IMPORTANT DATE

Contents of Final report:

• Background, goal
• Details of Dataset
• Details of network structure or your method
• Experimental results
• Conclusions

目的

This course aims to introduce recent technics for understanding multimedia information (image, audio, video, and text) by a machine. We will examine popular machine learning techniques with some practice problems for understanding of these topics. Hence, prepare your laptop for every class.

目標：画像や音声，およびそれらの組み合わせである映像の機械による内容理解に関する原理・技術やサービスについて知見を広げることを目的とし，新しいメディアのあり方について自らの研究テーマと関連付けて考えられるようにする．

This course is designed to give an overview of various basic machine learning methods for understanding multimedia information. The course will give students an overall understanding of these fields and would make them realize that machine learning plays important roles in information technologies. The course will be based on lectures, discussions, and doing small projects.

使用テキスト

Texts will be suggested at this website.

受講要件と成績評価

Grading Philosophy: Report (100%).

Deep Learningの仕組みに関する資料

• MIRU2014 tutorial deeplearning, 山下先生 (in Japanese)
• 画像認識分野でのディープラーニングの研究動向, 岡谷先生 (in Japanese)

Introduction for Deep Neural Networks

• Neural Networks and Deep Learning by M. Nielsen (in English, online book)
• Deep Learning Tutorial by Y. LeCun (in English)

Recent research trend & Tool

• ICCV 2019 papers
  CVPR 2020 papers
• ChainerCV: a Library for Computer Vision in Deep Learning
• Tutorial on Theory and Application of Generative Adversarial Networks
• Segmentation by Deep NN

List of DNN framework

• Deep Learning Frameworks

Dataset

• Kaggle: Your Home for Data Science

内容 資料 オリエンテーション・画像認識 画像認識の概要 Birth of DNN (Science Film Museum) キュウリの仕分け機 in CeBIT2017 ニューラルネットワークの概要 frameworks Keras TensorFlow Chainer PyTorch MatConvNet Amazon Machine Learning Amazon Web Service Educate Google Cloud Platform Intoroduction for Neural Networks pdf (in Japanese) Sample code (Octave or Matlab) For creating sin curve samples with labels (sin_data.m) Back Propagation for sin curve samples (bp.m) Introduction for Pattern Recognition Pattern Recognition and Machine Learning (C.M. Bishop, in English) パターン認識概要（in Japanese） Cross-validation Precision and recall F-score Confusion matrix Convolutional Neural Networks (CNNs / ConvNets), Object Detection Convolutional Neural Networks for Visual Recognition (in English) R-CNN YOLO SSD Recurrent Neural Networks (RNN) Recurrent Neural Networks Tutorial, Part 1 Introduction to RNNs (in English) [Tensorflow] Implementing Temporal Convolutional Networks ChainerによるRNN翻訳モデルの実装+@ - SlideShare 実験的ノウハウ 深層学習を用いた画像識別タスクの精度向上テクニック 畳み込みニューラルネットワークの研究動向 SSII2018講演スライド Fully Convolutional Networks (FCN) Fully Convolutional Networks for Semantic Segmentation Fully Convolutional Networks by Chainer (in English) Fully Convolutional Networks 〜 Chainerによる実装 〜 Generative Adversarial Networks (GAN) CVPR'17 GAN Tutorial DCGAN-tensorflowで自動画像生成をお手軽に試す chainer-DCGAN (GitHub) Reinforcement Learning 深層強化学習ライブラリChainerRL Unityが機械学習のライブラリを公開したので使ってみる　(和訳付き) Introducing: Unity Machine Learning Agents 【物理エンジン】強化学習で二足歩行させてみた Reinforcement Learning for Biped Locomotions Tutorial: Deep Reinforcement Learning Googleの人工知能「DQN」が人間より上手にプレイできるゲームとできないゲームの境界線 DQNの生い立ち＋Deep Q-NetworkをChainerで書いた Recursive Cortical Networks (RCN) Full paper Theory Sample code SuperPoint:local feature description, featrue tracking GitHub Demo Real time object detection with NN Windows and Linux version of Darknet Yolo v2 Neural Networks for object detection Audio Processing with NN NSynth: Neural Audio Synthesis Text Processing tf-seq2seq 今更ながらchainerでSeq2Seq（1） 自然言語処理における前処理の種類とその威力 Word2Vec：発明した本人も驚く単語ベクトルの驚異的な力 word2vecを使って、日本語wikipediaのデータを学習する 学習済みword2vecモデルを調べてみた Training Word2Vec Model on English Wikipedia by Gensim word2vec (GitHub) CapsNet What is a CapsNet or Capsule Network? Dynamic Routing Between Capsules CapsNetについての調べ dynamic_routing_between_capsules (chainer) illustration2vec illustration2vec alternative illustration2vec（高次元タグ予想器）について For comparison: scikit-learn scikit-learn Machine Learning in Python Nearest Neighbor rule scikit-learn+Chainer Poster Presentation Giving an Effective Poster Presentation Scientific Poster Design Ten Simple Rules for a Good Poster Presentation Schedule 01st: Introduction 02nd: Theme selection 03rd: Implementation & Experiments 1 04th: Implementation & Experiments 2 05th: Implementation & Experiments 3 06th: Implementation & Experiments 4 07th: Implementation & Experiments 5 08th: Implementation & Experiments 6 09th: Implementation & Experiments 7 10th: Implementation & Experiments 8 11th: Implementation & Experiments 9 12th: Implementation & Experiments 10 13th.:Implementation & Experiments 11 14th.:Implementation & Experiments 12 15th.:Implementation & Experiments 13 Report submission via e-mail (s-hotta@cc.tuat.ac.jp)