perClass course

We provide basic information on pattern recognition and practical hands-on exercises enabling you to formulate pattern recognition problems and design working solutions. Emphasis is paid to development of practical skills in the area of data handling, visualization, classifier design and validation.

  • The lecturers Dr Pavel Paclik and Dr Carmen Lai have years of experience with the industrial pattern recognition courses at Delft Technical University.
  • The course is based on perClass package
  • The course may be offered in-house and thereby tailored to focus on specific topics of your interest.
  • The in-house course cost depends on the number of participants and the modules of your interest.

Would you like to know more? Please contact us for details!

Participants

The course can be tailored both for:

  • Beginners who are new to the field and would like to develop both a fundamental understanding of pattern recognition methods, and learn how to efficiently use them.
  • Advanced users who are already familiar with pattern recognition principles, but would like to deepen their proficiency in designing and optimizing practical solutions.

Course schedule

Day 1

morning

Module 1: Introduction to Pattern Recognition and data handling. How to handle data and meta-data information (e.g. patients, acquisition day, video frame, object ids...). How to visualize and inspect data interactively? How to work with multiple sets of labels? Many problems can be foreseen already at this early step!

afternoon

Module 2: Supervised classification. How to design detectors and discriminants? What are the assumptions, strengths and weaknesses of the different classifiers? How to turn classifier outputs into decisions?

Day 2

morning

Module 3: Evaluation. How to estimate classifier performance? How to optimize your classifier so that it provides desired performance in your application (e.g. max 5% error on cancer)? How to find out samples with a specific type of error?

afternoon

Module 4: Feature reduction. How to judge the quality of your features? How to select or extract informative data representation performing well together with a specific classifier?

Day 3

full day

Module 5: Pattern Recognition system design How to validate what objects are "sticking out" from your data distribution (scanner/normalization errors)? How to construct a multi stage classification system (detector + classifier)? How to tune it based on application requirements? How to write custom algorithms and evaluate them (e.g. leave-one-patient-out)?

optional

half day

Module 6: Unsupevised Learning: Cluster Analysis. How to get a feeling from your data even if you do not have ground truth? How to estimate number of clusters? How to interpret clustering results (e.g. connecting the scatter plot with image visualization in image segmentation problems)

half day

Module 7: Classifier embedding in your application. How to bring your trained classifiers from Matlab into your application? How to re-train/update your classifier in the production application without recompilation? No need for extra coding, the solution is right here!

Detailed module's content