Recommender Systems III

General Where we focus User-Item RecommendationItem-Item RecommendationContent based recommendationCollaborative Filtering Recall: Notes on Item-Item Recommendations recommendation based on last item:“If you are interested in this, you probably also like …”doesn’t need a user modelItem model needed→ How to model similarities between items? Recall: Collaborative Filter … … …. Content based Recommendation we have more information…… Continue reading Recommender Systems III

Feature Extraction and Deep Learning

Feature Extraction Recall Supervised Classification the chosen vector space representation aligns the chosen model and how we make a classificationearlier, we took pixel by pixel and write it in the vector space Problem: If we took a slightly other camera angle, there is still a coffee mug, but it’s a whole different point in the…… Continue reading Feature Extraction and Deep Learning

SVMs, Model Selection und Outlier Detection

Non-linear Models II Support Vector Machines Summary: work with small data setsfor classification and regressiongives us the “one” best solutionnon-linear → many possibilitiescan be computing intensive with many classes, large data sets Basic model Linear classificationSupport only two classes {-1, 1}Parametrization:wx – b = 0(w is orthogonal to the line) New optimization problem: “Max Margin”…… Continue reading SVMs, Model Selection und Outlier Detection

Linear and Non-Linear ML Models

Linear Models Recall Classification: → there are other possibilities than Gaussian, e.g. geometrically Parameterization e.g. straight line (linear, 2D) How to find the Parameters? → we “draw” a line between the examples → many optimization methods to get better parameters Minimize the error (the Loss function L): Loss function has to be differential logistic function: Values…… Continue reading Linear and Non-Linear ML Models

Machine Learning

Introduction to Machine Learning We are here: Basic Definitions an Terminology: Basic Types of ML Algorithms: Supervised LearningLabeled dataDirect and quantitative evaluationmore present, but has limitsUnsupervised LearningLearn models from “ground truth” examplesPredict unseen examplesalgorithm that we want, but difficult to realizeReinforcement Learningnot that important for us, more present in robotics Supervised Learning: General: Classification Example:…… Continue reading Machine Learning

Basic Statistics and Data Wrangling

Revision of Basic Statistics Types of Data Continuous Data→ Data that can take any value in an intervall (also float, numeric, interval data)Discrete Data→ Data that can take only integer values (also integer, count)→ is our most used data typeCategorial Data→ Data that can take only predefined values representing a set of categories (also enums,…… Continue reading Basic Statistics and Data Wrangling

Recommender Systems

Intro Recommender Systems Definition: make product/service recommendations to people. Recommender systems want to identify items that are more relevant, so people consume more omnipresent in every big online store, streaming platformAmazon Example: “Customers who liked this, also liked….” In Practice: large systemsdifficult to buildexpensive to maintain Key problems: need to be fastprocess huge amount of data“good” recommendations…… Continue reading Recommender Systems