uses ImageNet)
11
GoogleNetImage recognition1
12
Seq2Seq11Encoding and decoding LSTM, but the input to the second is not human interpretable so only one moduleShould encoding and decoding actually be treated this way?
13
Neural Machine Translation2-32-3Encoder LSTM, decoder LSTM, and attention module which receives human-interpretable output from the decoderShould encoder and decoder be treated as one or two modules, given presence of attention module?
Is the softmax another module? Is its input interpretable?
14
VGG11Very deep conv net
15
XceptionImage recognitionHe has things he calls “inception modules”?
16
Deep Speech 211Deep conv+recurrent net
17
DQNGame11Deep Q networkShould the initial preprocessing filter be included?
And is there some exploration noise when choosing from the outputted Q-values? (I presume e.g. such softmax policies are not treated as modules)
18
Auto keras
19
20
This spreadsheet was compiled to provide an outside view for the Metaculus AI questions: "By end of 2021, how many modules will the state-of-the-art AI system for architecture-search for image recognition have?"
21
(You can sign up to be a beta user of Metaculus AI here.)
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
Loading...