Viacheslav Komisarenko

Mr. Viacheslav Komisarenko

University of Tartu
orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
19 June 2020 in Remote Sensing
Verified
Journal Article
Open access
PDF file
IBIuliia Burdun
Iuliia Burdun
Michel BechtoldMichel Bechtold
Michel Bechtold
VSValentina Sagris
Valentina Sagris
VKViacheslav Komisarenko
Viacheslav Komisarenko
GLGabrielle De Lannoy
Gabrielle De Lannoy
Ülo ManderÜlo Mander
Ülo Mander
This study explored the potential of optical and thermal satellite imagery to monitor temporal and spatial changes in the position of the water table depth (WTD) in the peat layer of northern bogs. We evaluated three different trapezoid models that are proposed in the literature for soil moisture monitoring in regions with mineral soils. Due to the tight capillary connection between water table and surface soil moisture, we hypothesized that the soil moisture indices retrieved from these models would be correlated with WTD measured in situ. Two trapezoid models were based on optical and thermal imagery, also known as Thermal-Optical TRApezoid Models (TOTRAM), and one was based on optical imagery alone, also known as the OPtical TRApezoid Model (OPTRAM). The models were applied to Landsat imagery from 2008 to 2019 and the derived soil moisture indices were compared with in-situ WTD from eight locations in two Estonian bogs. Our results show that only the OPTRAM index was significantly (p-value < 0.05) correlated in time with WTD (average Pearson correlation coefficient of 0.41 and 0.37, for original and anomaly time series, respectively), while the two tested TOTRAM indices were not. The highest temporal correlation coefficients (up to 0.8) were observed for OPTRAM over treeless parts of the bogs. An assessment of the spatial correlation between soil moisture indices and WTD indicated that all three models did not capture the spatial variation in water table depth. Instead, the spatial patterns of the indices were primarily attributable to vegetation patterns.

3 April 2025 in Machine Learning
Verified
Journal Article
Open access
PDF file
VKViacheslav Komisarenko
Viacheslav Komisarenko
MKMeelis Kull
Meelis Kull
In many binary classification applications, the costs of false positives and negatives are imbalanced. Furthermore, there is often uncertainty about the exact costs of these errors. A natural measure-of-interest to be minimised in such scenarios is the expected misclassification cost. We identify many situations where this measure has analytic gradients, and thus it can be used as a training loss and optimised directly using empirical risk minimisation. In particular, we derive such losses from the Beta, Gamma and Gaussian distributions to model different kinds of cost uncertainty. The Beta family includes commonly used losses such as cross-entropy, squared error and 0–1 loss as special cases. The question then arises as to when it is appropriate to directly optimize the measure-of-interest, versus using a standard surrogate like cross-entropy or focal loss during training. After revisiting the theory of surrogate losses, proper losses and cost-sensitive learning to obtain good candidate surrogates out of derived families, we conduct an empirical comparison of derived training losses that, to our knowledge, were never tried on deep neural networks before, with the aim to minimise cost-sensitive measures-of-interest. The findings show that using Beta losses in training leads to improved performance compared to traditional training objectives like cross-entropy, label smoothing, and focal loss. This improvement is seen not only in terms of misclassification cost metrics, but (perhaps surprisingly) also in conventional metrics such as accuracy, mean squared error, and the area under the ROC curve.

16 October 2024
Verified
Book Chapter
Open access
PDF file
VKViacheslav Komisarenko
Viacheslav Komisarenko
MKMeelis Kull
Meelis Kull
Proper losses such as cross-entropy incentivize classifiers to produce class probabilities that are well-calibrated on the training data. Due to the generalization gap, these classifiers tend to become overconfident on the test data, mandating calibration methods such as temperature scaling. The focal loss is not proper, but training with it has been shown to often result in classifiers that are better calibrated on test data. Our first contribution is a simple explanation about why focal loss training often leads to better calibration than cross-entropy training. For this, we prove that focal loss can be decomposed into a confidence-raising transformation and a proper loss. This is why focal loss pushes the model to provide under-confident predictions on the training data, resulting in being better calibrated on the test data, due to the generalization gap. Secondly, we reveal a strong connection between temperature scaling and focal loss through its confidence-raising transformation, which we refer to as the focal calibration map. Thirdly, we propose focal temperature scaling - a new post-hoc calibration method combining focal calibration and temperature scaling. Our experiments on three image classification datasets demonstrate that focal temperature scaling outperforms standard temperature scaling.

21 August 2024 in arXiv
Verified
Preprint
Open access
PDF file
VKViacheslav Komisarenko
Viacheslav Komisarenko
MKMeelis Kull
Meelis Kull
Proper losses such as cross-entropy incentivize classifiers to produce class probabilities that are well-calibrated on the training data. Due to the generalization gap, these classifiers tend to become overconfident on the test data, mandating calibration methods such as temperature scaling. The focal loss is not proper, but training with it has been shown to often result in classifiers that are better calibrated on test data. Our first contribution is a simple explanation about why focal loss training often leads to better calibration than cross-entropy training. For this, we prove that focal loss can be decomposed into a confidence-raising transformation and a proper loss. This is why focal loss pushes the model to provide under-confident predictions on the training data, resulting in being better calibrated on the test data, due to the generalization gap. Secondly, we reveal a strong connection between temperature scaling and focal loss through its confidence-raising transformation, which we refer to as the focal calibration map. Thirdly, we propose focal temperature scaling - a new post-hoc calibration method combining focal calibration and temperature scaling. Our experiments on three image classification datasets demonstrate that focal temperature scaling outperforms standard temperature scaling.