2019年、俺の読んだ論文50本全部解説(俺的ベスト3付き) 綿岡晃輝 @wataoka
https://qiita.com/wataoka/items/ae782defabc3706b5c93
<この項は書きかけです。順次追記します。>
はじめに
2017年, 2018年は深層学習の読書会に参加するにあたって、
「ゼロから作るDeep Learning ２自然言語処理編」読書会に参加する前に読んで置くとよい資料とプログラム
https://qiita.com/kaizen_nagoya/items/537b1810265bbbc70e73
docker(19) 言語処理100本ノックをdockerで。python覚えるのに最適。
https://qiita.com/kaizen_nagoya/items/7e7eb7c543e0c18438c4
docker(18) なぜdockerで機械学習するか 書籍・ソース一覧作成中 (目標100)
https://qiita.com/kaizen_nagoya/items/ddd12477544bf5ba85e2
資料を整理していました。2019年は、量子計算機に焦点を絞り、深層学習は少しほおったらかしにしてきました。
「量子アニーリングの基礎」を読む 第１日
https://qiita.com/kaizen_nagoya/items/2bc284faaf0f61278778
そんな後ろめたい気持ちで迎えた年末に、ちょうどねがってもない記事から論文を被験しています。
５０論文の単語帳作成
いつもは同じ領域の１００論文で作成する。固有名詞の幅も狭い。
今回は５０論文だが単語数は14302と数が多い。
文献、図表などの本文以外の文字処理上の課題があるかどうかは未確認。
単語帳を作成する趣旨は、自分の知らない単語はあらかじめ辞書などで調べておくためです。
４５年くらい前から、100ページくらいの文章を読むのに、記憶力が悪く、何度も同じ単語を辞書で引いていました。
PCを使うようになって、あらかじめ単語帳を作り、辞書で引いておく習慣にしています。
英語(18) 英語文献を入手したら
https://qiita.com/kaizen_nagoya/items/cb3d0e299c3c96370a31
短縮名(short name) 略号(abbreviation)
@iKaz888 さんのご意見に基づき作成しはじめました。
まだ全体の 20 10 4 2分の１くらいの作業量です。
過去に自分たちの作った資料で、複数担当で作ったら、資料の中で、R(red)とR(right)の２つの異なる意味で同じ記号の略号が存在したという反省に基づいています。
著者の違う複数の論文を読むときには、同じ記号で別の意味で使っている可能性を最大限調査するのが経験則です。
例えば、MCBNはMonte Carlo Bayesian networkかと思って読んでました。
| abbreviation | full spelling | |
|---|---|---|
| 3D | three dimensions | |
| aaai | Advancement of Artificial Intelligence | |
| ACC | accuracy | |
| ACM | Association for Computing Machinery | |
| ai | Artificial intelligence | |
| AMT | Amazon Mechanical Turk | |
| ARL | Army Research Laboratory | |
| AUC | area under the curve, area under the ROC curve | |
| BCA | Balanced Classification Accuracy | |
| BER | Bit Error Rate | |
| BMI | Body Mass Index | |
| BN | Batch Normalization | |
| CB | Class-Balanced | |
| CE | cross entropy | |
| CFI | Canada Foundation for Innovation | |
| CFP | Celebrities in Frontal Profile | |
| CFS | correlation-based feature selection | |
| cifar | Canadian Institute for Advanced Research | |
| CMC | Cumulative Match Characteristic | |
| cnn | Cellular neural network | |
| cnn(s) | Convolutional Neural Network(s) | |
| COMPAS | Correctional Offender Management Profiling for Alternative Sanctions | |
| CRPS | Continuous Ranked Probability Score | |
| CRRA | Constant Relative Risk Aversion | |
| CSFS | cost-sensitive feature selection | |
| cTMU | Conditional Transient Mapping Unit | |
| CUDO | Constant Uncertainty Dropout | |
| CUHK | Chinese University of Hong Kong | http://www.cuhk.edu.hk/english/index.html |
| CV | cross-validation | |
| cvpr | Conference on Computer Vision and Pattern Recognition | |
| DA | domain adaptation | |
| DAE | Discrimination-Aware Ensemble | |
| dann | Domain Adversarial Neural Network | |
| DAQL | Density Aware Quadruplet Loss | |
| DARPA | Defense Advanced Research Projects Agency | |
| DATL | Density Aware Triplet Loss | |
| DI | Disparate Impact | |
| DiF | Diversity in Faces | |
| DL | Deep Learning | |
| DM | Deep Metric Learning | |
| DM's | decision-maker's | |
| DNN(s) | deep neural network(s) | |
| DOS | deep oversampling framework | |
| DR | Diabetic Retinopathy | |
| DRO | distributionally robust optimization | |
| ECOA | Equal Credit Opportunity Acts | |
| EDE | Equally Distributed Equivalent | |
| EEOC | US Equal Employment Opportunity Commission | |
| EGT | Evolutionary Game Theory | |
| EO | Equal Opportunity | |
| ERM | empirical risk minimization | |
| EUS | Evolutionary Undersampling | |
| FAD | Fair adversarial discriminative | |
| FAIR | Facebook AI Research | |
| FERM | Fair Empirical Risk Minimization | |
| FF | frontal-frontal | |
| FGSM | Fast Gradient Sign Method | |
| FHA | Fair Housing Acts | |
| FL | Focal Loss | |
| FMMR | Fairness Maximal Marginal Relevance | |
| FN | false negative | |
| FNB | Fair Naive-Bayes | |
| FNR | false negative rate | |
| FP | false positive | |
| FP | frontal-profile | |
| FPN | Feature Pyramid Network | |
| FPR | false positive rates | |
| FRQNT | Fonds de recherche du Quebec Nature et technologies | |
| GAMO | Generative Adversarial Minority Oversampling | |
| GAN(s) | Generative adversarial network(s) | |
| GM | geometric mean | |
| GPU | Graphics Processing Unit | |
| GRL | gradient reversal layer | |
| GRU | gated recurrent unit | |
| http(s) | hyper text transfer protocol (secure) | |
| HUD. | Housing and Urban Development | |
| I-FGSM | Iterative FGSM | |
| icml | International Conference on Machine Learning | |
| IEEE | Institute of Electrical and Electronics Engineers | |
| ITFR | Information-Theoretic Feature Ranking | |
| KDE | Kernel Density Estimate | |
| KL | Kullback-Leibler | |
| KNN | k-nearest neighbor | |
| LDS | local distributional smoothness | |
| LFR | Learned Fair Representations | |
| LFW | labeled faces in the wild | |
| LIBSVM1 | https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets | |
| LR | Logistic Regression | |
| LR | linear predictor | |
| MCAR | Missing Completely At Random | |
| MCBN | Monte Carlo Batch Normalization | |
| MCC | Matthews correlation coefficient | |
| MCDO | Monte Carlo Dropout | |
| MD | minibatch diversity | |
| MDP | Markov decision process | |
| MIFR | Mutual Information-based Fair Representations | |
| MINECO | Spanish Ministry of Economy and Competitiveness | |
| minist | Modified National Institute of Standards and Technology | |
| MIT | Massachusetts Institute of Technology | http://www.mit.edu/ |
| ML | machine learning | |
| MLReliefF | Multi-Label ReliefF | |
| MMD | maximum mean discrepancy | |
| MMR | Maximal Marginal Relevance | |
| MNF | Multiplicative Normalizing Flows for variational Bayesian networks | |
| mRMR | Minimum Redundancy Maximum Relevance | |
| MSA | minimum subgroup accuracy | |
| mSDA | marginalized Stacked Autoencoders | |
| MUG | Multiplayer version of the Ultimatum Game | |
| NBS | naive Bayes | |
| NCCL | NVIDIA Collective Communication Library | |
| NCFS | Non-Convex Feature Learning | |
| NGO | Non Governmental Organization | |
| nips | Neural Information Processing Systems | |
| nn | neural network | |
| NSERC. | Natural Sciences and Engineering Research Council of Canada | |
| NSF | National Science Foundation | |
| NYPD | New York Police Devision | |
| OHEM | online hard example mining | |
| OPT | optimization | |
| OTOS | Optimal Transport for OverSampling | |
| PAD | Proxy A-distances | |
| PBP | Probabilistic backpropagation | |
| PCA | Principle Component Analysis | |
| PCG | primary condition group | |
| PLL | Log likelihood | |
| POF | price of fairness | |
| PRID | Progesterone Releasing Intravaginal Device | |
| RBA | reducing bias amplification | |
| ReLU | rectified linear unit | |
| resnet | Residual neural network | |
| ResNets | Residual neural network | |
| RFS | Robust Feature Selection | |
| RKHS | Reproducing Kernel Hilbert Space | |
| RLR | Regularized Logistic Regression | |
| ROC | Reject Option based Classification | |
| RPT | random perturbation training | |
| RUS | random majority undersampling | |
| SE | Squeeze and Excitation | |
| SEAT | Sentence Encoder Association Test | |
| SGD | stochastic gradient descent | |
| SGM | Sigmoid | |
| SM | Softmax | |
| SMOTE | Synthetic Minority Oversampling TEchnique | |
| SSL | semi-supervised learning | |
| STEM | Science, Technology, Engineering, and Math | |
| SUM. | sample-level uncertainty modeling | |
| SUNDIAL | SUrvey Network for Deep Imaging Analysis and Learning | |
| SVHN | Street View House Numbers | |
| svm | Support vector machine, | |
| SVM-RFE | support vector machine recursive feature elimination | |
| TN | true negative | |
| TP | true positive | |
| UCI | University of Californiaj Irvine | |
| UG | Ultimatum Game | |
| ui | user interface | |
| UMM | uncertainty-based margin | |
| US/USA | Unites States of America | |
| USD | US Dollar | |
| VAE | Variational Auto Encoders | |
| VAEs | virtual adversarial examples | |
| VAT | virtual adversarial training | |
| VI | variational inference | |
| VIPeR | viewpoint invariant pedestrian recognition | |
| WEAT | Word Embedding Association Test | |
| YTF | YouTube Faces |
主な同一表現の別の言葉は、
CA: California
CA: Canada
FP: false positive
FP: frontal-profile
LR: Logistic Regression
LR : linear predictor , Learning Fair Representations, Richard Zemel, 2013
VAE: Variational Auto Encoders
VAEs: virtual adversarial examples
英語(2) まぎらわしい、間違えやすい、行き違いの多い略号worst 10(候補24)
https://qiita.com/kaizen_nagoya/items/0bff5dbb72208053489b
英語(3) 仮説・検証(88)用語の衝突(用語・用例募集中)
https://qiita.com/kaizen_nagoya/items/6a8eb7ffaa45eeb16624
LRを
https://en.wikipedia.org/wiki/LR#Science%2C_technology_and_mathematics
に追記してみました。
固有名詞用語集(onomasticon)
まだ全体の 50 45 10 分の１くらいの作業量です。
英語(24) アン(-an)の部屋（28人名から学ぶ数学・物理）
https://qiita.com/kaizen_nagoya/items/e02cbe23b96d5fb96aa1
単語出現頻度(word count)
型(type): 略号（abbreviation）, 固有名詞（onomasticon）,変数(variable)
| word | 日本語 | count | type |
|---|---|---|---|
| the | その | 21520 | |
| of | の | 10920 | |
| and | そして | 9426 | |
| a | a | 8575 | |
| to | に | 8345 | |
| in | に | 7392 | |
| is | は | 5409 | |
| we | 私達 | 5238 | |
| for | にとって | 5016 | |
| x | x | 4168 | variable |
| that | それ | 3919 | |
| on | に | 3366 | |
| with | と | 2940 | |
| i | 私 | 2732 | |
| as | なので | 2694 | |
| by | 沿って | 2365 | |
| this | この | 2288 | |
| t | t | 2259 | |
| f | f | 2209 | variable |
| s | s | 2147 | |
| are | は | 2055 | |
| y | y | 1924 | variable |
| be | なる | 1920 | |
| learning | 学習 | 1854 | |
| d | d | 1845 | |
| fairness | 公平 | 1704 | |
| p | p | 1703 | |
| data | 与件 | 1688 | |
| from | から | 1660 | |
| k | k | 1635 | |
| g | g | 1563 | |
| can | できる | 1556 | |
| our | 私たちの | 1492 | |
| m | m | 1453 | |
| j | j | 1431 | |
| an | an | 1428 | |
| training | 訓練 | 1407 | |
| class | 級 | 1374 | |
| b | b | 1352 | |
| r | r | 1287 | |
| loss | 損失 | 1283 | |
| h | h | 1277 | |
| c | c | 1261 | |
| z | z | 1203 | |
| model | 模型 | 1193 | |
| which | どれ | 1134 | |
| e | e | 1119 | |
| n | n | 1116 | |
| adversarial | 敵対者 | 1096 | |
| it | それ | 1063 | |
| et | ら | 990 | |
| not | じゃない | 988 | |
| al | al | 987 | |
| samples | 例 | 974 | |
| each | 各 | 918 | |
| l | l | 891 | |
| w | w | 884 | |
| or | または | 856 | |
| set | 設定する | 854 | |
| at | で | 844 | |
| all | すべて | 819 | |
| have | 持ってる | 797 | |
| classification | 分類 | 788 | |
| where | どこ | 776 | |
| using | を使用して | 755 | |
| one | 一 | 722 | |
| if | もし | 696 | |
| between | の間に | 693 | |
| results | 結果 | 692 | |
| fair | 公平 | 690 | |
| two | 二 | 689 | |
| u | あなたは | 663 | |
| algorithm | 解法 | 656 | |
| accuracy | 正確さ | 642 | |
| classifier | 分類器 | 642 | |
| these | これら | 641 | |
| dataset | 与件集合 | 639 | |
| more | もっと | 631 | |
| used | 中古 | 628 | |
| figure | 図 | 603 | |
| different | 違う | 591 | |
| also | また | 590 | |
| such | そのような | 588 | |
| when | いつ | 587 | |
| function | 関数 | 584 | |
| neural | 神経の | 578 | |
| examples | 例 | 571 | |
| then | それから | 570 | |
| q | q | 569 | |
| distribution | 分布 | 568 | |
| group | 集まり | 562 | |
| network | 通信網 | 558 | |
| over | 以上 | 551 | |
| deep | 深い | 542 | |
| v | v | 533 | |
| use | 使用する | 520 | |
| networks | 網 | 516 | |
| number | 数 | 515 | |
| performance | 性能 | 509 | |
| models | 模型 | 508 | |
| has | 持っている | 502 | |
| only | のみ | 502 | |
| than | より | 495 | |
| their | 彼らの | 491 | |
| based | 基づく | 490 | |
| input | 入力 | 485 | |
| feature | 特徴 | 482 | |
| problem | 問題 | 481 | |
| sample | 例 | 477 | |
| method | 方法 | 472 | |
| metric | 測定 | 467 | |
| will | 意志 | 458 | |
| methods | 方法 | 456 | |
| other | その他 | 456 | |
| given | 与えられた | 455 | |
| error | 誤り | 452 | |
| softmax | 軟最大 | 445 | |
| any | どれか | 441 | |
| machine | 機械 | 439 | |
| large | 大 | 435 | |
| classes | 級 | 434 | |
| domain | 領域 | 433 | |
| while | ながら | 427 | |
| work | 作業 | 427 | |
| sensitive | 敏感な | 425 | |
| both | どちらも | 424 | |
| information | 情報 | 421 | |
| but | だが | 418 | |
| conference | 会議 | 409 | |
| trained | 訓練された | 409 | |
| features | 特徴 | 405 | |
| show | 公演 | 405 | |
| rate | 割合 | 403 | |
| datasets | 与件集合 | 402 | |
| test | 試験 | 402 | |
| first | 最初 | 399 | |
| proposed | 提案された | 389 | |
| target | 目標 | 388 | |
| approach | やり方 | 383 | |
| arxiv | arxiv | 380 | onomasticon |
| its | その | 378 | |
| xi | xi | 376 | variable |
| case | 場合 | 373 | |
| section | 節 | 362 | |
| decision | 決定 | 360 | |
| linear | 線形 | 355 | |
| same | 同じ | 351 | |
| label | 名札 | 349 | |
| there | そこ | 349 | |
| however | しかしながら | 348 | |
| optimal | 最適な | 348 | |
| no | 番号 | 346 | |
| experiments | 実験 | 341 | |
| following | 以下 | 340 | |
| example | 例 | 339 | |
| table | 表 | 338 | |
| image | 画像 | 337 | |
| some | いくつか | 336 | |
| how | どうやって | 334 | |
| layer | 層 | 333 | |
| dnn | dnn | 332 | abbreviation |
| probability | 確率 | 328 | |
| distance | 距離 | 327 | |
| systems | 系 | 320 | |
| images | 画像 | 319 | |
| positive | 前向き | 319 | |
| consider | 考える | 318 | |
| values | 値 | 311 | |
| representations | 表現 | 310 | |
| may | 五月 | 309 | |
| size | 大きさ | 307 | |
| impact | 影響 | 305 | |
| note | 注意 | 305 | |
| log | 記録 | 297 | |
| value | 値 | 293 | |
| they | 彼ら | 291 | |
| time | 時間 | 291 | |
| output | 出力 | 288 | |
| similar | 同様の | 288 | |
| into | に | 287 | |
| thus | かくして | 287 | |
| most | 最も | 286 | |
| pages | ページ数 | 286 | |
| space | 空間 | 286 | |
| so | そう | 285 | |
| recognition | 認識 | 283 | |
| cost | 費用 | 281 | |
| parameters | 引数 | 281 | |
| small | 小さい | 279 | |
| yi | yi | 278 | variable |
| was | だった | 275 | |
| measure | 測る | 273 | |
| let | させる | 269 | |
| disparate | 異なる | 267 | |
| do | 行う | 266 | |
| labels | 名札 | 263 | |
| discrimination | 差別 | 261 | |
| would | するだろう | 261 | |
| ieee | IEEE | 260 | abbreviation |
| bias | 偏り | 251 | |
| optimization | 最適化 | 251 | |
| points | 点 | 250 | |
| negative | 負 | 249 | |
| high | 高い | 248 | |
| max | 最大 | 248 | |
| source | 源 | 248 | |
| cifar | cifar | 246 | abbreviation |
| mean | 平均 | 246 | |
| algorithms | 解法 | 244 | |
| proceedings | 議事録 | 244 | |
| prediction | 予測 | 243 | |
| representation | 表現 | 242 | |
| new | 新着 | 241 | |
| random | 無作為 | 241 | |
| us | 我ら | 241 | |
| see | 見る | 240 | |
| vector | 有向量 | 240 | |
| does | する | 238 | |
| pr | pr | 238 | |
| research | 研究 | 237 | |
| under | 下 | 237 | |
| without | なしで | 237 | |
| o | o | 235 | |
| well | 上手 | 235 | |
| ml | ml | 234 | |
| uncertainty | 不確実性 | 232 | |
| mnist | mnist | 231 | abbreviation |
| original | 元の | 231 | |
| risk | 危険 | 231 | |
| individuals | 個人 | 230 | |
| theorem | 定理 | 230 | |
| objective | 目的 | 229 | |
| paper | 論文 | 229 | |
| since | 以来 | 229 | |
| defined | 定義済み | 228 | |
| minority | 少数 | 228 | |
| face | 面 | 226 | |
| batch | 束 | 225 | |
| constraints | 制約 | 225 | |
| sampling | 標本化 | 223 | |
| learned | 学んだ | 221 | |
| classifiers | 分類子 | 220 | |
| definition | 定義 | 220 | |
| individual | 個人 | 220 | |
| task | 仕事 | 220 | |
| gradient | 勾配 | 218 | |
| standard | 標準 | 218 | |
| bound | 行きの | 217 | |
| non | 非 | 217 | |
| been | されている | 215 | |
| groups | 集まり | 212 | |
| knowledge | 知識 | 212 | |
| were | だった | 210 | |
| attribute | 属性 | 209 | |
| better | より良い | 208 | |
| learn | 学ぶ | 208 | |
| parameter | 引数 | 207 | |
| international | 国際 | 206 | |
| because | なぜなら | 205 | |
| should | すべき | 203 | |
| imbalanced | 不均衡 | 201 | |
| adversary | 敵 | 200 | |
| respect | 尊敬 | 200 | |
| about | 約 | 199 | |
| fig | 図 | 197 | |
| proof | 証明 | 197 | |
| eq | eq | 196 | |
| many | たくさんの | 196 | |
| equal | 等しい | 194 | |
| train | 列車 | 194 | |
| architecture | 建築 | 193 | |
| here | ここに | 193 | |
| setting | 設定 | 193 | |
| shows | 示す | 192 | |
| min | 分 | 191 | |
| perturbation | 摂動 | 191 | |
| real | 現実 | 191 | |
| distributions | 分布 | 190 | |
| analysis | 分析 | 189 | |
| margin | 余白 | 189 | |
| constraint | 制約 | 187 | |
| even | でも | 186 | |
| functions | 関数 | 186 | |
| result | 結果 | 185 | |
| second | 第二 | 185 | |
| binary | 二進 | 184 | |
| now | 今 | 183 | |
| substitute | 代替 | 182 | |
| during | 中に | 181 | |
| make | 作る | 181 | |
| terms | 条項 | 181 | |
| th | 番目 | 181 | |
| against | に対して | 180 | |
| shown | 示す | 180 | |
| point | 点 | 179 | |
| step | 段階 | 178 | |
| supervised | 監督 | 178 | |
| through | 使って | 178 | |
| multi | 複数の | 177 | |
| computer | 計算機 | 176 | |
| parity | 一致 | 176 | |
| hypothesis | 仮説 | 175 | |
| wk | 週 | 175 | |
| larger | 大きい | 174 | |
| sets | 集合 | 174 | |
| statistical | 統計的 | 174 | |
| long | 長いです | 173 | |
| assume | 想定する | 172 | |
| state | 状態 | 172 | |
| transfer | 転送 | 172 | |
| above | 上 | 171 | |
| follows | 従う | 171 | |
| lower | 低い | 171 | |
| processing | 処理 | 171 | |
| single | 単一 | 171 | |
| corresponding | 対応する | 170 | |
| distillation | 蒸留 | 170 | |
| minibatch | 小束 | 170 | |
| study | 調査 | 170 | |
| across | に渡って | 169 | |
| framework | 枠組み | 169 | |
| imbalance | 不均衡 | 169 | |
| acm | acm | 168 | |
| human | 人間 | 168 | |
| vat | バット | 168 | |
| TRUE | 真 | 168 | |
| best | 最善 | 167 | |
| level | 水準 | 167 | |
| multiple | 複数 | 167 | |
| particular | 特定の | 167 | |
| term | 期間 | 166 | |
| biased | 偏った | 165 | |
| convolutional | たたみ込み | 164 | |
| due | 期限 | 164 | |
| layers | 層 | 163 | |
| average | 平均 | 162 | |
| define | 定義する | 161 | |
| denote | 示す | 161 | |
| opportunity | 機会 | 161 | |
| difference | 差 | 160 | |
| find | 見つける | 159 | |
| possible | 可能 | 159 | |
| process | 処理する | 159 | |
| after | 後 | 158 | |
| regression | 退行 | 158 | |
| selection | 選択 | 158 | |
| several | いくつかの | 158 | |
| provide | 提供する | 157 | |
| respectively | それぞれ | 157 | |
| good | 良い | 156 | |
| scale | 規模 | 156 | |
| very | 非常に | 155 | |
| validation | 検証 | 154 | |
| order | 注文 | 153 | |
| adaptation | 適応 | 152 | |
| synthetic | 合成 | 152 | |
| convex | 凸面 | 151 | |
| gender | 性別 | 151 | |
| pairs | 対 | 151 | |
| protected | 保護された | 151 | |
| therefore | したがって | 151 | |
| per | あたり | 150 | |
| achieve | 成し遂げる | 149 | |
| approaches | やり方 | 149 | |
| making | 作り | 149 | |
| regularization | 正則化 | 149 | |
| attributes | 属性 | 148 | |
| fixed | 修繕 | 148 | |
| gi | gi | 148 | |
| generated | 生成された | 147 | |
| predictions | 予測 | 147 | |
| dt | dt | 146 | |
| preprint | 前刷り | 145 | |
| ds | ds | 144 | |
| further | さらに | 144 | |
| lemma | 補題 | 144 | |
| system | 系 | 144 | |
| weight | 重量 | 144 | |
| cnn | cnn | 143 | abbreviation |
| instance | 実例 | 143 | |
| oracle | 神託 | 143 | |
| svm | svm | 143 | abbreviation |
| gf | GF | 142 | |
| higher | より高い | 142 | |
| instead | 代わりに | 142 | |
| specific | 明確な | 142 | |
| empirical | 経験的 | 141 | |
| measures | 対策 | 141 | |
| much | ずっと | 141 | |
| every | 毎回 | 140 | |
| general | 一般的な | 140 | |
| low | 低い | 140 | |
| metrics | 指標 | 140 | |
| tasks | 仕事 | 139 | |
| being | であること | 138 | |
| like | お気に入り | 138 | |
| expected | 期待される | 137 | |
| mixup | 混同 | 137 | |
| need | 必要 | 137 | |
| often | しばしば | 137 | |
| baseline | 基準線 | 136 | |
| erm | えーっと | 136 | |
| three | 三 | 136 | |
| constant | 絶え間ない | 135 | |
| way | 仕方 | 135 | |
| classi | classi | 134 | |
| effective | 効果的 | 134 | |
| inputs | 入力 | 134 | |
| lipschitz | lipschitz | 134 | onomasticon |
| utility | 効用 | 134 | |
| simple | 簡単な | 133 | |
| theoretical | 理論的 | 133 | |
| cross | 十字 | 132 | |
| less | もっと少なく | 132 | |
| resnet | resnet | 132 | abbreviation |
| via | 経由で | 132 | |
| pp | pp | 131 | |
| them | それら | 131 | |
| university | 大学 | 131 | |
| up | 上に | 131 | |
| appendix | 付録 | 130 | |
| aware | 気がついて | 130 | |
| those | それら | 130 | |
| could | たぶん......だろう | 129 | |
| decisions | 決定 | 129 | |
| dm | dm | 129 | |
| propose | 提案する | 129 | |
| solution | 解決 | 129 | |
| evaluation | 評価 | 128 | |
| obtain | 入手します | 128 | |
| dnns | dnns | 127 | abbreviation |
| object | 対象 | 127 | |
| out | でる | 127 | |
| procedure | 手順 | 127 | |
| re | 再 | 127 | |
| another | 別の | 126 | |
| compared | 比較した | 126 | |
| comparison | 比較 | 126 | |
| previous | 前 | 126 | |
| race | 人種 | 126 | |
| balanced | 均衡のとれた | 125 | |
| design | 設計 | 125 | |
| epochs | 世紀 | 125 | |
| matrix | 行列 | 124 | |
| existing | 既存の | 123 | |
| search | 探す | 123 | |
| what | 何 | 123 | |
| whether | かどうか | 123 | |
| top | 上 | 122 | |
| com | com | 121 | |
| compute | 計算する | 121 | |
| form | 形 | 121 | |
| goal | 目標 | 121 | |
| hidden | 隠された | 121 | |
| improve | 改善する | 121 | |
| majority | 過半数 | 121 | |
| applied | 適用された | 120 | |
| effect | 効果 | 120 | |
| labeled | 名札付き | 120 | |
| least | 少なくとも | 120 | |
| li | li | 120 | |
| observe | 観察する | 120 | |
| predict | 予測する | 120 | |
| weights | 重り | 120 | |
| evaluate | 評価する | 119 | |
| notion | 概念 | 119 | |
| related | 関連する | 119 | |
| selected | 選択された | 119 | |
| problems | 問題点 | 118 | |
| according | に応じて | 117 | |
| able | できる | 116 | |
| art | 芸術 | 116 | |
| ez | ez | 116 | |
| qi | 気 | 116 | |
| rule | 規則 | 116 | |
| attack | 攻撃 | 115 | |
| bayesian | baysesian | 115 | onomasticon |
| hard | 硬い | 115 | |
| he | 彼 | 115 | |
| cases | 場合 | 114 | |
| demographic | 人口統計 | 114 | |
| equation | 方程式 | 113 | |
| rates | 率 | 113 | |
| close | 閉じる | 112 | |
| forward | 進む | 112 | |
| generalization | 一般化 | 112 | |
| perform | 演じる | 112 | |
| mapping | 対応づけ | 111 | |
| must | しなければならない | 111 | |
| norm | 規範 | 111 | |
| still | まだ | 111 | |
| cvpr | cvpr | 110 | abbreviation |
| requires | が必要です | 110 | |
| social | 社会 | 110 | |
| techniques | 技術 | 110 | |
| triplet | 三揃 | 110 | |
| compare | 比べる | 109 | |
| estimate | 見積り | 109 | |
| future | 未来 | 109 | |
| computing | 計算 | 108 | |
| region | 領域 | 108 | |
| theory | 理論 | 108 | |
| allocation | 割り当て | 107 | |
| instances | 実例 | 107 | |
| FALSE | 偽 | 106 | |
| robust | 壮健 | 105 | |
| strategy | 戦略 | 105 | |
| success | 成功 | 105 | |
| threshold | 敷居 | 105 | |
| described | 説明した | 104 | |
| experiment | 実験 | 104 | |
| important | 重要 | 104 | |
| made | 製 | 104 | |
| means | 手段 | 104 | |
| science | 理科 | 104 | |
| settings | 設定 | 104 | |
| ai | ai | 103 | abbreviation |
| architectures | 建築 | 103 | |
| fully | 完全に | 103 | |
| obtained | 得た | 103 | |
| similarly | 同様に | 103 | |
| take | 取る | 103 | |
| finally | 最後に | 102 | |
| world | 世界 | 102 | |
| compas | 羅針盤 | 101 | |
| directly | 直接 | 101 | |
| goodfellow | 仲良し | 101 | |
| notions | 概念 | 101 | |
| perturbations | 摂動 | 101 | |
| stochastic | 確率的 | 101 | |
| within | 以内に | 101 | |
| detection | 検出 | 100 | |
| experimental | 実験的 | 100 | |
| including | 含む | 100 | |
| present | 提示 | 100 | |
| access | 接続 | 99 | |
| advances | 前進 | 99 | |
| artificial | 人工的な | 99 | |
| focal | 焦点 | 99 | |
| quality | 品質 | 99 | |
| significantly | かなり | 99 | |
| support | 支援 | 99 | |
| benefit | 恩恵 | 98 | |
| computed | 計算された | 98 | |
| domains | 領域 | 98 | |
| inequality | 不平等 | 98 | |
| maximum | 最大 | 98 | |
| might | かもしれない | 98 | |
| range | 範囲 | 98 | |
| similarity | 類似性 | 98 | |
| upper | 上の | 98 | |
| user | 利用者 | 98 | |
| end | 終わり | 97 | |
| entropy | エントロピー | 97 | |
| few | 少ない | 97 | |
| found | 見つけた | 97 | |
| literature | 文献 | 97 | |
| pi | pi | 97 | |
| recent | 最近 | 97 | |
| robustness | 堅牢性 | 97 | |
| vision | 資格 | 97 | |
| zhang | 張 | 97 | |
| algorithmic | 解法の | 96 | |
| di | di | 96 | |
| intelligence | 知性 | 96 | |
| smaller | 小さい | 96 | |
| testing | 検査 | 96 | |
| total | 合計 | 96 | |
| towards | に向かって | 96 | |
| who | 誰 | 96 | |
| vectors | 有向量 | 95 | |
| condition | 調子 | 94 | |
| defer | 延期する | 94 | |
| equality | 平等 | 94 | |
| focus | 絞る | 94 | |
| normalized | 正規化 | 94 | |
| right | 正しい | 94 | |
| shot | 撃つ | 94 | |
| center | 中央 | 92 | |
| confidence | 信頼 | 92 | |
| either | どちらか | 92 | |
| imagenet | 像網 | 92 | |
| main | 主 | 92 | |
| nips | nips | 92 | abbreviation |
| rather | むしろ | 92 | |
| works | 働く | 92 | |
| among | の間で | 91 | |
| chen | 陳 | 91 | |
| cos | cos | 91 | |
| dwork | dwork | 91 | onomasticon |
| eo | eo | 91 | |
| population | 人口 | 91 | |
| recall | 想起 | 91 | |
| stage | 段階 | 91 | |
| gaussian | gaussian | 90 | onomasticon |
| line | 線 | 90 | |
| semi | 副 | 90 | |
| specifically | 具体的に | 90 | |
| version | 版 | 90 | |
| wi | 版 | 90 | |
| zemel | zemel | 90 | onomasticon |
| bn | 十億 | 89 | |
| fact | 事実 | 89 | |
| generate | 生む | 89 | |
| hand | 手 | 89 | |
| https | https | 89 | abbreviation |
| increasing | 増加している | 89 | |
| known | 知られている | 89 | |
| next | 次 | 89 | |
| resulting | 結果として | 89 | |
| vi | vi | 89 | |
| angular | 角張った | 88 | |
| current | 電流 | 88 | |
| dropout | 脱落 | 88 | |
| head | 頭 | 88 | |
| trade | 取引 | 88 | |
| adult | 大人 | 87 | |
| attacks | 攻撃 | 87 | |
| conv | conv | 87 | |
| dann | dann | 87 | abbriviation |
| last | 最終 | 87 | |
| mining | 採掘 | 87 | |
| practice | 練習 | 87 | |
| reported | 報告 | 87 | |
| additional | 追加 | 86 | |
| allows | 許可する | 86 | |
| apply | 適用する | 86 | |
| approximate | 近似 | 86 | |
| approximation | 近似 | 86 | |
| box | 箱 | 86 | |
| considered | 考えた | 86 | |
| increases | 増加する | 86 | |
| predictive | 予測的 | 86 | |
| prior | 先に | 86 | |
| unfair | 不公平 | 86 | |
| units | 単位 | 86 | |
| wang | 王 | 86 | |
| zero | 零 | 86 | |
| context | 環境 | 85 | |
| crafting | 工芸 | 85 | |
| equivalent | 同等の | 85 | |
| natural | 自然の | 85 | |
| pair | 対 | 85 | |
| available | 利用可能 | 84 | |
| change | 変化する | 84 | |
| ci | ci | 84 | |
| formulation | 定式化 | 84 | |
| introduce | 導入する | 84 | |
| issues | 課題 | 84 | |
| ber | ビール | 83 | |
| cannot | できない | 83 | |
| ensure | 確保する | 83 | |
| equalized | 均等化 | 83 | |
| hence | したがって | 83 | |
| indeed | 確かに | 83 | |
| iq | iq | 83 | |
| likely | たぶん | 83 | |
| odds | 勝算 | 83 | |
| ut | ut | 83 | |
| although | でも | 82 | |
| makes | 作ります | 82 | |
| bounds | 境界 | 81 | |
| cs | cs | 81 | |
| introduced | 導入された | 81 | |
| left | 左 | 81 | |
| needs | 必要性 | 81 | |
| nn | nn | 81 | abbreviation |
| off | オフ | 81 | |
| welfare | 福祉 | 81 | |
| augmentation | 増強 | 80 | |
| control | 制御 | 80 | |
| dh | dh | 80 | |
| journal | 雑誌 | 80 | |
| noise | 雑音 | 80 | |
| reduce | 減らす | 80 | |
| sign | 符号 | 80 | |
| ui | ui | 80 | abbreviation |
| uses | 使用する | 80 | |
| wt | wt | 80 | |
| address | 住所 | 79 | |
| bounded | 跳ねる | 79 | |
| choice | 選択 | 79 | |
| chosen | 選ばれた | 79 | |
| discriminative | 差別的 | 79 | |
| distortion | ねじれ | 79 | |
| increase | 増加する | 79 | |
| logistic | 論理的 | 79 | |
| outcomes | 結果 | 79 | |
| part | 部 | 79 | |
| pattern | 類型 | 79 | |
| performed | 行った | 79 | |
| provides | 提供する | 79 | |
| yes | はい | 79 | |
| achieved | 達成した | 78 | |
| agents | 代理 | 78 | |
| aj | aj | 78 | |
| classified | 分類された | 78 | |
| demonstrate | 実証する | 78 | |
| dimension | 寸法 | 78 | |
| direction | 方向 | 78 | |
| hardt | 辛い | 78 | |
| require | 必要とする | 78 | |
| round | 円形 | 78 | |
| technique | 技術 | 78 | |
| unfairness | 不公平 | 78 | |
| zafar | zafar | 78 | onomasticon |
| complexity | 複雑 | 77 | |
| details | 詳細 | 77 | |
| difficult | 難しい | 77 | |
| easy | かんたん | 77 | |
| fraction | 分数 | 77 | |
| http | http | 77 | abbreviation |
| icml | icml | 77 | abbreviation |
| members | 要員 | 77 | |
| proposition | 命題 | 77 | |
| pytorch | pytorch | 77 | onomasticon |
| team | 班 | 77 | |
| various | いろいろ | 77 | |
| xt | xt | 77 | |
| around | 周り | 76 | |
| black | 黒 | 76 | |
| suppose | 仮定する | 76 | |
| achieves | 達成する | 75 | |
| computation | 計算 | 75 | |
| dp | dp | 75 | |
| interviewees | 面談対象者 | 75 | |
| satisfy | 満足させる | 75 | |
| sensitivity | 感度 | 75 | |
| survey | 調査 | 75 | |
| fast | 速い | 74 | |
| implementation | 実装 | 74 | |
| meta | 超 | 74 | |
| moreover | さらに | 74 | |
| question | 質問 | 74 | |
| regret | 後悔 | 74 | |
| relu | relu | 74 | |
| biases | 偏り | 73 | |
| definitions | 定義 | 73 | |
| er | er | 73 | |
| estimates | 見積り | 73 | |
| liu | u | 73 | |
| lr | lr | 73 | |
| provided | 提供された | 73 | |
| significant | 重要な | 73 | |
| unit | 単位 | 73 | |
| certain | 確かな | 72 | |
| credit | 信用 | 72 | |
| estimation | 推定 | 72 | |
| generative | 原動力 | 72 | |
| income | 所得 | 72 | |
| minimization | 最小化 | 72 | |
| optimize | 最適化する | 72 | |
| org | 組織 | 72 | |
| otherwise | さもないと | 72 | |
| program | 計画 | 72 | |
| reject | 拒絶する | 72 | |
| strategies | 戦略 | 72 | |
| users | 利用者 | 72 | |
| virtual | 仮想 | 72 | |
| bayes | bayes | 71 | onomasticon |
| before | 前 | 71 | |
| david | david | 71 | |
| efficient | 効率的 | 71 | |
| full | 一杯 | 71 | |
| implies | 含意する | 71 | |
| kn | kn | 71 | |
| mcbn | mcbn | 71 | abbreviation |
| minimizing | 最小化 | 71 | |
| observed | 観察された | 71 | |
| pa | pa | 71 | |
| patterns | 類型 | 71 | |
| subset | 部分集合 | 71 | |
| errors | 誤り | 70 | |
| four | 四 | 70 | |
| get | 取得する | 70 | |
| overall | 全体 | 70 | |
| people | 人 | 70 | |
| properties | 財産 | 70 | |
| repair | 修復 | 70 | |
| residual | 残余 | 70 | |
| tail | 尾 | 70 |
docker
プログラムちょい替え（10）英語(14) 単語帳作成 dockerで（文字コード対応）量子計算機 arXiv掲載 西森 秀稔 論文(shell, awk), docker(82)
https://qiita.com/kaizen_nagoya/items/319672853519990cee42
作成時のスクリプトなどを利用。
macOS
$ docker run -v /tmp/docker:/tmp/docker -it kaizenjapan/qc-nakamori /bin/bash
-v /tmp/docker:/tmep/dockerは、dockerを起動するシステムに/tmp/dockerというフォルダが利用可能な場合に記載。フォルダ名は、利用できるもので、docker側が空のものであればよい。ファイル共有するつもりがなければ不要。
docker/ubuntu
# cd /home/wataoka/work
# ./ptt.sh
# cat ../text/* > all.txt
# tr 'a-z' 'A-Z' < all.txt > small.txt
# awk -f wc.awk small.txt > wc.txt
# ls ../
pdf text work
# ls
all.txt ptt.sh small.txt wc.awk wc.csv wc.org wc.txt wc.xlsx
wc.csvが14302語の単語帳csv(comma separated value)ファイル。
同じ内容のwc.xlsxはExcelファイル
上記表の90回以上出現する単語は日本語は未整備。
ptt.sh
#!/bin/bash
cd ../nakamori
for File in *; do
case ${File##*.} in
pdf|PDF )
echo "Convert the PDF:" ${File}
pdftotext -q ${File} ../text/${File}.txt ;;
*) ;;
esac
done
固有名詞などの確認
docker/ubuntu
# cd work
# grep zemel all.txt
University of Toronto, Department of Computer Science. Supported by NSERC. Email: zemel@cs.toronto.edu.
mlr.press/v28/zemel13.html.
zemel@cs.toronto.edu
共出現率計算プログラム検索中。
awk
wc.awk
# Print list of word frequencies
# https://researchmap.jp/jomd7nobo-45644/
# This is for papers, not for programming.
# remove _ from gsub list 20200101 @kaizen_nagoya
{
$0 = tolower($0) #
gsub(/[^a-z \t]/, " ", $0) #
for (i = 1; i <= NF; i++)
freq[$i]++
}
END {
for (word in freq)
printf "%s\t%d\n", word, freq[word]
}
ファイル一覧
標題名順
| title | file |
|---|---|
| A Survey on Bias and Fairness in Machine Learning | 1908.09635.pdf |
| Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour | 1706.02677.pdf |
| Bayesian Uncertainty Estimation for Batch Normalized Deep Networks | 1802.06455.pdf |
| Certifying and removing disparate impact | 1412.3756.pdf |
| Class-Balanced Loss Based on Effective Number of Samles | 1901.05555.pdf |
| Classification with Fairness Constraints: A Meta-Algorithm with Probable Guarantees | 1806.06055.pdf |
| Cost-Sensitive Feature Selection by Optimizing F-measures | 1904.02301v1.pdf |
| Data Decision and Theoretical Implications when Adversarially Learning Fair Representations | 1707.00075.pdf |
| Decision Theory for Discrimination-aware Classification | ICDM_2012.pdf |
| Distillation as a Defense to Adversarial Perturbations against Deep Neural Networks | 1511.04508.pdf |
| Domain-Adversarial Training of Neural Networks | 1505.07818.pdf |
| Empirical Risk Minimization under Fairness Constraints | 1802.08626.pdf |
| Enhancing the Accuracy and Fairness of Human Decision Making | 1805.10318.pdf |
| Evolution of collective fairness in hybrid populations of humans and agents | 4450.pdf |
| Explaining and Harnessing Adversarial Examples | 1412.6572.pdf |
| Exploiting Synthetically Generated Data with Semi-Supervised Learning for Small and Imbalanced Datasets | 1903.10022.pdf |
| Fairness Behind a Veil of Ignorance: A Welfare Analysis for Automated Decision Making | 1806.04959.pdf |
| Fairness Through Awareness | 1104.3913.pdf |
| Fairness Through Computationally-Bounded Awareness | 1803.03239.pdf |
| Fairness Without Demographics in Repeated Loss Minimization | 1806.0801.pdf |
| Focal Loss for Dense Object Detection | 1708.02002.pdf |
| Generative Adversarial Minority Oversampling | 1903.09730v1.pdf |
| Group Fairness for Indivisible Goods Allocation | group-fairness-aaai.pdf |
| Imperceptible Adversarial Attacks on Tabular Data | 1911.03274.pdf |
| Improving Fairness in Machine Learning Systems: What Do Industry Practitioners Need? | 1812.05239.pdf |
| Large-Margin Softmax Loss for Convolutional Neural Networks | 1612.02295.pdf |
| Learning Adversarially Fair and Transferable Representations | 1802.06309.pdf |
| Learning Controllable Fair Representations | 1812.04218.pdf |
| Learning Fair Representations | icml-final.pdf |
| Learning to Model the Tail | 7278-learning-to-model-the-tail.pdf |
| Limitations of the Lipschits constant as a defense against adversarial examples | 1807.09705.pdf |
| Max-margin Class Imbalanced Learning with Gaussian Affinity | 1901.07711.pdf |
| mixup: Beyond Empirical Risk Minimization | 1710.09412.pdf |
| On Learning Density Aware Embeddings | 1904.03911.pdf |
| On the Legal Comparability of Fairness Definitions | 1912.00761.pdf |
| One-network Adversarial Fairness | AAAI2019_OneNetworkAdversarialFairness.pdf |
| Online Learning with an Unknown Fairness Metric | 1802.06936.pdf |
| Oversampling for Imbalanced Data via Optimal Transport | 4381.pdf |
| Practical Black-Box Attacks against Machine Learning | 1602.02697.pdf |
| Predict Responsibly: Improving Fairness and Accuracy by Learning to Defer | 1711.06664.pdf |
| PyTorch: An Imperative Style, High-Performance Deep Learning Library | 1912.01703.pdf |
| Recovering from Biased Data: Can Fairness Constraints Improve Accuracy? | 1912.01094.pdf |
| SphereFace: Deep Hypersphere Embedding for Face Recognition | 1704.08063.pdf |
| Striking the Right Balance with Uncertainty | 1901.07590.pdf |
| The Cost of Fairness in Binary Classification | menon18a.pdf |
| The Limitations of Deep Learning in Adversarial Settings | 1511.07528.pdf |
| Trainable Undersampling for Class-Imbalance Learning | aaai.2019.classimbalance.pdf |
| Transfer of Machine Learning Fairness across Domains | 1906.09688.pdf |
| Using Image Fairness Representations in Diversity-Based Re-ranking for Recommendations | 1809.03577.pdf |
| Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning | 1704.03976.pdf |
参考文献一覧(reference list)
上記書籍の参考文献の一覧を作成中。
各論文で出てくる略号も一覧中。
参考文献駆動執筆の分類
https://qiita.com/kaizen_nagoya/items/fc2300b3717f61d39dee
参考文献駆動執筆(references driven writing)・デンソークリエイト編
https://qiita.com/kaizen_nagoya/items/b27b3f58b8bf265a5cd1
参考資料(reference)
dockerで機械学習 with anaconda(1)「ゼロから作るDeep Learning - Pythonで学ぶディープラーニングの理論と実装」斎藤 康毅 著
https://qiita.com/kaizen_nagoya/items/a7e94ef6dca128d035ab
「ゼロから作るDeep Learning ２自然言語処理編」読書会に参加する前に読んで置くとよい資料とプログラム
https://qiita.com/kaizen_nagoya/items/537b1810265bbbc70e73
docker(18) なぜdockerで機械学習するか 書籍・ソース一覧作成中 (目標100)
https://qiita.com/kaizen_nagoya/items/ddd12477544bf5ba85e2
docker(19) 言語処理100本ノックをdockerで。python覚えるのに最適。
https://qiita.com/kaizen_nagoya/items/7e7eb7c543e0c18438c4
ゼロから作るDeepLearning２自然言語処理編 読書会の進め方（例）
https://qiita.com/kaizen_nagoya/items/025eb3f701b36209302e
深層学習・機械学習 出願特許
https://qiita.com/kaizen_nagoya/items/38085a4fb38cd1d9e4d8
英語(1) プログラマが知っているとよい英単語の語源
https://qiita.com/kaizen_nagoya/items/9de6d47c47e2c211222b
＜この記事は個人の過去の経験に基づく個人の感想です。現在所属する組織、業務とは関係がありません。＞
文書履歴(document history)
ver. 0.01 初稿 20191231
ver. 0.02 略号、固有名詞追記 20200101 午後
ver. 0.03 用語頻度90まで一覧 略号、固有名詞追記 20200101 夕
ver. 0.04 用語頻度80まで一覧 略号、固有名詞追記 20200101 夜
ver. 0.05 用語頻度70まで一覧 略号、固有名詞追記 20200101 真夜中
ver. 0.06 参考文献一覧作成中 略号、固有名詞追記 20200102 午前
ver. 0.07 参考資料追記 20200102 昼
ver. 0.08 略号100語到達 20200102 午後
ver. 0.09 略号160語 20200103 午後
ver. 0.10 はじめに追記 20200103 夜
作業場
まだfull spellingを記載していない略号等
CASIA
LFW
PCA
RBF
L-BFGS
LSTMs
MP-DBM
EU
FEDER
LSP
LSPFS
SMOTE
USP
SIG
AALIM
AIDS.
CA
DMNS
ECGs
HIV
COLT
ICML
ITCS
NIPS
SIAM
GF1A
GF1B
COCO
HOG
SSD
YOLO
FGSM
NSF
ONR
AFOSR
SPAM
MLP
LAFTR
LeNet ?
PCA
SUN
t-SNE
CoRR
CRA
VGG-11
ADASYN
DANGER
MWMOTE
GRF
HKRGC
API
APL
CNTK
NumPy
SciPy
Pandas
FIFO
CPU
AUCPRC
ALLKNN
最後までおよみいただきありがとうございました。
いいね 💚、フォローをお願いします。
Thank you very much for reading to the last sentence.
Please press the like icon 💚 and follow me for your happy life.