GDGoran Delac

Goran Delac

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Goran Delac has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Large Language Models as Attribution Regularizers for Efficient Model Training

    by Davor Vukadin, Marin Šilić, Goran Delač

    Large Language Models (LLMs) have demonstrated remarkable performance across diverse domains. However, effectively leveraging their vast knowledge for training smaller downstream models remains an open challenge, especially in domains like tabular data learning, where simpler models are often preferred due to interpretability and efficiency. In this paper, we introduce a novel yet straightforward method for incorporating LLM-generated global task feature attributions into the training process of smaller networks. Specifically, we propose an attribution-matching regularization term that aligns the training dynamics of the smaller model with the insights provided by the LLM. By doing so, our approach yields superior performance in few-shot learning scenarios. Notably, our method requires only black-box API access to the LLM, making it easy to integrate into existing training pipelines with minimal computational overhead. Furthermore, we demonstrate how this method can be used to address common issues in real-world datasets, such as skewness and bias. By integrating high-level knowledge from LLMs, our approach improves generalization, even when training data is limited or imbalanced. We validate its effectiveness through extensive experiments across multiple tasks, demonstrating improved learning efficiency and model robustness.


  • Open Access

    Advancing Attribution-Based Neural Network Explainability through Relative Absolute Magnitude Layer-Wise Relevance Propagation and Multi-Component Evaluation

    by Davor Vukadin, Petar Afrić, Marin Šilić, Goran Delač

    Recent advancement in deep-neural network performance led to the development of new state-of-the-art approaches in numerous areas. However, the black-box nature of neural networks often prohibits their use in areas where model explainability and model transparency are crucial. Over the years, researchers proposed many algorithms to aid neural network understanding and provide additional information to the human expert. One of the most popular methods being Layer-Wise Relevance Propagation (LRP). This method assigns local relevance based on the pixel-wise decomposition of nonlinear classifiers. With the rise of attribution method research, there has emerged a pressing need to assess and evaluate their performance. Numerous metrics have been proposed, each assessing an individual property of attribution methods such as faithfulness, robustness, or localization. Unfortunately, no single metric is deemed optimal for every case, and researchers often use several metrics to test the quality of the attribution maps. In this work, we address the shortcomings of the current LRP formulations and introduce a novel method for determining the relevance of input neurons through layer-wise relevance propagation. Furthermore, we apply this approach to the recently developed Vision Transformer architecture and evaluate its performance against existing methods on two image classification datasets, namely ImageNet and PascalVOC. Our results clearly demonstrate the advantage of our proposed method. Furthermore, we discuss the insufficiencies of current evaluation metrics for attribution-based explainability and propose a new evaluation metric that combines the notions of faithfulness, robustness, and contrastiveness. We utilize this new metric to evaluate the performance of various attribution-based methods. Our code is available at: https://github.com/davor10105/relative-absolute-magnitude-propagation


  • Open Access

    Empirical Study: How Issue Classification Influences Software Defect Prediction

    by Petar Afric, Davor Vukadin, Marin Silic, Goran Delac

    Software defect prediction aims to identify potentially defective software modules to better allocate limited quality assurance resources. Practitioners often do this by utilizing supervised models trained using historical data. This data is gathered by mining version control and issue tracking systems. Version control commits are linked to issues they address. If the linked issue is classified as a bug report, the change is considered as bug fixing. The problem arises from the fact that issues are often incorrectly classified within issue tracking systems. This introduces noise into the gathered datasets. In this paper, we investigate the influence issue classification has on software defect prediction dataset quality and resulting model performance. To do this, we mine data from 7 popular open-source repositories, create issue classification and software defect prediction datasets for each of them. We investigate issue classification using four different methods; a simple keyword heuristic, an improved keyword heuristic, the FastText model and the RoBERTa model. Our results show that using the RoBERTa model for issue classification produces the best software defect prediction datasets, containing on average 14.3641% of mislabeled instances. SDP models trained on such datasets achieve superior performance, to those trained on SDP datasets created using other issue classification methods, in 65 out of 84 experiments, with 55 of them being statistically relevant. Furthermore, in 17 out of 28 experiments we could not show a statistically relevant performance difference between SDP models trained on RoBERTa derived software defect prediction datasets and those created using manually labeled issues.


  • Agent-Based Modelling: Parallel and Distributed Simulation of Product Development Team

    by Ivan Čeh, Mario Štorga, Goran Delač

    The successful collaboration of team members plays a crucial role in product development. Research of different teamwork factors may help understand the process, and computer simulation can be a suitable method of achieving this goal. This work aims...


  • Open Access

    Named Entity Recognition for Addresses: An Empirical Study

    by Helena Ceovic, Adrian Satja Kurdija, Goran Delac, Marin Silic

    In this paper, several model architectures are explored in order to design a high-performing named entity recognition model for addresses which deals with challenges such as diversity, ambiguity and complexity of the address entity. Different types of neural networks are used for training the classifier, including the bidirectional LSTM network in combination with a convolutional layer, a conditional random field layer and different word embeddings. Experiments are conducted on two types of corpora specifically constructed and tagged for tackling this challenge: unstructured and semi-structured datasets. For model evaluation, two versions of the unstructured dataset are used that are tagged differently based on the granularity of address entity: entire address, and address consisting of subparts. For both types of corpora, the best results are achieved on a BiLSTM-CRF architecture model with a single RNN layer trained with BERT embeddings.


  • Open Access

    Repdx - Source Code Defect Prediction

    by Petar Afrić, Adrian Satja Kurdija, Marin Silic, Goran Delac

    This paper presents a new defect prediction model based on the REPD model we have previously proposed. The new model is named REPDX, which stands for ”Reconstruction Error Probability Distribution Expanded”. The model achieves the F1 score improvement over REPD of up to 16.82%. We investigate how the model allows for putting emphasis on either precision or recall. Given the practical importance of crossproject defect prediction, we investigate the performance of REPDX compared to TCA+. Our experiments show that REPDX achieves superior statistically relevant F1 scores in 25 out of 25 scenarios, improving the F1 score by as much as ≈30%.

  • Open Access

    Information Extraction From Free-Form CV Documents in Multiple Languages

    by Davor Vukadin, Adrian Satja Kurdija, Goran Delac, Marin Silic

    This paper proposes two natural language processing models for extracting useful information from multilingual, unstructured (free form) CV documents. The model identifies the relevant document sections (personal information, education, employment, etc.) and the corresponding specific information at the lower hierarchy level (names, addresses, roles, skill competences, etc.). Our approach employs the transformer architecture and its multilingual implementation of the encoder part in the form of the BERT language model. The models are trained and tested on a large, manually annotated CV dataset, achieving high scores on standard accuracy measures. The proposed models exhibit important properties of end-to-end training and interpretability, which was investigated by visualizing the model attention and its vector representations.


  • Fast Multi-Criteria Service Selection for Multi-User Composite Applications

    by Adrian Satja Kurdija, Marin Silic, Goran Delac, Klemo Vladimir

    As cloud computing becomes the prevailing aspect of software engineering, paradigms such as Service-Based Systems (SBSs) or Software as a Service (SaaS) are coming into focus. They are based on cloud services responding to numerous client requests. Selecting the actual service instance for request can be an issue, if requirements for multiple Quality of Service (QoS) attributes need to be satisfied for many users simultaneously. The problem becomes more complex if we take into account the compositeness of users' applications, consisting of many tasks, where QoS properties are calculated over the whole composition. The existing approaches for this problem lack either efficiency or generality. In this paper, we propose a fast heuristic method for multi-criteria service selection, designed for multi-user composite workflows with the goal of satisfying all, or as many as possible, of the given QoS requirements. The proposed method reduces the problem to several independent transportation problems, using a global-aware utility cost based on expected compositional QoS, and iterative solution improvements. Apart from being more general than the existing approaches, the proposed method turns out to be more efficient than the alternatives (up to 5x faster), as shown by extensive experiments covering both special and more general cases.