With predictive modeling techniques, it is possible to predict anything from clients’ shopping habits and illnesses to a golfer’s handicap. The only prerequisite is to have enough examples. In a doctoral thesis from the University of Borås in Sweden, Rikard König has adapted the technique of genetic programming so it can be used for such purposes.
The doctoral thesis, Enhancing Genetic Programming for Predictive Modeling, is about machine learning, more specifically predictive modeling, a field of computer science. Machine learning entails getting a computer to learn something, to become intelligent. Predictive modeling is a broad area of machine learning where a computer learns things on the basis of positive and negative examples, finds connections and explains why things turn out in a certain way.
Within predictive modeling, there is an array of techniques that are used to produce models that can predict practically anything, for instance, how people might be expected to respond to advertisements. Since these are general techniques, it is possible to predict just about anything as long as there are enough previous examples, i.e. sufficient information. The goal of predictive modeling is to find an accurate model and preferably one that explains something that was not previously known.
Genetic programming (GP) is a general optimization technique that is based on Darwin’s theories on evolution and natural selection. It is a technique that was not really designed for predictive modelling.
”In my thesis, I present several improvements that increase the accuracy and comprehensibility of models created with GP. There are many researchers who work with GP but my solutions are unique,” says Rikard König, PhD student at the School of Business and IT at the University of Borås.
In order to produce a model with the help of GP, you start off with, say, a thousand randomly chosen models and let them compete with each other. You work out how many errors the models make on known examples and then base a natural selection on the results. The most accurate models have a greater chance of surviving and having “children” – you pair off two models. These “children” are then a combination of their parents and form a new generation which is hopefully stronger. A small number of models can also be subjected to mutation, just like in nature.
”The new generation is assessed in the same way, using the known examples. They compete, pair off and give rise to an even stronger new generation. The process is repeated until a sufficiently accurate model has been found. The fascinating thing is that evolution is such a powerful way of searching through all possible solutions,” says Rikard König.
GP has several properties that make it suitable for predictive modeling. One example is that the search is independent of the representation of the model. This means that the exact representation and way of measuring errors can be adapted to individual problems. This is not normally the case with traditional predictive techniques. At the same time, the technique is problematic when a highly complex model is needed since the search goes through all possible solutions and the number of solutions increases exponentially with the complexity of the models.
”One of my improvements is a hybrid technique for creating an accurate and comprehensible model when the search space is extremely large, i.e. when a model with high complexity is required. The solution is to send relatively strong models created by a traditional predictive technique into a generation to guide the search in a promising direction.”
As part of his research, Rikard König has also produced an application that realizes his research results. The programme can be downloaded from www.grex.se
Rikard König is working on several research projects where these solutions may be put to use. For instance, one project is in collaboration with Scania where data from tens of thousands of lorries have been saved and will be analysed in order to explain what effect the driver has on fuel consumption. Another example, which also shows how generic the technique is, is a new project where golf swings from 500 golfers will be analysed. Here, the aim is to find general explanations for what distinguishes good swings from bad swings. Another aim is to be able to automatically recommend exercises for individual golfers on the basis of each person’s particular needs.
Thesis title: Enhancing Genetic Programming for Predictive Modeling
Contact: Rikard König, firstname.lastname@example.org, +46-33 435 5945, +46- 73922 96 56
Pressofficer Anna Kjellsson, email@example.com, +46-734 61 20 01
Anna Kjellsson | idw - Informationsdienst Wissenschaft
Rainbow colors reveal cell history: Uncovering β-cell heterogeneity
22.09.2017 | DFG-Forschungszentrum für Regenerative Therapien TU Dresden
The pyrenoid is a carbon-fixing liquid droplet
22.09.2017 | Max-Planck-Institut für Biochemie
Plants and algae use the enzyme Rubisco to fix carbon dioxide, removing it from the atmosphere and converting it into biomass. Algae have figured out a way to increase the efficiency of carbon fixation. They gather most of their Rubisco into a ball-shaped microcompartment called the pyrenoid, which they flood with a high local concentration of carbon dioxide. A team of scientists at Princeton University, the Carnegie Institution for Science, Stanford University and the Max Plank Institute of Biochemistry have unravelled the mysteries of how the pyrenoid is assembled. These insights can help to engineer crops that remove more carbon dioxide from the atmosphere while producing more food.
A warming planet
Our brains house extremely complex neuronal circuits, whose detailed structures are still largely unknown. This is especially true for the so-called cerebral cortex of mammals, where among other things vision, thoughts or spatial orientation are being computed. Here the rules by which nerve cells are connected to each other are only partly understood. A team of scientists around Moritz Helmstaedter at the Frankfiurt Max Planck Institute for Brain Research and Helene Schmidt (Humboldt University in Berlin) have now discovered a surprisingly precise nerve cell connectivity pattern in the part of the cerebral cortex that is responsible for orienting the individual animal or human in space.
The researchers report online in Nature (Schmidt et al., 2017. Axonal synapse sorting in medial entorhinal cortex, DOI: 10.1038/nature24005) that synapses in...
Whispering gallery mode (WGM) resonators are used to make tiny micro-lasers, sensors, switches, routers and other devices. These tiny structures rely on a...
Using ultrafast flashes of laser and x-ray radiation, scientists at the Max Planck Institute of Quantum Optics (Garching, Germany) took snapshots of the briefest electron motion inside a solid material to date. The electron motion lasted only 750 billionths of the billionth of a second before it fainted, setting a new record of human capability to capture ultrafast processes inside solids!
When x-rays shine onto solid materials or large molecules, an electron is pushed away from its original place near the nucleus of the atom, leaving a hole...
For the first time, physicists have successfully imaged spiral magnetic ordering in a multiferroic material. These materials are considered highly promising candidates for future data storage media. The researchers were able to prove their findings using unique quantum sensors that were developed at Basel University and that can analyze electromagnetic fields on the nanometer scale. The results – obtained by scientists from the University of Basel’s Department of Physics, the Swiss Nanoscience Institute, the University of Montpellier and several laboratories from University Paris-Saclay – were recently published in the journal Nature.
Multiferroics are materials that simultaneously react to electric and magnetic fields. These two properties are rarely found together, and their combined...
19.09.2017 | Event News
12.09.2017 | Event News
06.09.2017 | Event News
22.09.2017 | Life Sciences
22.09.2017 | Medical Engineering
22.09.2017 | Physics and Astronomy