Sunday, April 30, 2017

Big video data, deep learning and forensic science

The increase of digital video data is still growing very fast, it is stated that 90 percent of the data on earth is created in the last two years. Also sensors of a self driving car could make 100 Gigabytes per second and suppose only a fraction is send to the cloud, then we have huge amounts of data that can be analyzed.

When this has to be analyzed we need fast methods for selection of relevant data. For humans it would not be feasible to process these amounts of data manually, so machine learning is one of the options to solve this. As chair of forensic data science at the Institute for Informatics of the University of Amsterdam this is one of the topics of research. Also combined with Biometrics where the privacy protection are top priority this makes new solutions possible for law enforcement.

It is also expected that more relevant statistical information is deducted from the data. However as always most of the data is heterogeneous and might be contaminated, so before drawing conclusion one should know a measure of uncertainty of the data.

One of the issues with deep learning is that part of it is a black box, and methods to explain how the network learns from the training sets are under development. However at the other side the human brain of an expert can also be seen as a black box, since by visual comparison the expert also uses previous experiences and is sensitive to bias. Research in this field is conducted and should also provide solutions to cope with this bias within forensic science.

Saturday, January 14, 2017

Antiforensic tools and criminal networks

In November I was the second reader of the PhD defense of the thesis of Michael Gruhn at the FAU University in Erlangen on rootkit and anti forensics software and how this can impact forensic science.
In December I was one of the promotors of the PhD defense of the thesis of Paul Duyn on criminal networks and a data driven approach on the different criminal networks as a complex adaptive system at the University of Amsterdam.
The combination of both approaches might even give more new insights, and nowadays there appears to be a growing interest in forensic data science since new approaches can be developed for preventing crimes from happening and examining crimes after they were committed. A multi-disciplinary approach is important to learn from each other fields and work on new solutions for example on cybercrime or any new crime that is developing. Even if antiforensics solutions have been used, possibilities exist to find forensic relevant information that can be used in court.
I look forward to many new multidisciplinary approaches, for example one of the approaches on forensic big data analysis is with the consortium Essential, were 15 PhD positions are available that will work on a range of topics within information policy and law.