Press Release, 07.02.2020

Medialer Temporallappen

Medial temporal lobe

How memory sharpens perception
Computer model provides new insight into the function of this brain region

A group of computational neuroscientists from Ruhr-Universität Bochum (RUB) showed with the help of a computer model that the medial temporal lobe could be indirectly involved in perception processes. Their results were published in the journal Hippocampus, online on December 30, 2019.

Hints of perceptual processes in the MTL

The medial temporal lobe (MTL) contains important brain structures such as the hippocampus and other anatomically related structures, which together ensure that we can consciously remember facts and experiences. However, a number of studies suggested that the MTL is also involved in perceptual processes. Studies with patients with a brain lesion in the MTL show that such an injury not only affects memory, but also the ability to perceptually differentiate between different objects.

Some neuroscientists see these results as an indication that the MTL could also be involved in the processing of visual information. Other researchers only conclude from the results that the assigned tasks also require memory and therefore the patients with the limited memory capacity performed less well.

Computer algorithm with a memory

With a computer model, Prof. Dr. Sen Cheng, Prof. Dr. Laurenz Wiskott and Richard Görler from Ruhr-Universität Bochum are now contributing new insights into the function of the MTL to the debate. Their theory: sensory representations - and thus the ability to interpret sensory impressions - are initially learned through the processing of sensory perceptions, but they can be improved down the line by replaying experiences from memory.

To test this theory, the researchers had a computer algorithm - slow feature analysis - perform a visual discrimination task. First, the algorithm had to learn to recognize two different images - it created visual representations of the two images. One showed the letter T and one the letter L. When creating the visual representations, the algorithm had a kind of memory available in one part of the experiment but not in the second part.

Now both algorithms had to face the visual discrimination task, where they should tell whether different images were more like the letter T or L. The letters were presented with background noise. They also overlapped each other (see graphic). "The algorithm works better the less attention it pays to the background noise," explains Richard Görler, first author of the study.


Medialer Temporallappen

Examples for the pictures in the study

The researchers found that the algorithm that could create the visual representation of the letters with the help of a memory performed better in the discrimination task than the algorithm that was trained without memories. To the RUB researchers, this result means that the medial temporal lobe does not have a direct role in the processing of sensory information, but because of its function in memory it helps to ensure that sensory perceptions can be interpreted more accurately.

Funding:
The study was supported by SFB 874, Project B2, which the German Research Foundation has been funding at Ruhr-Universität Bochum since 2010. SFB 874 "Integration and Representation of Sensory Processes" investigates how sensory signals generate neuronal maps, resulting in complex behaviour and memory formation. The study was also supported by the Federal Ministry of Education and Research and the funds of the DFG research unit 2812.

Reference:
Richard Görler, Laurenz Wiskott, Sen Cheng (2019) Improving sensory representations using episodic memory. Hippocampus. DOI: 10.1002/hipo.23186

Link to the publication:
https://onlinelibrary.wiley.com/doi/full/10.1002/hipo.23186

Contact:
Prof. Dr. Sen Cheng
Institut für Neuroinformatik
Computational Neuroscience
Ruhr-Universität Bochum
Phone: +49 - (0)234 - 32 - 29486
E-Mail: sen.cheng@ruhr-uni-bochum.de

Text: Judith Merkelt-Jedamzik

Translation: Judith Merkelt-Jedamzik

Downloads

Article as a PDF file