@inproceedings{Grum2023,
  author    = {Grum, Marcus},
  title     = {Learning representations by crystallized back-propagating errors},
  series = {Artificial intelligence and soft computing},
  booktitle = {Artificial intelligence and soft computing},
  editor    = {Rutkowski, Leszek and Scherer, Rafał and Korytkowski, Marcin and Pedrycz, Witold and Tadeusiewicz, Ryszard and Zurada, Jacek M.},
  publisher = {Springer},
  address   = {Cham},
  isbn      = {978-3-031-42504-2},
  doi       = {10.1007/978-3-031-42505-9_8},
  pages     = {78 -- 100},
  year      = {2023},
  abstract  = {With larger artificial neural networks (ANN) and deeper neural architectures, common methods for training ANN, such as backpropagation, are key to learning success. Their role becomes particularly important when interpreting and controlling structures that evolve through machine learning. This work aims to extend previous research on backpropagation-based methods by presenting a modified, full-gradient version of the backpropagation learning algorithm that preserves (or rather crystallizes) selected neural weights while leaving other weights adaptable (or rather fluid). In a design-science-oriented manner, a prototype of a feedforward ANN is demonstrated and refined using the new learning method. The results show that the so-called crystallizing backpropagation increases the control possibilities of neural structures and interpretation chances, while learning can be carried out as usual. Since neural hierarchies are established because of the algorithm, ANN compartments start to function in terms of cognitive levels. This study shows the importance of dealing with ANN in hierarchies through backpropagation and brings in learning methods as novel ways of interacting with ANN. Practitioners will benefit from this interactive process because they can restrict neural learning to specific architectural components of ANN and can focus further development on specific areas of higher cognitive levels without the risk of destroying valuable ANN structures.},
  language  = {en}
}
@phdthesis{Mey2016,
  author    = {Mey, J{\"u}rgen},
  title     = {Intermontane valley fills},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-103158},
  school      = {Universit{\"a}t Potsdam},
  pages     = {xii, 111},
  year      = {2016},
  abstract  = {Sedimentary valley fills are a widespread characteristic of mountain belts around the world. They transiently store material over time spans ranging from thousands to millions of years and therefore play an important role in modulating the sediment flux from the orogen to the foreland and to oceanic depocenters. In most cases, their formation can be attributed to specific fluvial conditions, which are closely related to climatic and tectonic processes. Hence, valley-fill deposits constitute valuable archives that offer fundamental insight into landscape evolution, and their study may help to assess the impact of future climate change on sediment dynamics. In this thesis I analyzed intermontane valley-fill deposits to constrain different aspects of the climatic and tectonic history of mountain belts over multiple timescales. First, I developed a method to estimate the thickness distribution of valley fills using artificial neural networks (ANNs). Based on the assumption of geometrical similarity between exposed and buried parts of the landscape, this novel and highly automated technique allows reconstructing fill thickness and bedrock topography on the scale of catchments to entire mountain belts. Second, I used the new method for estimating the spatial distribution of post-glacial sediments that are stored in the entire European Alps. A comparison with data from exploratory drillings and from geophysical surveys revealed that the model reproduces the measurements with a root mean squared error (RMSE) of 70m and a coefficient of determination (R2) of 0.81. I used the derived sediment thickness estimates in combination with a model of the Last Glacial Maximum (LGM) icecap to infer the lithospheric response to deglaciation, erosion and deposition, and deduce their relative contribution to the present-day rock-uplift rate. For a range of different lithospheric and upper mantle-material properties, the results suggest that the long-wavelength uplift signal can be explained by glacial isostatic adjustment with a small erosional contribution and a substantial but localized tectonic component exceeding 50\% in parts of the Eastern Alps and in the Swiss Rh{\^o}ne Valley. Furthermore, this study reveals the particular importance of deconvolving the potential components of rock uplift when interpreting recent movements along active orogens and how this can be used to constrain physical properties of the Earth's interior. In a third study, I used the ANN approach to estimate the sediment thickness of alluviated reaches of the Yarlung Tsangpo River, upstream of the rapidly uplifting Namche Barwa massif. This allowed my colleagues and me to reconstruct the ancient river profile of the Yarlung Tsangpo, and to show that in the past, the river had already been deeply incised into the eastern margin of the Tibetan Plateau. Dating of basal sediments from drill cores that reached the paleo-river bed to 2-2.5 Ma are consistent with mineral cooling ages from the Namche Barwa massif, which indicate initiation of rapid uplift at ~4 Ma. Hence, formation of the Tsangpo gorge and aggradation of the voluminous valley fill was most probably a consequence of rapid uplift of the Namche Barwa massif and thus tectonic activity. The fourth and last study focuses on the interaction of fluvial and glacial processes at the southeastern edge of the Karakoram. Paleo-ice-extent indicators and remnants of a more than 400-m-thick fluvio-lacustrine valley fill point to blockage of the Shyok River, a main tributary of the upper Indus, by the Siachen Glacier, which is the largest glacier in the Karakoram Range. Field observations and 10Be exposure dating attest to a period of recurring lake formation and outburst flooding during the penultimate glaciation prior to ~110 ka. The interaction of Rivers and Glaciers all along the Karakorum is considered a key factor in landscape evolution and presumably promoted headward erosion of the Indus-Shyok drainage system into the western margin of the Tibetan Plateau. The results of this thesis highlight the strong influence of glaciation and tectonics on valley-fill formation and how this has affected the evolution of different mountain belts. In the Alps valley-fill deposition influenced the magnitude and pattern of rock uplift since ice retreat approximately 17,000 years ago. Conversely, the analyzed valley fills in the Himalaya are much older and reflect environmental conditions that prevailed at ~110 ka and ~2.5 Ma, respectively. Thus, the newly developed method has proven useful for inferring the role of sedimentary valley-fill deposits in landscape evolution on timescales ranging from 1,000 to 10,000,000 years.},
  language  = {en}
}
@article{MeyScherlerZeilingeretal.2015,
  author    = {Mey, J{\"u}rgen and Scherler, Dirk and Zeilinger, Gerold and Strecker, Manfred},
  title     = {Estimating the fill thickness and bedrock topography in intermontane valleys using artificial neural networks},
  series = {Journal of geophysical research : Earth surface},
  volume    = {120},
  journal   = {Journal of geophysical research : Earth surface},
  number    = {7},
  publisher = {American Geophysical Union},
  address   = {Washington},
  issn      = {2169-9003},
  doi       = {10.1002/2014JF003270},
  pages     = {1301 -- 1320},
  year      = {2015},
  abstract  = {Thick sedimentary fills in intermontane valleys are common in formerly glaciated mountain ranges but difficult to quantify. Yet knowledge of the fill thickness distribution could help to estimate sediment budgets of mountain belts and to decipher the role of stored material in modulating sediment flux from the orogen to the foreland. Here we present a new approach to estimate valley fill thickness and bedrock topography based on the geometric properties of a landscape using artificial neural networks. We test the potential of this approach following a four-tiered procedure. First, experiments with synthetic, idealized landscapes show that increasing variability in surface slopes requires successively more complex network configurations. Second, in experiments with artificially filled natural landscapes, we find that fill volumes can be estimated with an error below 20\%. Third, in natural examples with valley fill surfaces that have steeply inclined slopes, such as the Unteraar and the Rhone Glaciers in the Swiss Alps, for example, the average deviation of cross-sectional area between the measured and the modeled valley fill is 26\% and 27\%, respectively. Finally, application of the method to the Rhone Valley, an overdeepened glacial valley in the Swiss Alps, yields a total estimated sediment volume of 9711km(3) and an average deviation of cross-sectional area between measurements and model estimates of 21.5\%. Our new method allows for rapid assessment of sediment volumes in intermontane valleys while eliminating most of the subjectivity that is typically inherent in other methods where bedrock reconstructions are based on digital elevation models.},
  language  = {en}
}