Week 1 Flashcards
What is the aim of pattern recognition ?
Aim is to build algorithms that can:
- recognise useful patterns in data
- extract useful information from data
To make decisions based on data.
Disciplines related to pattern recognition
ML - broader
NN - narrower
Statistic - overlapping
Disciplines to which pattern recognition is applied
Signal processing
Computer vision
Data mining
Why is pattern recognition important ?
Lots of data (data is cheap)
Finding patterns in data can be useful/lucrative
Retail applications of pattern recognition
Recommended algorithms
Financial applications of pattern recognition
Forecasting
Detecting fraud
Internet applications of pattern recognition
Spam detection for email
Medical applications of pattern recognition
DNA sequencing
Identifying proteins from AA sequences
Finding patterns in diagnostic tests
Classifier design cycle
What is learning ?
Acquiring and improving performance through experience
Define semi supervised learning ? When is it useful ? What does it need?
Training data consists of both labelled and unlabelled examples
For some samples we have {x, w} for some samples we have {x}
Useful when cost of having human label data is high
Need specialist learning algorithms able to update parameters when a label is and isn’t provided
Define RL
For each input, algo makes a decision w* =g(x)
A critic/teacher compares w* to correct answer and (occasionally) tells learner if it was correct (through rewards)
Transfer learning
Train classifier on 1 task (for which data is abundant)
Then further train on 2nd task for which data is less abundant
Define supervised learning
2 main methods for choosing hyper params
Exhaustive search (grid search)
Learning (meta learning)
Evaluation metrics for classifier
What is a dichotomiser
Classifier with only 2 classes
Evaluation metrics for dichotomizer
Recall and precision
