Deep Learning on Graphs
This book covers comprehensive contents in developing deep learning techniques for graph structured data with a specific focus on Graph Neural Networks (GNNs). The foundation of the GNN models are introduced in detail including the two main building operations: graph filtering and pooling operations. We then discuss the robustness and scalability of the GNNs, which are extremely important for util
OpenAIãŒç™ºè¦‹ã—ãŸScaling Lawã®ç§˜å¯† - ディープラーニングブãƒã‚°
OpenAIã¯GPT-3ã®æ¬¡ã®ç ”究を始ã‚ã¦ã„ã¾ã™ï¼Ž 世間ãŒGPT-3ã®ãƒ‡ãƒ¢ã«æ¹§ã上ãŒã‚‹ä¸ï¼ŒOpenAIã¯Scaling Lawã«é–¢ã™ã‚‹2本ã®è«–文をã²ã£ãã‚Šã¨å…¬é–‹ã—ã¾ã—ãŸï¼Ž Scaling Lawを一言ã§èª¬æ˜Žã™ã‚‹ãªã‚‰ã€ŒTransformerã®æ€§èƒ½ã¯ãŸã£ãŸ3ã¤ã®å¤‰æ•°ã®ã¹ã乗則ã«æ”¯é…ã•ã‚Œã¦ã„ã‚‹ã€ã¨ã„ã†ã‚‚ã®ã§ã™ï¼Ž Scaling Lawã¯ãã‚Œå˜ä½“ã ã‘ãªã‚‰èˆˆå‘³æ·±ã„話ã§çµ‚ã‚ã‚‹ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“．実際ã«è‹±èªžåœã§ã‚‚ã‚ã¾ã‚Šè©±é¡Œã«ã‚ãŒã£ã¦ã„ã¾ã›ã‚“．ã—ã‹ã—，ã“ã®æ³•å‰‡ã®æœ¬å½“ã®å‡„ã•ã«æ°—ã¥ã„ã¦ã„ã‚‹ç ”ç©¶è€…ã‚‚ã„ã¦ï¼Œãªãœè©±é¡Œã«ãªã‚‰ãªã„ã®ã‹å›°æƒ‘ã—ã¦ã„ã¾ã™ï¼Ž I am curious why people are not talking more about the OpenAI scaling law papers. For me, they seem very significant. What I heard so far:
深層å¦ç¿’ã®æ•°ç†ï¼šã‚«ãƒ¼ãƒãƒ«æ³•, スパース推定ã¨ã®æŽ¥ç‚¹
2. 深層å¦ç¿’ã®åºƒãŒã‚Š 2 [Glow: Generative Flow with Invertible 1x1 Convolutions. Kingma and Dhariwal, 2018] AlphaGo/Zero ç”»åƒã®ç”Ÿæˆ ç”»åƒã®å¤‰æ› ç”»åƒèªè˜ 自動翻訳 [Zhu, Park, Isola, and Efros: Unpaired image-to-image translation using cycle-consistent adversarial networks. ICCV2017.] 様々ãªã‚¿ã‚¹ã‚¯ã§é«˜ã„精度 [Silver et al. (Google Deep Mind): Mastering the game of Go with deep neural networks and tree search, Nature, 529, 484—489, 2016] [W
Deep Learning ideas that have stood the test of time
Deep Learning is such a fast-moving field and the huge number of research papers and ideas can be overwhelming. The goal of this post is to review ideas that have stood the test of time. These ideas, or improvements of them, have been used over and over again. They’re known to work. If you were to start in Deep Learning today, understanding and implementing each of these techniques would probably
SGDã®æ±ŽåŒ–性能ã®è§£æž: ãƒãƒƒãƒã‚µã‚¤ã‚ºã¨å¦ç¿’率ã®åŠ¹çŽ‡çš„ãªè¨å®š
3ã¤ã®è¦ç‚¹ ✔ï¸Â SGDã§äºˆæ¸¬ãƒ¢ãƒ‡ãƒ«ã®ãƒ‘ラメータ探索を行ãªã£ãŸéš›ã®æ±ŽåŒ–性能を導出 âœ”ï¸ ãƒãƒƒãƒã‚µã‚¤ã‚ºã¨å¦ç¿’率ã®æ¯”率ãŒå°ã•ã‘ã‚Œã°æ±ŽåŒ–性能ãŒå‘上ã™ã‚‹ã“ã¨ã‚’証明 âœ”ï¸ ã“ã®ã‚ˆã†ãªæ€§è³ªã‚’計1,600個ã®äºˆæ¸¬ãƒ¢ãƒ‡ãƒ«ã«ã‚ˆã‚‹å®Ÿé¨“ã§æ¤œè¨¼ Control Batch Size and Learning Rate to Generalize Well: Theoretical and Empirical Evidence written by Fengxiang He, Tongliang Liu, Dacheng Tao Comments: Published by NeurIPS 2019. Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)) 背景 深層å¦ç¿’モデルã¯ãã®è¡¨ç¾åŠ›ã®é«˜ã•ã‹ã‚‰æ§˜ã€…ãªåˆ†é‡Žã§å¿œç”¨ã•ã‚Œã¦ãŠã‚Šã€ã“れらã®ãƒ¢ãƒ‡ãƒ«ã‚’訓
データã®ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆã‚‚å¦ç¿’ã™ã‚‹End-To-End新音声åˆæˆã€EATS!
3ã¤ã®è¦ç‚¹ ✔ï¸Â 真ã®End-to-Endã«è¿‘ã¥ã„ãŸDeepMindã®æ–°éŸ³å£°åˆæˆæ–¹å¼ âœ”ï¸ ãƒ†ã‚ストã¨éŸ³å£°ã®ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆã‚’自動ã§å¦ç¿’å¯èƒ½ã« âœ”ï¸ GANè˜åˆ¥å™¨ã‚’使ã£ãŸå·§å¦™ãªæ失関数è¨è¨ˆ End-to-End Adversarial Text-to-Speech written by Jeff Donahue,Sander Dieleman,MikoÅ‚aj BiÅ„kowski,Erich Elsen,Karen Simonyan (Submitted on 5 Jun 2020) Comments: Published by arXiv Subjects: Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS) ã¯ã˜ã‚ã« DeepMindã‹ã‚‰ã¾ãŸæ–°ã—ã„音声åˆæˆãŒå‡ºã¾ã—ãŸ! ã™ã”ã„ã®? ã‚‚ã†T
Monte Carlo Dropoutã«ã‚ˆã‚‹åŒ»ç™‚AIã®è§£é‡ˆæ€§ã®å¯èƒ½æ€§ã¯ã‚ã‚‹ã‹ã€‚ãƒãƒªãƒ¼ãƒ—ã®ã‚»ã‚°ãƒ¡ãƒ³ãƒ†ãƒ¼ã‚·ãƒ§ãƒ³ã§ç¢ºèª
3ã¤ã®è¦ç‚¹ âœ”ï¸ ä¸ç¢ºå®Ÿæ€§æŽ¨å®šã¨è§£é‡ˆå¯èƒ½æ€§ã«ãŠã‘る最近ã®é€²æ©ã‚’大腸内視é¡æ¤œæŸ»ç”»åƒã‹ã‚‰ãƒãƒªãƒ¼ãƒ—をセグメンテーションタスクã«çµ„ã¿è¾¼ã‚€ âœ”ï¸ Montecarlo dropoutを使用ã—ã¦äºˆæ¸¬ã«é–¢é€£ã™ã‚‹ä¸ç¢ºå®Ÿæ€§ã‚’モデル化ã—ãŸä¸ç¢ºå®Ÿæ€§ãƒžãƒƒãƒ—を出力 âœ”ï¸ ã‚¬ã‚¤ãƒ‰ä»˜ããƒãƒƒã‚¯ãƒ—ãƒãƒ‘ゲーションを使用ã—ã¦é‡è¦ã¨è¦‹ãªã•ã‚Œã‚‹ãƒ”クセルを視覚化ã—ãŸè§£é‡ˆå¯èƒ½æ€§ãƒžãƒƒãƒ—を出力 Uncertainty and Interpretability in Convolutional Neural Networks for Semantic Segmentation of Colorectal Polyps written by Kristoffer Wickstrøm,Michael Kampffmeyer,Robert Jenssen (Submitted on 16 Jul 2018) Comments: Published
åã©ã‚‚ã®å¦ç¿’方法ã«ç€æƒ³ã€ãƒ‡ãƒ¼ã‚¿åŠ¹çŽ‡ã‚’高ã‚る機械å¦ç¿’ã®æ–°ã‚¢ãƒ—ãƒãƒ¼ãƒ
深層å¦ç¿’ã§ã¯ã€æ©Ÿæ¢°å¦ç¿’モデルã®æ£ç¢ºã•ã‚’高ã‚ã‚‹ã«ã¯è¨“練用データã®é‡ã‚’増やã™ã“ã¨ãŒä¸€èˆ¬çš„ã 。カーãƒã‚®ãƒ¼ãƒ¡ãƒãƒ³å¤§å¦ã®ç ”究者らã¯ã€è¦ªãŒåã©ã‚‚ã«æ•™ãˆã‚‹æ–¹æ³•ã«ãƒ’ントを得ã¦ã€ãƒ‡ãƒ¼ã‚¿é‡ã‚’増やã•ãªãã¦ã‚‚モデルã®æ£ç¢ºã•ã‚’高ã‚られるå¦ç¿’アルゴリズムを見ã„出ã—ãŸã€‚ by Karen Hao2020.05.11 48 17 19 7 機械å¦ç¿’アルゴリズムを訓練ã™ã‚‹æ¨™æº–çš„ãªæ–¹æ³•ã¯ã€ã™ã¹ã¦ã®è©³ç´°æƒ…å ±ã‚’ä¸€åº¦ã«ä¸Žãˆã‚‹ã“ã¨ã 。ã•ã¾ã–ã¾ãªå‹•ç‰©ã®ç¨®é¡žã‚’èªè˜ã™ã‚‹ç”»åƒåˆ†é¡žã‚·ã‚¹ãƒ†ãƒ を構築ã—ã¦ã„ã‚‹å ´åˆã§ã‚ã‚Œã°ã€å„種類ã®ä¾‹ã‚’示ã—ã€ãã‚Œã«å¿œã˜ã¦ãƒ©ãƒ™ãƒ«ã‚’付ã‘る。ãŸã¨ãˆã°ã€çŠ¬ã®å ´åˆã§ã‚ã‚Œã°ã€Œã‚¸ãƒ£ãƒ¼ãƒžãƒ³ã‚·ã‚§ãƒ‘ードã€ã‚„「プードルã€ã¨ã„ã£ãŸå…·åˆã 。 ã ãŒã€äººé–“ã®è¦ªãŒåã©ã‚‚ã«ä½•ã‹ã‚’æ•™ãˆã‚‹ã¨ãã®ã‚¢ãƒ—ãƒãƒ¼ãƒã¯ã¾ã£ãŸãç•°ãªã‚‹ã€‚親ã¯ã¯ã‚‹ã‹ã«åºƒç¯„ãªãƒ©ãƒ™ãƒ«ã§å§‹ã‚る。ã©ã‚“ãªçŠ¬ç¨®ã§ã‚ã£ã¦ã‚‚最åˆã¯å˜ã«ã€ŒçŠ¬ã€ã§ã‚る。åã©ã‚‚ãŒã“ã†ã—ãŸã‚ˆã‚Šå˜ç´”ãªã‚«ãƒ†ã‚´ãƒªã‚’区別ã™ã‚‹æ–¹æ³•ã‚’å¦ç¿’ã—
訓練ã—ã¦ã„ãªã„クラスã®ç”»åƒã‚’生æˆå¯èƒ½ï¼ï¼Ÿãƒ¡ãƒˆãƒªãƒƒã‚¯ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã‚’å°Žå…¥ã—ãŸOpenGAN
3ã¤ã®è¦ç‚¹ ✔ï¸Â 訓練ã—ã¦ã„ãªã„クラスã®ç”»åƒã‚’生æˆå¯èƒ½ ✔ï¸Â GANã«ãƒ¡ãƒˆãƒªãƒƒã‚¯ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã‚’å°Žå…¥ ✔ï¸Â OpenGANãŒç”Ÿæˆã—ãŸç”»åƒã¯ãƒ‡ãƒ¼ã‚¿æ°´å¢—ã—ã«ã‚‚有効 OpenGAN: Open Set Generative Adversarial Networks written by Luke Ditria, Benjamin J. Meyer, Tom Drummond (Submitted on 18 Mar 2020) Comments: Published by arXiv Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) GAN (Generative Adversarial Nets) ã®1ã¤ã¨ã—ã¦ã€cGAN (Conditional GAN) ãŒå˜åœ¨ã—ã¾ã™ã€‚cGANã¯
ä¼¼åˆã†é«ªåž‹ã¨ä¼¼ã¦ã„る芸能人をAIãŒè¨ºæ–ã™ã‚‹ã‚¢ãƒ—リã€é«˜æ ¡ç”ŸãŒé–‹ç™º | Ledge.ai
æ ªå¼ä¼šç¤¾ã‚¢ãƒ¼ã‚¹ãƒ›ãƒ¼ãƒ«ãƒ‡ã‚£ãƒ³ã‚°ã‚¹ã¯3月19æ—¥ã€å†™çœŸã‚’撮るã ã‘ã§ã€AI（人工知能）ãŒä¼¼åˆã†é«ªåž‹ã¨ä¼¼ã¦ã„る芸能人を診æ–ã™ã‚‹ã‚¢ãƒ—リ「AI STYLISTã€ã‚’リリースã—ãŸã€‚ AI STYLISTã¯ã€ã‚¹ãƒžãƒ¼ãƒˆãƒ•ã‚©ãƒ³ã®ã‚«ãƒ¡ãƒ©ã§å†™çœŸã‚’撮るã ã‘ã§ã€é¡”ç«‹ã¡ã‚„顔型をAIãŒåˆ†æžã—ã€ä¼¼åˆã†é«ªåž‹ã‚’æ案ã™ã‚‹ã‚µãƒ¼ãƒ“スã 。男女ã©ã¡ã‚‰ã«ã‚‚対応ã—ã¦ã„ã‚‹ã€‚ä¾¡æ ¼ã¯ç„¡æ–™ã€‚ç¾åœ¨ã¯iOS版ã®ã¿ã§ã€Android版も開発予定ã 。 ＞＞プレスリリース ç¾å½¹é«˜æ ¡ç”ŸãŒãƒ‡ã‚£ãƒ¼ãƒ—ラーニングを使ã£ãŸç‹¬è‡ªã®AIを開発AI STYLISTをリリースã™ã‚‹ã‚¢ãƒ¼ã‚¹ãƒ›ãƒ¼ãƒ«ãƒ‡ã‚£ãƒ³ã‚°ã‚¹ã¯ã€ç¾Žå®¹å®¤ã€ŒHAIR&MAKE EARTHã€ãªã©ã‚’é‹å–¶ã—ã¦ã„る。 プレスリリースã«ã‚ˆã‚Œã°ã€ã€Œæ˜¥ã¯ã‚¤ãƒ¡ãƒ¼ã‚¸ãƒã‚§ãƒ³ã‚¸ã‚’ã—ãŸã„ã¨æ€ã†äººãŒå¢—ãˆã‚‹æ™‚期。ヘアスタイルを変ãˆã¦ã¿ã‚ˆã†ï¼ã¨ã„ã†æ–¹ã‚‚多ãã„ã¾ã™ã€ã¨ã„ã†ã€‚ãã®éš›ã«ã€AI STYLISTを使ãˆã°ç†æƒ³ã®é«ªåž‹ã‚’見ã¤ã‘られるãã†ã 。美容室をé‹å–¶ã™ã‚‹
フォワードコンピュテーションを並列化ï¼ï¼Ÿã“ã‚Œã¾ã§ã®å›ºå®šæ¦‚念を崩ã™ç ”究ã®ç™»å ´ï¼
3ã¤ã®è¦ç‚¹ ✔ï¸Â 深層å¦ç¿’ã®é€æ¬¡çš„ãªè¨ˆç®—ã§ã‚るフォワードコンピュテーションを並列化ã™ã‚‹æ‰‹æ³•ã®æ案 âœ”ï¸ ãƒ•ã‚©ãƒ¯ãƒ¼ãƒ‰ã‚³ãƒ³ãƒ”ãƒ¥ãƒ†ãƒ¼ã‚·ãƒ§ãƒ³ã‚’è¤‡æ•°ã®éžç·šå½¢ãªæ–¹ç¨‹å¼ã¨è¦‹ãªã—ã€ã‚½ãƒ«ãƒãƒ¼ã‹ã‚‰è¨ˆç®— ✔ï¸Â フォワードコンピュテーションを最大33å€é«˜é€ŸåŒ–ã™ã‚‹ã“ã¨ã«æˆåŠŸ Nonlinear Equation Solving: A Faster Alternative to Feedforward Computation written by Yang Song, Chenlin Meng, Renjie Liao, Stefano Ermon (Submitted on 10 Feb 2020) Comments: accepted by arXiv Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML) ã¯ã˜ã‚㫠深層å¦ç¿’ã¯ç”»åƒèªè˜ã‚’ã¯ã˜ã‚
GANã®ç™ºå±•ã®æ´å²ã‚’振り返るï¼GANã®åŒ…括的ãªã‚µãƒ¼ãƒ™ã‚¤è«–æ–‡ã®ç´¹ä»‹(応用編)
3ã¤ã®è¦ç‚¹ ✔ï¸æ§˜ã€…ãªåˆ†é‡Žã§ä½¿ç”¨ã•ã‚Œã¦ã„ã‚‹ 「GANã€ã®åŒ…括的ãªã‚µãƒ¼ãƒ™ã‚¤è«–æ–‡ã®ç´¹ä»‹ ✔ï¸å¿œç”¨ç·¨ã§ã¯ã€GANã®å“質を測る指標ã¨æ§˜ã€…ãªå¿œç”¨ã‚’紹介 âœ”ï¸ ã“ã®è¨˜äº‹ã§GANãŒä½•ã«å¿œç”¨ã§ãã‚‹ã‹ç¶²ç¾…çš„ã«æŠŠæ¡ãŒå¯èƒ½ A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications written by Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, Jieping Ye (Submitted on 20 Jan 2020) subjects : Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) ã¯ã˜ã‚㫠今回ã®è¨˜äº‹ã¯ã€GANã®ã‚µãƒ¼ãƒ™ã‚¤è«–æ–‡(アルゴリズム編)ã«ç¶šãã€
ã¤ã„ã«èª•ç”Ÿï¼æœŸå¾…ã®æ–°ã—ã„活性化関数「Mishã€è§£èª¬
3ã¤ã®è¦ç‚¹ âœ”ï¸ ReLUã€Swishã«æ¬¡ãæ–°ãŸãªæ´»æ€§åŒ–関数Mishã‚’æ案 ✔ï¸Â MNISTã‚„CIFAR-10/100ãªã©ã§ReLUã¨Swishを圧倒 ✔ï¸Â è«–æ–‡ç†è€…実装ã®GitHubレãƒã¯æ—©é€Ÿ600以上ã®ã‚¹ã‚¿ãƒ¼ã‚’æŒã¡ã€éžå¸¸ã«ç°¡å˜ã«ä½¿ãˆã‚‹ Mish: A Self Regularized Non-Monotonic Neural Activation Function written by Diganta Misra (Submitted on 23 Aug 2019 (v1), last revised 2 Oct 2019 (this version, v2)) Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Neural and Evolutionary Comp
パラメータ数を激減ã•ã›ã‚‹æ–°ã—ã„畳ã¿è¾¼ã¿ã€ŒMixConvã€è§£èª¬ï¼
３ã¤ã®è¦ç‚¹ ãã®ï¼‘ パラメータ数を激減ã•ã›ã‚‹æ–°ã—ã„畳ã¿è¾¼ã¿MixConvã‚’æ案 ãã®ï¼’ MixConv層をå«ã‚“ã モデルをAIã«è‡ªå‹•ç”Ÿæˆ(=NAS)ã•ã›ã‚‹ã“ã¨ã§MixNetを開発 ãã®ï¼“ MixNetã¯MobileNet-V3ã‚„MnasNetãªã©ã®å°åž‹ç”»åƒèªè˜ãƒ¢ãƒ‡ãƒ«ã®ã¿ãªã‚‰ãšResNet-153ã«å¯¾ã—ã¦ã¯ãƒ‘ラメータ数1/9程度ã§æ€§èƒ½ã‚’凌ã„ã MixConv: Mixed Depthwise Convolutional Kernels written by Mingxing Tan, Quoc V. Le (Submitted on 22 Jul 2019 (v1), last revised 1 Dec 2019 (this version, v3)) Journal reference: BMVC 2019 Subjects: Computer Vision and Pattern
実務ã®å°‚門家ã¨ã—ã¦æ©Ÿæ¢°å¦ç¿’や統計分æžã‚’手掛ã‘ãŸã„人ã«ã‚ªã‚¹ã‚¹ãƒ¡ã®æ›¸ç±åˆç´š5冊＆ä¸ç´š8冊＋テーマ別11冊（2020å¹´2月版） - 渋谷駅å‰ã§åƒãデータサイエンティストã®ãƒ–ãƒã‚°
(Image by Pixabay) ã“ã®è¨˜äº‹ã¯ä»¥ä¸‹ã®ã‚ªã‚¹ã‚¹ãƒ¡æ›¸ç±ãƒªã‚¹ãƒˆè¨˜äº‹ã®ã‚¢ãƒƒãƒ—デートã§ã™ã€‚ 毎回ã®æ–り書ãã§æ縮ã§ã™ãŒã€ã“ã®è¨˜äº‹ã§ã¯ã€Œãƒ‡ãƒ¼ã‚¿ã‚µã‚¤ã‚¨ãƒ³ãƒ†ã‚£ã‚¹ãƒˆã‚„機械å¦ç¿’エンジニアãªã©ãƒ‡ãƒ¼ã‚¿åˆ†æžã®å®Ÿå‹™ã®å°‚門家ã¨ã—ã¦ã€*1機械å¦ç¿’や統計分æžã‚’手掛ã‘ã¦ã„ããŸã„ã¨ã„ã†ã€ä¸»ã«åˆç´šãªã„ã—ä¸ç´šãらã„ã®ã‚¹ã‚ルレベルã®äººãŸã¡ã«ãŠè–¦ã‚ã—ãŸã„書ç±ã‚’ã€åˆç´šå‘ã‘5冊・ä¸ç´šå‘ã‘8冊åŠã³ç´°ã‹ã„テーマ別ã«11冊ã€ãã‚Œãžã‚ŒæŒ™ã’ã¦ã„ãã¾ã™ã€‚スタンスã¨ã—ã¦ã¯ç›¸å¤‰ã‚らãšã€Œå½“座ã®æœ€çµ‚到é”点を『ä¸ç´šã€ã«ç½®ã„ãŸæ™‚ã«æœ€åˆã«èªã‚“ã§å†…容をマスターã—ã¦ãŠãã¹ã書ç±ã€ã‚’åˆç´šã«ç½®ã„ã¦ã„ã‚‹ã®ã§ã€ä¸–ã®ä¸ã®ã“ã®æ‰‹ã®ãŠè–¦ã‚書ç±ãƒªã‚¹ãƒˆã«æ¯”ã¹ã‚‹ã¨è‹¥å¹²ãƒãƒ¼ãƒ‰ãªå†…容ã®ã‚‚ã®ãŒåˆç´šå‘ã‘ã«å¤šã„ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“。 後ã¯ã¡ã‚‡ã£ã¨æ°—ãŒæ—©ã„ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“ãŒã€æ©Ÿæ¢°å¦ç¿’パートã«é–¢ã—ã¦ã¯ã€ŒAutoML時代ã«ã‚ã£ã¦ã‚‚実務ã®å°‚門家ã§ã‚ã‚Œã°çŸ¥ã£ã¦ãŠãã¹ã知è˜ã€ã‚’åŽã‚ãŸæ›¸ç±ã‚’é¸ã‚“ã§ãŠãã¾
教師ã‚ã‚Šå¦ç¿’ã®ç²¾åº¦ã‚’超ãˆãŸï¼ï¼Ÿç›¸äº’æƒ…å ±é‡ã®æœ€å¤§åŒ–ã«ã‚ˆã‚‹æ•™å¸«ãªã—å¦ç¿’手法IICã®ç™»å ´ï¼
3ã¤ã®è¦ç‚¹ ✔ï¸ç›¸äº’æƒ…å ±é‡ã‚’最大化ã™ã‚‹æž 組ã¿ã§ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆã‚’å¦ç¿’ã™ã‚‹æ•™å¸«ãªã—å¦ç¿’手法IICã®æ案 ✔ï¸äºˆæ¸¬å€¤ã‚’ãã®ã¾ã¾å‡ºåŠ›ã™ã‚‹ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆã‚’å¦ç¿’å¯èƒ½ã§ã‚ã‚‹ãŸã‚ã€ã‚¯ãƒ©ã‚¹ã‚¿ãƒªãƒ³ã‚°ãŒä¸è¦ ✔ï¸å¾“æ¥ã®æ•™å¸«ãªã—å¦ç¿’手法ã®ã€Œã‚¯ãƒ©ã‚¹ã‚¿ãŒä¸€ã¤ã«ã¾ã¨ã¾ã£ã¦ã—ã¾ã†å•é¡Œã€ãŠã‚ˆã³ã€ŒãƒŽã‚¤ã‚ºã«å¼±ã„ã¨ã„ã†å•é¡Œã€ã‚’解決 Invariant Information Clustering for Unsupervised Image Classification and Segmentation written by Xu Ji et.al (Submitted on 22 Aug 2019) subjects : Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) ã¯ã˜ã‚ã« è¿‘å¹´ã€æ§˜ã€…ãªå ´é¢ã«ãŠã„ã¦ã€æ·±å±¤å¦ç¿’手法ãŒä½¿ç”¨ã•
畳ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã®é«˜ç²¾åº¦åŒ–ã¨é«˜é€ŸåŒ–
2012å¹´ã®ç”»åƒèªè˜ã‚³ãƒ³ãƒšãƒ†ã‚£ã‚·ãƒ§ãƒ³ILSVRCã«ãŠã‘ã‚‹AlexNetã®ç™»å ´ä»¥é™ï¼Œç”»åƒèªè˜ã«ãŠã„ã¦ã¯ç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ (CNN) を用ã„ã‚‹ã“ã¨ãŒãƒ‡ãƒ•ã‚¡ã‚¯ãƒˆã‚¹ã‚¿ãƒ³ãƒ€ãƒ¼ãƒ‰ã¨ãªã£ãŸï¼ŽCNNã¯ç”»åƒåˆ†é¡žã ã‘ã§ã¯ãªã，セグメンテーションや物体検出ãªã©æ§˜ã€…ãªã‚¿ã‚¹ã‚¯ã‚’解ããŸã‚ã®ãƒ™ãƒ¼ã‚¹ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã¨ã—ã¦ã‚‚広ã利用ã•ã‚Œã¦ãã¦ã„る．本講演ã§ã¯ï¼ŒAlexNet以é™ã®ä»£è¡¨çš„ãªCNNã®å¤‰é·ã‚’振り返るã¨ã¨ã‚‚ã«ï¼Œè¿‘å¹´æ案ã•ã‚Œã¦ã„る様々ãªCNNã®æ”¹è‰¯æ‰‹æ³•ã«ã¤ã„ã¦ã‚µãƒ¼ãƒ™ã‚¤ã‚’è¡Œã„，ãれらを幾ã¤ã‹ã®ã‚¢ãƒ—ãƒãƒ¼ãƒã«åˆ†é¡žã—，解説ã™ã‚‹ï¼Žæ›´ã«ï¼Œå®Ÿç”¨ä¸Šé‡è¦ãªé«˜é€ŸåŒ–手法ã«ã¤ã„ã¦ã€ç•³ã¿è¾¼ã¿ã®åˆ†è§£ã‚„æžåˆˆã‚Šç‰ã®åˆ†é¡žã‚’è¡Œã„，ãã‚Œãžã‚Œè§£èª¬ã‚’è¡Œã†ï¼Ž Recent Advances in Convolutional Neural Networks and Accelerating DNNs 第21回ステアラボ人工知能セミナー講演資料 http
生æˆãƒ¢ãƒ‡ãƒ«ã‚’ä¸å¿ƒã¨ã—ãŸAI創薬最å‰ç·š / Elix CBI 2019
AI創薬ã§åˆ©ç”¨ã•ã‚Œã‚‹æ§˜ã€…ãªç”Ÿæˆãƒ¢ãƒ‡ãƒ«ã«ã¤ã„ã¦ã¾ã¨ã‚ã¦ã„ã¾ã™ã€‚CBIå¦ä¼š2019ã§ã®è¬›æ¼”スライドã§ã™ã€‚
ã€è«–文紹介】医用画åƒã¸ã®è»¢ç§»å¦ç¿’ã®æœ‰åŠ¹æ€§ã«ã¤ã„㦠Transfusion: Understanding Transfer Learning for Medical Imaging
特ã«åŒ»ç”¨ç”»åƒã‚’対象ã¨ã—ãŸã‚¿ã‚¹ã‚¯ã«ãŠã„ã¦ã€ImageNetãªã©ã®è‡ªç„¶ç”»åƒã‹ã‚‰å¦ç¿’ã—ãŸãƒ¢ãƒ‡ãƒ«ã‚’用ã„ãŸè»¢ç§»å¦ç¿’ãŒã‚ˆãè¡Œã‚ã‚Œã¦ã„ã¾ã™ãŒãã‚Œã¯æœ¬å½“ã«æœ‰åŠ¹ãªã®ï¼Ÿã¨ã„ã†ç–‘å•ã‚’検証ã—ãŸè«–文を紹介ã—ã¾ã™ã€‚ çµæžœã¨ã—ã¦ã€è»¢ç§»å¦ç¿’ã¯ãƒ‘フォーマンスã«ã»ã¨ã‚“ã©å½±éŸ¿ãŒãªãã€åŽæŸé€Ÿåº¦ãŒæ—©ããªã‚‹ã¨ã„ã†é™å®šçš„ãªæ©æµã§ã‚ã‚‹ã“ã¨ãŒæ˜Žã‚‰ã‹ã«â€¦
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
事å‰å¦ç¿’言語モデルã®å‹•å‘ / Survey of Pretrained Language Models
{{#tags}}- {{label}}
{{/tags}}