Intelligent Systems in Computer Science

Perception

Interaction with environment requires cognitive processes, e.g. computer vision, speech recognition, motion detection, scene analysis, object classification

Decision making

Processing of incoming information, analysis of a situation, selection of possible actions in order to achieve a goal, e.g. path finding, sorting of objects

Learning

Data driven adaptation of the system based on observations only (unsupervised) or together with feedback from the environment (supervised), e.g. classification, regression.

Types of machine learning problems

• Unsupervised Learning: input data, no labels

  • Clustering: group similar data points
  • Density estimation
  • Dimensionality reduction
  • Outlier/novelty detection

• Supervised Learning: labels are provided

  • Classification
  • Regression

• Semi-Supervised learning

  • labels for just part of the data

• Reinforcement learning

  • find a sequence of actions (policy) that reaches a target

Classification vs Regression

Classification: Predict a discrete label from features

Examples:

• Medicine: classify X-rays as \cancer" or \healthy"
• SPAM detection: classify emails as spam or not
• Face recognition, speech recognition,

Remark the picture of from the lecture slides

Regression: Predict a continuous value

Examples:

• Weather forecasting (wind speed, mm, rainall)
• In financial markets: predict tomorrow's stock price from past evolution and external factors
• A robot learning its loccartion in an environment

Semi-supervised Training

Partially labelled data sets, first training based on labelled subset, extend by making predictions for unlabeled data, accept examples with high confidence for next training ... iterative procedure

Unsupervised Training

Clustering

Given a set of data points: partition it into groups such that points within each group are similar (low inter-group variability) and groups are dissimilar (high intra-group variability)