Cambios
En el instante 11 de octubre de 2025, 1:23:42 UTC,
-
Añadido recurso Efficient training of deep learning models through improved adaptive sampling a Efficient training of deep learning models through improved adaptive sampling
f | 1 | { | f | 1 | { |
2 | "author": "JI Avalos-L\u00f3pez, A Rojas-Dom\u00ednguez, M | 2 | "author": "JI Avalos-L\u00f3pez, A Rojas-Dom\u00ednguez, M | ||
3 | Ornelas-Rodr\u00edguez, M Carpio, ...", | 3 | Ornelas-Rodr\u00edguez, M Carpio, ...", | ||
4 | "author_email": null, | 4 | "author_email": null, | ||
5 | "creator_user_id": "a3da3ec9-3fd4-47a4-8d04-0a90b09614e0", | 5 | "creator_user_id": "a3da3ec9-3fd4-47a4-8d04-0a90b09614e0", | ||
6 | "extras": [ | 6 | "extras": [ | ||
7 | { | 7 | { | ||
8 | "key": "Publicaci\u00f3n", | 8 | "key": "Publicaci\u00f3n", | ||
9 | "value": "Cap\u00edtulo" | 9 | "value": "Cap\u00edtulo" | ||
10 | }, | 10 | }, | ||
11 | { | 11 | { | ||
12 | "key": "Tipo", | 12 | "key": "Tipo", | ||
13 | "value": "Publicaci\u00f3n" | 13 | "value": "Publicaci\u00f3n" | ||
14 | } | 14 | } | ||
15 | ], | 15 | ], | ||
16 | "groups": [ | 16 | "groups": [ | ||
17 | { | 17 | { | ||
18 | "description": "Este grupo integra las publicaciones | 18 | "description": "Este grupo integra las publicaciones | ||
19 | acad\u00e9micas derivadas de los proyectos de investigaci\u00f3n del | 19 | acad\u00e9micas derivadas de los proyectos de investigaci\u00f3n del | ||
20 | Observatorio Metropolitano CentroGeo. Incluye art\u00edculos | 20 | Observatorio Metropolitano CentroGeo. Incluye art\u00edculos | ||
21 | presentados en congresos nacionales e internacionales, manuscritos en | 21 | presentados en congresos nacionales e internacionales, manuscritos en | ||
22 | formato preprint, cap\u00edtulos de libro y trabajos publicados en | 22 | formato preprint, cap\u00edtulos de libro y trabajos publicados en | ||
23 | revistas cient\u00edficas especializadas. Estos materiales reflejan la | 23 | revistas cient\u00edficas especializadas. Estos materiales reflejan la | ||
24 | labor de investigaci\u00f3n, desarrollo metodol\u00f3gico y | 24 | labor de investigaci\u00f3n, desarrollo metodol\u00f3gico y | ||
25 | an\u00e1lisis territorial del observatorio, contribuyendo al avance | 25 | an\u00e1lisis territorial del observatorio, contribuyendo al avance | ||
26 | del conocimiento en temas urbanos, metropolitanos y geoespaciales.", | 26 | del conocimiento en temas urbanos, metropolitanos y geoespaciales.", | ||
27 | "display_name": "Publicaciones", | 27 | "display_name": "Publicaciones", | ||
28 | "id": "a15a6b77-ddf5-4594-acab-7e772938a5b0", | 28 | "id": "a15a6b77-ddf5-4594-acab-7e772938a5b0", | ||
29 | "image_display_url": "", | 29 | "image_display_url": "", | ||
30 | "name": "publicaciones", | 30 | "name": "publicaciones", | ||
31 | "title": "Publicaciones" | 31 | "title": "Publicaciones" | ||
32 | } | 32 | } | ||
33 | ], | 33 | ], | ||
34 | "id": "dd4b450d-e41a-40f9-b9ac-50ce286b734b", | 34 | "id": "dd4b450d-e41a-40f9-b9ac-50ce286b734b", | ||
35 | "isopen": false, | 35 | "isopen": false, | ||
36 | "license_id": null, | 36 | "license_id": null, | ||
37 | "license_title": null, | 37 | "license_title": null, | ||
38 | "maintainer": null, | 38 | "maintainer": null, | ||
39 | "maintainer_email": null, | 39 | "maintainer_email": null, | ||
40 | "metadata_created": "2025-10-11T01:23:41.807764", | 40 | "metadata_created": "2025-10-11T01:23:41.807764", | ||
n | 41 | "metadata_modified": "2025-10-11T01:23:41.807773", | n | 41 | "metadata_modified": "2025-10-11T01:23:42.330841", |
42 | "name": | 42 | "name": | ||
43 | deep-learning-models-through-improved-adaptive-sampling-5aadeb713593", | 43 | deep-learning-models-through-improved-adaptive-sampling-5aadeb713593", | ||
44 | "notes": "Training of Deep Neural Networks (DNNs) is very | 44 | "notes": "Training of Deep Neural Networks (DNNs) is very | ||
45 | computationally demanding and resources are typically spent on | 45 | computationally demanding and resources are typically spent on | ||
46 | training-instances that do not provide the most benefit to a | 46 | training-instances that do not provide the most benefit to a | ||
47 | network\u2019s learning; instead, the most relevant instances should | 47 | network\u2019s learning; instead, the most relevant instances should | ||
48 | be prioritized during training. Herein we present an improved version | 48 | be prioritized during training. Herein we present an improved version | ||
49 | of the Adaptive Sampling (AS) method (Gopal, 2016) extended for the | 49 | of the Adaptive Sampling (AS) method (Gopal, 2016) extended for the | ||
50 | training of DNNs. As our main contribution we formulate a probability | 50 | training of DNNs. As our main contribution we formulate a probability | ||
51 | distribution for data instances that minimizes the variance of the | 51 | distribution for data instances that minimizes the variance of the | ||
52 | gradient-norms w.r.t. the network\u2019s loss function. Said | 52 | gradient-norms w.r.t. the network\u2019s loss function. Said | ||
53 | distribution is combined with the optimal distribution for the data | 53 | distribution is combined with the optimal distribution for the data | ||
54 | classes previously derived by Gopal and the improved AS is used to | 54 | classes previously derived by Gopal and the improved AS is used to | ||
55 | replace uniform sampling with the objective of accelerating the | 55 | replace uniform sampling with the objective of accelerating the | ||
56 | training of DNNs. Our proposal is comparatively evaluated against | 56 | training of DNNs. Our proposal is comparatively evaluated against | ||
57 | uniform sampling and against Online Batch Selection (Loshchilov & | 57 | uniform sampling and against Online Batch Selection (Loshchilov & | ||
58 | Hutter, 2015). Results from training a Convolutional Neural Network on | 58 | Hutter, 2015). Results from training a Convolutional Neural Network on | ||
59 | the MNIST dataset with the Adadelta and Adam optimizers over different | 59 | the MNIST dataset with the Adadelta and Adam optimizers over different | ||
60 | training batch-sizes show the effectiveness and superiority of our | 60 | training batch-sizes show the effectiveness and superiority of our | ||
61 | proposal.", | 61 | proposal.", | ||
n | 62 | "num_resources": 0, | n | 62 | "num_resources": 1, |
63 | "num_tags": 17, | 63 | "num_tags": 17, | ||
64 | "organization": { | 64 | "organization": { | ||
65 | "approval_status": "approved", | 65 | "approval_status": "approved", | ||
66 | "created": "2022-05-19T00:10:30.480393", | 66 | "created": "2022-05-19T00:10:30.480393", | ||
67 | "description": "Observatorio Metropolitano CentroGeo", | 67 | "description": "Observatorio Metropolitano CentroGeo", | ||
68 | "id": "b3b3a79d-748a-4464-9471-732b6c74ec53", | 68 | "id": "b3b3a79d-748a-4464-9471-732b6c74ec53", | ||
69 | "image_url": | 69 | "image_url": | ||
70 | "2022-05-19-001030.456616FullColor1280x1024LogoOnly.png", | 70 | "2022-05-19-001030.456616FullColor1280x1024LogoOnly.png", | ||
71 | "is_organization": true, | 71 | "is_organization": true, | ||
72 | "name": "observatorio-metropolitano-centrogeo", | 72 | "name": "observatorio-metropolitano-centrogeo", | ||
73 | "state": "active", | 73 | "state": "active", | ||
74 | "title": "Observatorio Metropolitano CentroGeo", | 74 | "title": "Observatorio Metropolitano CentroGeo", | ||
75 | "type": "organization" | 75 | "type": "organization" | ||
76 | }, | 76 | }, | ||
77 | "owner_org": "b3b3a79d-748a-4464-9471-732b6c74ec53", | 77 | "owner_org": "b3b3a79d-748a-4464-9471-732b6c74ec53", | ||
78 | "private": false, | 78 | "private": false, | ||
79 | "relationships_as_object": [], | 79 | "relationships_as_object": [], | ||
80 | "relationships_as_subject": [], | 80 | "relationships_as_subject": [], | ||
t | 81 | "resources": [], | t | 81 | "resources": [ |
82 | { | ||||
83 | "cache_last_updated": null, | ||||
84 | "cache_url": null, | ||||
85 | "created": "2025-10-11T01:23:42.411079", | ||||
86 | "datastore_active": false, | ||||
87 | "description": "Training of Deep Neural Networks (DNNs) is very | ||||
88 | computationally demanding and resources are typically spent on | ||||
89 | training-instances that do not provide the most benefit to a | ||||
90 | network\u2019s learning; instead, the most relevant instances should | ||||
91 | be prioritized during training. Herein we present an improved version | ||||
92 | of the Adaptive Sampling (AS) method (Gopal, 2016) extended for the | ||||
93 | training of DNNs. As our main contribution we formulate a probability | ||||
94 | distribution for data instances that minimizes the variance of the | ||||
95 | gradient-norms w.r.t. the network\u2019s loss function. Said | ||||
96 | distribution is combined with the optimal distribution for the data | ||||
97 | classes previously derived by Gopal and the improved AS is used to | ||||
98 | replace uniform sampling with the objective of accelerating the | ||||
99 | training of DNNs. Our proposal is comparatively evaluated against | ||||
100 | uniform sampling and against Online Batch Selection (Loshchilov & | ||||
101 | Hutter, 2015). Results from training a Convolutional Neural Network on | ||||
102 | the MNIST dataset with the Adadelta and Adam optimizers over different | ||||
103 | training batch-sizes show the effectiveness and superiority of our | ||||
104 | proposal.", | ||||
105 | "format": "HTML", | ||||
106 | "hash": "", | ||||
107 | "id": "25abfb23-4d55-4f3b-bc2e-e1c0efdc392d", | ||||
108 | "last_modified": null, | ||||
109 | "metadata_modified": "2025-10-11T01:23:42.334426", | ||||
110 | "mimetype": null, | ||||
111 | "mimetype_inner": null, | ||||
112 | "name": "Efficient training of deep learning models through | ||||
113 | improved adaptive sampling", | ||||
114 | "package_id": "dd4b450d-e41a-40f9-b9ac-50ce286b734b", | ||||
115 | "position": 0, | ||||
116 | "resource_type": null, | ||||
117 | "size": null, | ||||
118 | "state": "active", | ||||
119 | "url": "https://doi.org/10.1007/978-3-030-77004-4_14", | ||||
120 | "url_type": null | ||||
121 | } | ||||
122 | ], | ||||
82 | "state": "active", | 123 | "state": "active", | ||
83 | "tags": [ | 124 | "tags": [ | ||
84 | { | 125 | { | ||
85 | "display_name": "adaptive-sampling", | 126 | "display_name": "adaptive-sampling", | ||
86 | "id": "8225c826-dbae-409d-9a77-db998c804b0b", | 127 | "id": "8225c826-dbae-409d-9a77-db998c804b0b", | ||
87 | "name": "adaptive-sampling", | 128 | "name": "adaptive-sampling", | ||
88 | "state": "active", | 129 | "state": "active", | ||
89 | "vocabulary_id": null | 130 | "vocabulary_id": null | ||
90 | }, | 131 | }, | ||
91 | { | 132 | { | ||
92 | "display_name": "artificial-intelligence", | 133 | "display_name": "artificial-intelligence", | ||
93 | "id": "47cb06d3-c2a4-42af-b633-03eea6181083", | 134 | "id": "47cb06d3-c2a4-42af-b633-03eea6181083", | ||
94 | "name": "artificial-intelligence", | 135 | "name": "artificial-intelligence", | ||
95 | "state": "active", | 136 | "state": "active", | ||
96 | "vocabulary_id": null | 137 | "vocabulary_id": null | ||
97 | }, | 138 | }, | ||
98 | { | 139 | { | ||
99 | "display_name": "artificial-neural-network", | 140 | "display_name": "artificial-neural-network", | ||
100 | "id": "d63d29ea-7b70-4ed5-83cc-66f3b9586c1c", | 141 | "id": "d63d29ea-7b70-4ed5-83cc-66f3b9586c1c", | ||
101 | "name": "artificial-neural-network", | 142 | "name": "artificial-neural-network", | ||
102 | "state": "active", | 143 | "state": "active", | ||
103 | "vocabulary_id": null | 144 | "vocabulary_id": null | ||
104 | }, | 145 | }, | ||
105 | { | 146 | { | ||
106 | "display_name": "computer-science", | 147 | "display_name": "computer-science", | ||
107 | "id": "29cae056-cd7e-43f7-be5b-b25869a3fbf2", | 148 | "id": "29cae056-cd7e-43f7-be5b-b25869a3fbf2", | ||
108 | "name": "computer-science", | 149 | "name": "computer-science", | ||
109 | "state": "active", | 150 | "state": "active", | ||
110 | "vocabulary_id": null | 151 | "vocabulary_id": null | ||
111 | }, | 152 | }, | ||
112 | { | 153 | { | ||
113 | "display_name": "convolutional-neural-network", | 154 | "display_name": "convolutional-neural-network", | ||
114 | "id": "b08eff8e-b411-4c56-ab0d-3ef47b79699c", | 155 | "id": "b08eff8e-b411-4c56-ab0d-3ef47b79699c", | ||
115 | "name": "convolutional-neural-network", | 156 | "name": "convolutional-neural-network", | ||
116 | "state": "active", | 157 | "state": "active", | ||
117 | "vocabulary_id": null | 158 | "vocabulary_id": null | ||
118 | }, | 159 | }, | ||
119 | { | 160 | { | ||
120 | "display_name": "deep-learning", | 161 | "display_name": "deep-learning", | ||
121 | "id": "2374322a-d83b-4c03-8b75-ba1d0b75384e", | 162 | "id": "2374322a-d83b-4c03-8b75-ba1d0b75384e", | ||
122 | "name": "deep-learning", | 163 | "name": "deep-learning", | ||
123 | "state": "active", | 164 | "state": "active", | ||
124 | "vocabulary_id": null | 165 | "vocabulary_id": null | ||
125 | }, | 166 | }, | ||
126 | { | 167 | { | ||
127 | "display_name": "deep-neural-networks", | 168 | "display_name": "deep-neural-networks", | ||
128 | "id": "4f23ca30-4b12-401d-a93c-958cf047df6e", | 169 | "id": "4f23ca30-4b12-401d-a93c-958cf047df6e", | ||
129 | "name": "deep-neural-networks", | 170 | "name": "deep-neural-networks", | ||
130 | "state": "active", | 171 | "state": "active", | ||
131 | "vocabulary_id": null | 172 | "vocabulary_id": null | ||
132 | }, | 173 | }, | ||
133 | { | 174 | { | ||
134 | "display_name": "function-biology", | 175 | "display_name": "function-biology", | ||
135 | "id": "50851f79-cad9-4419-99d1-daf7ff2dcf76", | 176 | "id": "50851f79-cad9-4419-99d1-daf7ff2dcf76", | ||
136 | "name": "function-biology", | 177 | "name": "function-biology", | ||
137 | "state": "active", | 178 | "state": "active", | ||
138 | "vocabulary_id": null | 179 | "vocabulary_id": null | ||
139 | }, | 180 | }, | ||
140 | { | 181 | { | ||
141 | "display_name": "importance-sampling", | 182 | "display_name": "importance-sampling", | ||
142 | "id": "5ab61561-5592-434e-9870-f97f65dd4378", | 183 | "id": "5ab61561-5592-434e-9870-f97f65dd4378", | ||
143 | "name": "importance-sampling", | 184 | "name": "importance-sampling", | ||
144 | "state": "active", | 185 | "state": "active", | ||
145 | "vocabulary_id": null | 186 | "vocabulary_id": null | ||
146 | }, | 187 | }, | ||
147 | { | 188 | { | ||
148 | "display_name": "machine-learning", | 189 | "display_name": "machine-learning", | ||
149 | "id": "99557dd8-2914-4e28-8278-7a042cfb23d0", | 190 | "id": "99557dd8-2914-4e28-8278-7a042cfb23d0", | ||
150 | "name": "machine-learning", | 191 | "name": "machine-learning", | ||
151 | "state": "active", | 192 | "state": "active", | ||
152 | "vocabulary_id": null | 193 | "vocabulary_id": null | ||
153 | }, | 194 | }, | ||
154 | { | 195 | { | ||
155 | "display_name": "mnist-database", | 196 | "display_name": "mnist-database", | ||
156 | "id": "2376b903-8a68-4a42-89f1-fd576aea48a2", | 197 | "id": "2376b903-8a68-4a42-89f1-fd576aea48a2", | ||
157 | "name": "mnist-database", | 198 | "name": "mnist-database", | ||
158 | "state": "active", | 199 | "state": "active", | ||
159 | "vocabulary_id": null | 200 | "vocabulary_id": null | ||
160 | }, | 201 | }, | ||
161 | { | 202 | { | ||
162 | "display_name": "sampling-signal-processing", | 203 | "display_name": "sampling-signal-processing", | ||
163 | "id": "ecc43bd8-306b-47f5-a170-175ba948e225", | 204 | "id": "ecc43bd8-306b-47f5-a170-175ba948e225", | ||
164 | "name": "sampling-signal-processing", | 205 | "name": "sampling-signal-processing", | ||
165 | "state": "active", | 206 | "state": "active", | ||
166 | "vocabulary_id": null | 207 | "vocabulary_id": null | ||
167 | }, | 208 | }, | ||
168 | { | 209 | { | ||
169 | "display_name": "selection-genetic-algorithm", | 210 | "display_name": "selection-genetic-algorithm", | ||
170 | "id": "5ae19ac6-8758-4ff8-ab84-a9b392758848", | 211 | "id": "5ae19ac6-8758-4ff8-ab84-a9b392758848", | ||
171 | "name": "selection-genetic-algorithm", | 212 | "name": "selection-genetic-algorithm", | ||
172 | "state": "active", | 213 | "state": "active", | ||
173 | "vocabulary_id": null | 214 | "vocabulary_id": null | ||
174 | }, | 215 | }, | ||
175 | { | 216 | { | ||
176 | "display_name": "stochastic-gradient-descent", | 217 | "display_name": "stochastic-gradient-descent", | ||
177 | "id": "128418a8-7e59-4368-83ea-992a4c9fb97f", | 218 | "id": "128418a8-7e59-4368-83ea-992a4c9fb97f", | ||
178 | "name": "stochastic-gradient-descent", | 219 | "name": "stochastic-gradient-descent", | ||
179 | "state": "active", | 220 | "state": "active", | ||
180 | "vocabulary_id": null | 221 | "vocabulary_id": null | ||
181 | }, | 222 | }, | ||
182 | { | 223 | { | ||
183 | "display_name": "training-meteorology", | 224 | "display_name": "training-meteorology", | ||
184 | "id": "5278a683-25af-4654-9242-1566903f1da0", | 225 | "id": "5278a683-25af-4654-9242-1566903f1da0", | ||
185 | "name": "training-meteorology", | 226 | "name": "training-meteorology", | ||
186 | "state": "active", | 227 | "state": "active", | ||
187 | "vocabulary_id": null | 228 | "vocabulary_id": null | ||
188 | }, | 229 | }, | ||
189 | { | 230 | { | ||
190 | "display_name": "training-set", | 231 | "display_name": "training-set", | ||
191 | "id": "33d3b3b3-3c56-4b13-8163-1582fd1678d1", | 232 | "id": "33d3b3b3-3c56-4b13-8163-1582fd1678d1", | ||
192 | "name": "training-set", | 233 | "name": "training-set", | ||
193 | "state": "active", | 234 | "state": "active", | ||
194 | "vocabulary_id": null | 235 | "vocabulary_id": null | ||
195 | }, | 236 | }, | ||
196 | { | 237 | { | ||
197 | "display_name": "variance-accounting", | 238 | "display_name": "variance-accounting", | ||
198 | "id": "ba557cbb-58d4-45a0-900c-d6fa002336ad", | 239 | "id": "ba557cbb-58d4-45a0-900c-d6fa002336ad", | ||
199 | "name": "variance-accounting", | 240 | "name": "variance-accounting", | ||
200 | "state": "active", | 241 | "state": "active", | ||
201 | "vocabulary_id": null | 242 | "vocabulary_id": null | ||
202 | } | 243 | } | ||
203 | ], | 244 | ], | ||
204 | "title": "Efficient training of deep learning models through | 245 | "title": "Efficient training of deep learning models through | ||
205 | improved adaptive sampling", | 246 | improved adaptive sampling", | ||
206 | "type": "dataset", | 247 | "type": "dataset", | ||
207 | "url": "https://doi.org/10.1007/978-3-030-77004-4_14", | 248 | "url": "https://doi.org/10.1007/978-3-030-77004-4_14", | ||
208 | "version": null | 249 | "version": null | ||
209 | } | 250 | } |