Uni-Logo
You are here: Home Laboratories Kuhl, Matthias Research Efficient Data Processing & Manipulation

Efficient Data Processing & Manipulation

Less is more!

Our work in data processing and manipulation is driven by the principle of "less is more". We aim to collect only the most essential data, while still achieving accurate and reliable results.

Our research spans a range of technologies, including data-driven compression, spatial compression, stochastic computing, and machine learning.

Data-Driven Compression
In data-driven compression, we collect only a few random measurements of natural signals, instead of the full set of high-dimensional data. We then find an appropriate compressible transform basis of the original signal and perform digital reconstruction of the signal through L1 minimization. This approach allows for efficient data processing without sacrificing accuracy.

Spatial Compression
In spatial compression, our multi-channel array technology allows for the monitoring of activity patterns with high spatial resolution and minimal spatial dimensions. We use smart interlinking of recording sites to reduce physical transmission lines and enable massively parallel sensing with a high density of recording sites.

spatial compression

exemplary data compression and reconstruction of a signal (source: own picture)

Stochastic Computing
Our research in stochastic computing offers an alternative to binary implementations of arithmetic operators, using stochastic number generation and only very few area-efficient logic gates. Having recently proven the realizability of arbitrary FIR filters, we are now investigating other filter structures and classification algorithms to expand the scope of this technology.

Filter

implementation of an FIR filter by means of stochastic computing (source: own picture)

Machine Learning
In machine learning, we focus on decision tree ensembles, which extract domain-specific features from raw data and use trained decision trees to vote for a result class. The resulting ultra-low power circuits and advanced classifier architectures enable efficient edge computing for long-life battery-powered and implantable devices.


Our work in data processing and manipulation offers a glimpse into the future of efficient and accurate data analysis. Whether you're a researcher or a student interested in studying microsystems technology, our technologies and methods offer innovative solutions for achieving more with less.

Personal tools