一些常见的半监督学习文章总结,包括纯半监督学习,直推学习,主动学习
哈佛大学校长350周年校庆讲话(1636-1986)
从我们上次聚会庆祝哈佛大学校庆300周年纪念日起,50年过去了。人们从这段历史中可以清楚得看出,在今后的几十年内,几乎不大可能有什么大的动荡会改变哈佛,美国和整个地球。美国在世界上演的角色也没多大可能发生变化。少数民族和妇女对于更多机会的要求,医疗保健的巨大变化,技术的急速发展都不会吞没美国。最重要的危险,仍然是对知识的极端重要性的估计不足。
Session Recommendation
Attention Mechanism in Session Recommendation
This report focuses on the attention mechanism used in some previous session recommendation works. The conclusion is summarized in the final part: Some Intuition of Using Candidate Item as Query
Paper_Notes_About_Recommendation_in_SIGIR18
Overview
•Online recommendation(1)
•Recommendation with Social Networks(2+1)
Group representation, community detection, sequence-aware Rec
•Recommendation with Knowledge Base(1)
•Improve traditional methods(3)
APR, CMN, Bandit problem
•Some specific tasks(5)
Recommend email, mention, citation, Wikipedia article section
Conversational recommender system
User modeling: Geo-social based(1)
Paper summary_work based on meta learning
- Meta Learning Framework
- ProtoNet: 一个和 Matching Net 十分相似的处理 few shot learning 任务的
模型 - MAML: MAML 模型通过 Meta learning 的方法,尝试为所有的子任务学
习一个初始参数,使得各个子任务能够在该参数基础上快速收敛。 - Meta learning for unsupervised learning: 一种利用 meta learning 框架来使
用无标签数据的方法。 - Meta learning for item cold-start recommendation: Meta learning 在物品冷
启动任务中的一个应用。
BP derivation for MLP and CNN
\section{Task description}
Please derive a backpropagation process
(1) for the multi-layer neural network with one hidden layer, where data are in a m-dimensional feature space with n classes. Loss functions can use L2 distance or cross entropy.
(2) for the LeNet-5 CNN.
ml_basic_classifiers
1 Task description
- Based on the MNIST dataset, design and implement a proper convolutional neural network.
- Based on the MNIST dataset, design and implement classifiers including: least squares with regularization, Fisher discriminant analysis (with kernels), Perceptron (with kernels), logistic regression, SVM (with kernels), MLP-NN with two different error functions.
- (optional)Based on CNN classifiers, please implement an object detection task (including face recognition).
- Design and implement a proper recurrent neural network based on LSTM or/and GRU for Sentiment Analysis. Data is available at http://deeplearning.net/tutorial/lstm.html
ml basic knowledge practice_regression problem
1 Task Description
Use the following dataset to do house price predicting work.
Details: design a model to do house price predicting work. Linear Regression models including basic linear model based on polynomial, Ridge Regression, Lasso Regression and regression model based Decision Tree must be implemented. Regression models based on SVM and Deep Learning is optional.