Optimization for Deep Learning Highlights in 2017
Optimization for Deep Learning Highlights in 2017 Different gradient descent optimization algorithms have been proposed in recent years but Adam is still most commonly used. This post discusses the most exciting highlights and most promising recent approaches that may shape the way we will optimize our models in the future. This post discusses the most exciting highlights and most promising direct
PyTorch or TensorFlow?
This is a guide to the main differences I’ve found between PyTorch and TensorFlow. This post is intended to be useful for anyone considering starting a new project or making the switch from one deep learning framework to another. The focus is on programmability and flexibility when setting up the components of the training and deployment deep learning stack. I won’t go into performance (speed / me
ソニーãŒæ·±å±¤å¦ç¿’ã®GUIè¨è¨ˆãƒ„ール「Neural Network Consoleã€ã‚’公開
ソニーã¯2017å¹´8月17æ—¥ã€ãƒ—ãƒã‚°ãƒ©ãƒ コードを記述ã›ãšã«æ·±å±¤å¦ç¿’ã®å¦ç¿’モデル開発ãŒã§ãã‚‹GUIツール「Neural Network Consoleã€ã‚’無償公開ã—ãŸã€‚åŒç¤¾ãŒ2017å¹´6月27æ—¥ã«ã‚ªãƒ¼ãƒ—ンソースソフトウエア（OSS）ã¨ã—ã¦å…¬é–‹ã—ãŸæ·±å±¤å¦ç¿’フレームワーク「Neural Network Libraries（NNabla）ã€ã‚’使ã£ãŸé–‹ç™ºä½œæ¥ã‚’効率化ã§ãる。 深層å¦ç¿’ã«ä½¿ãˆã‚‹GUIツールã¨ã—ã¦ã¯ã€ç±³ã‚¨ãƒŒãƒ“ディアã®ã€ŒDIGITSã€ãªã©ãŒæ—¢ã«å…¬é–‹ã•ã‚Œã¦ã„る。ソニーãŒå…¬é–‹ã—ãŸGUIツールã®æ—¢å˜ãƒ„ールã¨æ¯”ã¹ãŸå¼·ã¿ã¯ã€ã€Œç´°ã‹ãªãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã¾ã§GUIã§æ“作ã§ãã‚‹ã€ç‚¹ã§ã‚ã‚‹ã¨ã€ã‚½ãƒ‹ãƒ¼ã®å°æž—由幸シニアマシンラーニングリサーãƒãƒ£ãƒ¼ã¯è©±ã™ã€‚
Deep Learning for NLP Best Practices
Deep Learning for NLP Best Practices Neural networks are widely used in NLP, but many details such as task or domain-specific considerations are left to the practitioner. This post collects best practices that are relevant for most tasks in NLP. This post gives an overview of best practices relevant for most tasks in natural language processing. Update July 26, 2017: For additional context, the Ha
Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour - Meta Research | Meta Research
Accurate, Large Minibatch SGD: Training ImageNet in 1 HourData @ Scale Deep learning thrives with large neural networks and large datasets. However, larger networks and larger datasets result in longer training times that impede research and development progress. Distributed synchronous SGD offers a potential solution to this problem by dividing SGD minibatches over a pool of parallel workers. Yet
ã‚¿ã‚«ãƒã‚·æ˜¥ã® GAN ç¥ã‚Šï¼ã€œ 一日一GAN(๑•̀ㅂ•Ì)و✧ 〜 - ABEJA Tech Blog
ABEJAã§ãƒªã‚µãƒ¼ãƒãƒ£ãƒ¼ã‚’ã—ã¦ã„ã¾ã™é«˜æ©‹ã§ã™ã€‚ 昨今 deep learning 界隈ã§ã¯ Generative Adversarial Net(GAN) ãŒæµè¡Œã£ã¦ã„ã¦ã€ä¸–ã¯ã¾ã•ã«ã‚¬ãƒ³ã‚¬ãƒ³è¡Œã“ã†ãœæ™‚代ã§ã™ã。 GAN を用ã„ã‚‹ã¨ç¶ºéº—ãªçµµãŒä½œæˆã§ããŸã‚ŠäºŒã¤ã®çµµã®ä¸é–“ã®ã‚ˆã†ãªçµµã‚’生æˆã§ããŸã‚Šã—ã¾ã™ã€‚例ãˆã°ã“ã®è«–æ–‡ã®ã‚ˆã†ãªæ„Ÿã˜ã§ã™ã€‚ã“ã®ã‚ˆã†ã« GAN ã¯æœ‰ç”¨ãªãƒ¢ãƒ‡ãƒ«ã§ã‚る一方ã€æœ€è¿‘ã® GAN ã§ã¯æ€¥ã«ã‚ˆãã‚ã‹ã‚‰ãªã„å¼ãŒå‡ºã¦ããŸã‚Šã™ã‚‹ã®ã§ã€å‹‰å¼·ã‚‚å…¼ãã¦ã€Œä¸€æ—¥ä¸€GANã€ã‚’ã‚„ã£ã¦ã¿ã¾ã—ãŸã€‚今回èªã‚“ã è«–æ–‡ã®ãƒªã‚¹ãƒˆã¯ä»¥ä¸‹ã§ã™ã€‚ EBGAN (https://arxiv.org/abs/1609.03126 ) WGAN (https://arxiv.org/abs/1701.07875) LSGAN (https://arxiv.org/abs/1611.04076) f-GAN (https://ar
æ–°è¦ã‚¢ãƒ—リケーションã¸ã®æ·±å±¤å¦ç¿’é©ç”¨ã®ãƒ™ã‚¹ãƒˆãƒ—ラクティス - Qiita
Deleted articles cannot be recovered. Draft of this article would be also deleted. Are you sure you want to delete this article? ã©ã®ã‚ˆã†ã«ã‚µãƒ¼ãƒ“スã«ãƒ‡ã‚£ãƒ¼ãƒ—ラーニングをé©ç”¨ã™ã‚Œã°ã„ã„ã‹ã€ãƒ™ã‚¹ãƒˆãƒ—ラクティスãŒã¾ã¨ã¾ã£ã¦ã„ãŸè«–æ–‡ãŒã‚ã£ãŸã®ã§ã€è¨³ã—ã¦ã¿ã¾ã—ãŸã€‚ 元論文 Best Practices for Applying Deep Learning to Novel Applications Leslie N. Smith https://arxiv.org/ftp/arxiv/papers/1704/1704.01568.pdf アブスト ã“ã®ãƒ¬ãƒãƒ¼ãƒˆã¯ã€ç‰¹å®šã®ãƒ‰ãƒ¡ã‚¤ãƒ³ã®ã‚¨ã‚スパートã ãŒãƒ‡ã‚£ãƒ¼ãƒ—ラーニングã«è©³ã—ããªã„人å‘㑠ディープラーニングを試ã—ã¦ã¿ãŸã„ã¨æ€ã£ã¦
ディープラーニングã«ã‚ˆã‚‹ãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã‚¢ã‚¤ãƒ†ãƒ 検出ã¨æ¤œç´¢ - ZOZO Technologies TECH BLOG
データサイエンティストã®ä¸æ‘ã§ã™ã€‚VASILYã§ã¯ãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã«ç‰¹åŒ–ã—ãŸç”»åƒè§£æžã‚¨ãƒ³ã‚¸ãƒ³ã‚’開発ã—ã¦ã„ã¾ã™ã€‚本記事ã§ã¯ã€ã‚¹ãƒŠãƒƒãƒ—写真ã‹ã‚‰ãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã‚¢ã‚¤ãƒ†ãƒ を検出ã™ã‚‹ã‚·ã‚¹ãƒ†ãƒ を紹介ã—ãŸã„ã¨æ€ã„ã¾ã™ã€‚ æ¦‚è¦ ã“ã®ã‚·ã‚¹ãƒ†ãƒ ã®å…¥åŠ›ã¯ã‚¹ãƒŠãƒƒãƒ—写真ã§ã™ã€‚スナップ写真ãŒå…¥åŠ›ã•ã‚ŒãŸã¨ãã€ã‚·ã‚¹ãƒ†ãƒ ã¯ä»¥ä¸‹ã®ã‚¿ã‚¹ã‚¯ã‚’解ãã¾ã™ã€‚ 写真ä¸ã‹ã‚‰ãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã‚¢ã‚¤ãƒ†ãƒ ã«è©²å½“ã™ã‚‹é ˜åŸŸã‚’検出ã™ã‚‹ 検出ã—ãŸãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã‚¢ã‚¤ãƒ†ãƒ ã®ã‚«ãƒ†ã‚´ãƒªã‚’予測ã™ã‚‹ 検出ã—ãŸãƒ•ã‚¡ãƒƒã‚·ãƒ§ãƒ³ã‚¢ã‚¤ãƒ†ãƒ ã«ä¼¼ã¦ã„るアイテムをDBã‹ã‚‰æ¤œç´¢ã™ã‚‹ å„タスクを解ã方法ã¯æ§˜ã€…ã‚ã‚Šã¾ã™ãŒã€å¼Šç¤¾ã®ã‚·ã‚¹ãƒ†ãƒ ã§ã¯2種類ã®ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚’使ã£ã¦ã“れをé”æˆã—ã¦ã„ã¾ã™ã€‚ ファッションアイテムã®æ¤œå‡ºã¨ã‚«ãƒ†ã‚´ãƒªäºˆæ¸¬ 検出ã¯ç”»åƒèªè˜ã®åŸºæœ¬çš„ãªã‚¿ã‚¹ã‚¯ã§ç››ã‚“ã«ç ”究ã•ã‚Œã¦ã„ã¦æ§˜ã€…ãªæ‰‹æ³•ãŒæ案ã•ã‚Œã¦ã„ã¾ã™ãŒã€ä»Šå›žã¯Single Shot MultiBox Detector (SSD)*1 ã¨å‘¼ã°
Factorized convolutional neural networks, AKA separable convolutions - FastML
The paper in question proposes a way to reduce the amount of computation needed in convolutional networks roughly three times, while keeping the same accuracy. Here’s what you wanted to know about this method (already available in TensorFlow), reprinted from two smart folks. swibe: In traditional convolution layers, the convolution is tied up with cross-channel pooling: for each output channel, a
深層ニューラルãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ï¿½ã«ã‚ˆã‚‹çŸ¥è˜ã®è‡ªå‹•ç²å¾—・推論
2. 自然言語処ç†ã¨ã¯ • 言葉をæ“ã‚‹è³¢ã„コンピュータを作る • 応用: æƒ…å ±æ¤œç´¢ï¼Œæ©Ÿæ¢°ç¿»è¨³ï¼Œè³ªå•å¿œç”ï¼Œè‡ªå‹•è¦ ç´„ï¼Œå¯¾è©±ç”Ÿæˆï¼Œè©•åˆ¤åˆ†æžï¼ŒSNS分æžï¼Œâ€¦ • 基礎: å“詞タグ付ã‘ï¼ˆå½¢æ…‹ç´ è§£æžï¼‰ï¼Œãƒãƒ£ãƒ³ã‚ ング，固有表ç¾æŠ½å‡ºï¼Œæ§‹æ–‡è§£æžï¼Œå…±å‚照解æžï¼Œ æ„味役割付与，… • 多ãã®ã‚¿ã‚¹ã‚¯ã¯ã€Œå…¥åŠ›ð‘¥ð‘¥ã‹ã‚‰å‡ºåŠ›ï¿½ð‘¦ð‘¦ã‚’予測〠�ð‘¦ð‘¦ = argmax ð‘¦ð‘¦âˆˆð‘Œð‘Œ ð‘ƒð‘ƒ(ð‘¦ð‘¦|ð‘¥ð‘¥) ※確率ã§ã¯ãªã„モデルもã‚ã‚Šã¾ã™ 2016-06-30 深層ニューラルãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ã‚ˆã‚‹çŸ¥è˜ã®è‡ªå‹•ç²å¾—・推論 2 3. å˜èªžåˆ—ã‹ã‚‰ãƒ©ãƒ™ãƒ«: Ì‚ð‘¦ð‘¦ = argmax ð‘¦ð‘¦âˆˆð‘Œð‘Œ ð‘ƒð‘ƒ(ð‘¦ð‘¦|ð’™ð’™) 2016-06-30 深層ニューラルãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ã‚ˆã‚‹çŸ¥è˜ã®è‡ªå‹•ç²å¾—・推論 3 The movie is the best I’ve ever seen! The movie
ディープラーニングã«ã‚ˆã‚‹ è‡ªç„¶è¨€èªžå‡¦ç† (技術編)
åå¤å±‹å¤§å¦ç‰¹åˆ¥è¬›ç¾© 2016å¹´6月29æ—¥(æ°´) ディープラーニングã«ã‚ˆã‚‹ è‡ªç„¶è¨€èªžå‡¦ç† (技術編) æ—¥æœ¬ã‚¢ã‚¤ãƒ»ãƒ“ãƒ¼ãƒ»ã‚¨ãƒ æ ªå¼ä¼šç¤¾ æ±äº¬åŸºç¤Žç ”究所 åªäº• ç¥å¤ª [email protected] 1 ニューラルãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æŠ€è¡“詳細 • 目的関数 • 誤差関数 • 目的関数ã®æœ€å°åŒ– • 勾é…法 • 目的関数ã®å¾®åˆ†è¨ˆç®— • 誤差逆ä¼æ¬æ³• • 誤差ã®åˆ†è§£ã¨å¯¾å‡¦æ‰‹æ³• • 推定誤差ã«åŠ¹ã手法 • 最é©åŒ–誤差ã«åŠ¹ã手法 • RNNã®è©±é¡Œ 2 目的関数: 誤差 • 教師ã‚ã‚Šå¦ç¿’ã®ç›®çš„関数 • 𑥠∈ ð‘‹: 入力, 𑦠∈ ð‘Œ: 出力 • 入力xã‹ã‚‰yを予測ã—ãŸã„å•é¡Œè¨å®š • 真ã®ç›®çš„関数: ð¿âˆ— 𜃠= ∫ð‘ ð‘¥, 𑦠ℓ𜃠ð‘¥, 𑦠ð‘‘ð‘¥ð‘‘𑦠• â„“ðœƒã¯äº‹ä¾‹ã”ã¨ã®æ失関数（後述） • 訓練データã§ã®èª¤å·® • データ分布p(x,y)ã¯æ™®é€šã‚ã‹ã‚‰ãªã„ã®ã§è¨“練データN個: D=
Wide & Deep Learning for Recommender Systems
Generalized linear models with nonlinear feature transformations are widely used for large-scale regression and classification problems with sparse inputs. Memorization of feature interactions through a wide set of cross-product feature transformations are effective and interpretable, while generalization requires more feature engineering effort. With less feature engineering, deep neural networks
ゼãƒã‹ã‚‰Deepã¾ã§å¦ã¶å¼·åŒ–å¦ç¿’ - Qiita
Deleted articles cannot be recovered. Draft of this article would be also deleted. Are you sure you want to delete this article? ãƒãƒœãƒƒãƒˆã‹ã‚‰è‡ªå‹•é‹è»¢è»Šã€ã¯ã¦ã¯å›²ç¢ãƒ»å°†æ£‹ã¨ã„ã£ãŸã‚²ãƒ¼ãƒ ã¾ã§ã€æ˜¨ä»Šå¤šãã®ã€ŒAIã€ãŒä¸–é–“ã‚’ã«ãŽã‚ã›ã¦ã„ã¾ã™ã€‚ ãã®ä¸ã®ã‚ーワードã¨ã—ã¦ã€ã€Œå¼·åŒ–å¦ç¿’ã€ã¨ã„ã†ã‚‚ã®ãŒã‚ã‚Šã¾ã™ã€‚ãã†ã—ãŸæ„味ã§ã¯ã€æ•°ã‚る機械å¦ç¿’ã®æ‰‹æ³•ã®ä¸ã§æœ€ã‚‚注目ã•ã‚Œã¦ã„ã‚‹(ãã—ã¦èª‡å¼µã•ã‚Œã¦ã„る・・・)手法ã¨ã‚‚ã„ãˆã‚‹ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“。 今回ã¯ãã®å¼·åŒ–å¦ç¿’ã¨ã„ã†æ‰‹æ³•ã«ã¤ã„ã¦ã€åŸºç¤Žã‹ã‚‰æœ€è¿‘目覚ã¾ã—ã„精度を出ã—ã¦ã„ã‚‹Deep Q-learning(ã„ã‚ゆるドã‚ュンã€DQNã§ã™)ã¾ã§ã€ãã®ç™ºå±•ã®æµã‚Œã¨ä»•çµ„ã¿ã«ã¤ã„ã¦è§£èª¬ã‚’ã—ã¦ã„ããŸã„ã¨æ€ã„ã¾ã™ã€‚ 本記事ã®å†…容をベースã«ã€ãƒãƒ³ã‚ºã‚ªãƒ³ã‚¤ãƒ™ãƒ³ãƒˆã‚’é–‹
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
https://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture8.pdf
Natural Language Inference, Reading Comprehension and Deep Learning
GitHub - rouseguy/DeepLearning-NLP: Introduction to Deep Learning for Natural Language Processing
https://www.bloomberg.com/news/articles/2016-07-19/google-cuts-its-giant-electricity-bill-with-deepmind-powered-ai
Wide & Deep Learning: Better Together with TensorFlow
SLOBSERVER - L'actualité de Second Life
Deepdist.com is for sale | HugeDomains
{{#tags}}- {{label}}
{{/tags}}