Milan Dubljanin
Milan Dubljanin has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Milan Dejanović
Faculty of Science and Mathematics, University of Priština, Lole Ribara 29, 38220 Kosovska Mitrovica, Serbia
3 shared publications - Milan Savić
Faculty of Sciences and Mathematics, University of Pristina in Kosovska Mitrovica, Ive Lole Ribara 29, 38220 Kosovska Mitrovica, Serbia
3 shared publications - Stefan Panic
University of Priština in Kosovska Mitrovica, Faculty of Natural Sciences and Mathematics, Kosovska Mitrovica
2 shared publications - Danijel Djosic
Faculty of Science and Mathematics, University of Priština, Lole Ribara 29, 38220 Kosovska Mitrovica, Serbia
1 shared publication - Marko Smilić
Faculty of Sciences and Mathematics, University of Priština, Lole Ribara 29, 38220 Kosovska Mitrovica, Serbia
1 shared publication Mihajlo Stefanovic
Faculty of Electronic Engineering University of Nis Nis Serbia
1 shared publication
Latest Publications
- Open Access
- 29 June 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (7) , 521
Adaptive A-Semilogarithmic Gradient Quantization for Efficient Deep Neural Network Training
This paper introduces an adaptive A-semilogarithmic gradient quantization framework aimed at reducing memory overhead and computational complexity during the training of deep neural networks. The approach employs a semilogarithmic companding function parameterized by a dynamically adjusted scaling factor A, which evolves in response to the statistical properties of gradients throughout the training process. Two distinct quantization strategies are proposed and evaluated: The switching piecewise A-quantizer, which adaptively toggles between low-bit uniform and high-bit semilogarithmic quantization according to an exponentially weighted moving-average (EMA) estimate of gradient variance; and the hybrid A-quantizer, which statically partitions the gradient domain, applying uniform quantization in low-magnitude regions and semilogarithmic companding in high-magnitude regions. The proposed methods are empirically evaluated on both multilayer perceptron (MLP) and convolutional neural network (CNN) architectures using tabular and image-classification benchmarks, including DCCC, CIFAR-10, CIFAR-100, and ImageNet. Quantitative results demonstrate that both models achieve comparable classification accuracy to full-precision (FP32) baselines while significantly reducing gradient reconstruction error. Notably, the hybrid A-quantizer consistently yields better validation accuracy, reduced RMSE, and improved convergence behavior relative to its switching counterpart. These findings underscore the effectiveness of hybrid semilogarithmic quantization as a robust and efficient solution for training deep models in resource-constrained or bandwidth-limited environments, with strong potential for scalable deployment across diverse hardware platforms.
- 29 June 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (7) , 521
Open AccessAdaptive A-Semilogarithmic Gradient Quantization for Efficient Deep Neural Network Training
This paper introduces an adaptive A-semilogarithmic gradient quantization framework aimed at reducing memory overhead and computational complexity during the training of deep neural networks. The approach employs a semilogarithmic companding function parameterized by a dynamically adjusted scaling factor A, which evolves in response to the statistical properties of gradients throughout the training process. Two distinct quantization strategies are proposed and evaluated: The switching piecewise A-quantizer, which adaptively toggles between low-bit uniform and high-bit semilogarithmic quantization according to an exponentially weighted moving-average (EMA) estimate of gradient variance; and the hybrid A-quantizer, which statically partitions the gradient domain, applying uniform quantization in low-magnitude regions and semilogarithmic companding in high-magnitude regions. The proposed methods are empirically evaluated on both multilayer perceptron (MLP) and convolutional neural network (CNN) architectures using tabular and image-classification benchmarks, including DCCC, CIFAR-10, CIFAR-100, and ImageNet. Quantitative results demonstrate that both models achieve comparable classification accuracy to full-precision (FP32) baselines while significantly reducing gradient reconstruction error. Notably, the hybrid A-quantizer consistently yields better validation accuracy, reduced RMSE, and improved convergence behavior relative to its switching counterpart. These findings underscore the effectiveness of hybrid semilogarithmic quantization as a robust and efficient solution for training deep models in resource-constrained or bandwidth-limited environments, with strong potential for scalable deployment across diverse hardware platforms.
- Open Access
- 12 January 2026
- Published by MDPI AG in Information
- Vol. 17 (1) , 69
Efficient Quantization of Pretrained Deep Networks via Adaptive Block Transform Coding
This work investigates the effectiveness of block transform coding (BTC) as a lightweight, training-free quantization strategy for compressing the weights of pretrained deep neural networks. The proposed method applies a rule-based block transform with variance and root mean square error (RMSE)-driven stopping criteria, enabling substantial reductions in bit precision while preserving the statistical structure of convolutional and fully connected layer weights. Unlike uniform 8-bit quantization, BTC dynamically adjusts bit usage across layers and achieves significantly lower distortion for the same compression budget. We evaluate BTC across many pretrained architectures and tabular benchmarks. Experimental results show that BTC consistently reduces storage to 4–7.7 bits per weight while maintaining accuracy within 2–3% of the 32-bit floating point (FP32) baseline. To further assess scalability and baseline strength, BTC is additionally evaluated on large-scale ImageNet models and compared against a calibrated percentile-based uniform post-training quantization method. The results show that BTC achieves a substantially lower effective bit-width while incurring only a modest accuracy reduction relative to calibration-aware 8-bit quantization, highlighting a favorable compression–accuracy trade-off. BTC also exhibits stable behavior across successive post-training quantization (PTQ) configurations, low quantization noise, and smooth RMSE trends, outperforming naïve uniform quantization under aggressive compression. These findings confirm that BTC provides a scalable, architecture-agnostic, and training-free quantization mechanism suitable for deployment in memory- and computing-constrained environments.
- 12 January 2026
- Published by MDPI AG in Information
- Vol. 17 (1) , 69
Open AccessEfficient Quantization of Pretrained Deep Networks via Adaptive Block Transform Coding
This work investigates the effectiveness of block transform coding (BTC) as a lightweight, training-free quantization strategy for compressing the weights of pretrained deep neural networks. The proposed method applies a rule-based block transform with variance and root mean square error (RMSE)-driven stopping criteria, enabling substantial reductions in bit precision while preserving the statistical structure of convolutional and fully connected layer weights. Unlike uniform 8-bit quantization, BTC dynamically adjusts bit usage across layers and achieves significantly lower distortion for the same compression budget. We evaluate BTC across many pretrained architectures and tabular benchmarks. Experimental results show that BTC consistently reduces storage to 4–7.7 bits per weight while maintaining accuracy within 2–3% of the 32-bit floating point (FP32) baseline. To further assess scalability and baseline strength, BTC is additionally evaluated on large-scale ImageNet models and compared against a calibrated percentile-based uniform post-training quantization method. The results show that BTC achieves a substantially lower effective bit-width while incurring only a modest accuracy reduction relative to calibration-aware 8-bit quantization, highlighting a favorable compression–accuracy trade-off. BTC also exhibits stable behavior across successive post-training quantization (PTQ) configurations, low quantization noise, and smooth RMSE trends, outperforming naïve uniform quantization under aggressive compression. These findings confirm that BTC provides a scalable, architecture-agnostic, and training-free quantization mechanism suitable for deployment in memory- and computing-constrained environments.
- Open Access
- 1 January 2025
- Published by Centre for Evaluation in Education and Science (CEON/CEES)
- No. Saopštenja,p. 243-248
Application of neural networks in half marathon and marathon
Artificial intelligence is the fastest-growing field of computer science. Its use is growing and yet to be fully uncovered. Using deep learning and datasets of specific data, artificial neural networks can be trained to help users in everyday life. This paper aims to make neural networks to help get the best time in a half marathon and marathons. Many factors influence marathon race times, such as temperature, precipitation, sunshine or atmospheric pressure, sex, age and physical preparation and current condition of runners. A dataset from smartwatches will be used to determine the condition of runners, as well as a dataset Berlin marathon dataset. Combining the analysis of these two datasets, neural networks will predict if the ideal time was achieved.
- 1 January 2025
- Published by Centre for Evaluation in Education and Science (CEON/CEES)
- No. Saopštenja,p. 243-248
Open AccessApplication of neural networks in half marathon and marathon
Artificial intelligence is the fastest-growing field of computer science. Its use is growing and yet to be fully uncovered. Using deep learning and datasets of specific data, artificial neural networks can be trained to help users in everyday life. This paper aims to make neural networks to help get the best time in a half marathon and marathons. Many factors influence marathon race times, such as temperature, precipitation, sunshine or atmospheric pressure, sex, age and physical preparation and current condition of runners. A dataset from smartwatches will be used to determine the condition of runners, as well as a dataset Berlin marathon dataset. Combining the analysis of these two datasets, neural networks will predict if the ideal time was achieved.
- 28 September 2021
- Published by Wiley in International Journal of Numerical Modelling: Electronic Networks, Devices and Fields
- Vol. 35 (2) , e2958
Outage statistics of double gamma–gamma random process and its application to cooperative optical wireless communication relay systems
This article investigates outage statistics of double gamma–gamma random variable (RV) modeled as the product of two independent and identically distributed gamma–gamma RVs. The outage closed‐form statistics in terms of the first‐order statistical measures (probability density function and cumulative distribution function) and the second‐order statistical measures (level crossing rate and average fade duration) by Laplace approximation integral (LAI) and exponential LAI are derived. The statistical results are directly related to the performance evaluation of dual‐hop as well as cooperative dual‐hop optical wireless communications over moderate to strong turbulence‐induced fading channels.
- 28 September 2021
- Published by Wiley in International Journal of Numerical Modelling: Electronic Networks, Devices and Fields
- Vol. 35 (2) , e2958
Outage statistics of double gamma–gamma random process and its application to cooperative optical wireless communication relay systems
This article investigates outage statistics of double gamma–gamma random variable (RV) modeled as the product of two independent and identically distributed gamma–gamma RVs. The outage closed‐form statistics in terms of the first‐order statistical measures (probability density function and cumulative distribution function) and the second‐order statistical measures (level crossing rate and average fade duration) by Laplace approximation integral (LAI) and exponential LAI are derived. The statistical results are directly related to the performance evaluation of dual‐hop as well as cooperative dual‐hop optical wireless communications over moderate to strong turbulence‐induced fading channels.