FABENG4999H: Independent Study Honors Thesis

Open Courses

Aleix Martinez's ECE 5460 Image Processing at OSU (Full 2018)
Aleix Martinez's ECE 7868: Pattern Recognition and Machine Learning at OSU (Full 2018)
Michael Guerzhoy's CSC320: Introduction to Visual Computing at university of Toronto(Winter 2015)
Michael Guerzhoy's CSC321: Introduction to Machine Learning and Neural Networks at university of Toronto(Winter 2016)
Leonid Sigal's CPSC 425: Computer Vision at UBC(University of British Columbia) (2018 Full)
Leonid Sigal's CPSC 532S: Multimodal Learning with Vision, Language and Sound at UBC(University of British Columbia) (2018 Full)
Antonio Torralba's 6.869 Advances in Computer Vision at MIT
Ioannis Gkioulekas's 16-385 Computer Vision at CMU (Spring 2020)
Antonio Torralba's Computer Vision CSE 576, at Washinton Univeristy(Spring 2008)

Reading Schedule

Date	Topic	Notes	Reference
3/22	Perceptron	note1, note2, note3	CIML Chap4: The Perceptron, code: perceptron_demo.py
3/24	Gradient Descent	note1, note2, note3	CIML Chap7: Linear Models
3/26	Why does deep learning is not Overfitting?	note	CIML Chap2: Limits of Learning Overfitting or perfect fitting? Risk bounds for classification and regression rules that interpolate Reconciling modern machine-learning practice and the classical bias–variance trade-off
3/28	Understand CNN	note	CIML Chapter 10: Neural Networks Deep Learning Book Chapter 9: CNN Interesting Discover: Deep Neural Networks are Easily Fooled A Neural Algorithm of Artistic Style DEEP DREAM GENERATOR
5/4	ML: SVM -- recitation, lecture	formula,	Reference: cse3521, MIT 6.034, 老冯经典之作：纯白板手推SVM
5/5	ML: K-mean, KNN, Decision tree -- recitation, lecture	note1, note2	Reference: cse3521, MIT 6.034 人美声甜的数学系博士小姐姐带你读李航《统计学习方法》—— 第五章决策树
5/6	NN: feef-foward and back-prop	note	Reference: cse3521, cse5526
5/11	Practical Skills: One Hot Encoding and Label Encoding		Reference: Approaching (Almost) Any Machine Learning Problem, Mixing continuous and binary data with linear SVM?
6/1-6/30	Pipeline 1 -- Deep learning study. Topic Covers: Getting start: Machine Learning (2020): Course Introduction Use CNN to build a multi-class classifier for foods type. Use RNN to build a Seq2Seq text generator. Use GAN to build a ACGN character image generator. Use Reinforcement Learning to implement Policy Gradient and Actor-Critic(If time allow). (option) Explainable AI: AI can tell you this is a Cat, but why? (option) Adversarial Attack: How to prevent/filter the vicious noise? (option) Network Compression: How to compress the network/model, so could be embedded to smartphone, or portable device? (Option) Anomaly Detection: Could our network detect the anomaly image during training? Say "I don't know" why you need to. Transfer Learning(Domain Adversarial learning): Could our network still work well(Can identify the object) for a differnet backgroud scheme? (Option) Meta Learning: Can our machine learn the learning algorithm? (Option) Life-long Learning: Can our machine do continuous learning like Skynet in Terminator? aka Never ending learning, or Incremental learning. Pipeline 2 -- CSE 5525: Speech and Language Processing. Topic Covers: Apply Baive Baye classifier for IMDB sentiment analysis. Apply Perceptron and simple Feedforward network for IMDB sentiment analysis. Apply CNN for IMDB sentiment analysis. Implement the structured Perceptron and Viterbi algorithm for part-of-speech tagging on Twitter dataset.		Code Example Set: Deepnote(handsome_ML) Code Example Set: Deep learning with python
7/1-7/31	Pipeline 1 -- Computer Vision for Faces. Topic Covers: Getting start: AI Courses by OpenCV Pipeline 2 -- CSE 5524: Computer Vision for Human-Computer Interaction. Topic Covers: Pipeline 3 -- Udacitlabel: Introduction to Computer Vision. Topic Covers:		Code Example Set: Best practise for CV Code Example Set:OpenCV with Python CV Resources set: Awesome Computer Vision cs231n_fall2019 \| Github National Taiwan University_fall2018 \| Github
???	How to avoid overfitting in training Neural Network
???	Sergey Ioffe, Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
???	Regularization and variable selection via the elastic net
???	A fun topic: Inceptionism: Going deeper into Neural Networks		A Neural Algorithm of Artistic Style Artistic style transfer for videos Deep Dream/Style GENERATOR [Github] Greg Surma's project
???	CNN Receptive Field Computation Using Backprop		Email from Satya [Github]learnopencv
???	Ruslan talk's at Simons Institute: Tutorial on Deep Learning		Ruslan Salakhutdinov Deep Learning I Supervised Learning Deep Learning II Unsupervised Learning Deep Learning III Unsupervised Learning Deep Learning IV: Model Evaluation and Open Questions More