Scholarly Publications - Computer Engineering
Permanent URI for this collectionhttps://hdl.handle.net/11693/115582
Browse
Recent Submissions
Item Embargo Large-margin multiple kernel ℓp-SVDD using Frank–Wolfe algorithm for novelty detection(Elsevier BV, 2023-12-09) Rahimzadeh Arashloo, ShervinUsing a variable 𝓁𝑝≥1-norm penalty on the slacks, the recently introduced 𝓁𝑝-norm Support Vector Data Description (𝓁𝑝-SVDD) method has improved the performance in novelty detection over the baseline approach, sometimes remarkably. This work extends this modelling formalism in multiple aspects. First, a large-margin extension of the 𝓁𝑝-SVDD method is formulated to enhance generalisation capability by maximising the margin between the positive and negative samples. Second, based on the Frank–Wolfe algorithm, an efficient yet effective method with predictable accuracy is presented to optimise the convex objective function in the proposed method. Finally, it is illustrated that the proposed approach can effectively benefit from a multiple kernel learning scheme to achieve state-of-the-art performance. The proposed method is theoretically analysed using Rademacher complexities to link its classification error probability to the margin and experimentally evaluated on several datasets to demonstrate its merits against existing methods.Item Open Access Robust one-class classification using deep kernel spectral regression(ELSEVIER, 2024-03-07) Mohammad, Salman; Arashloo, Shervin RahimzadehThe existing one-class classification (OCC) methods typically presume the existence of a pure target training set and generally face difficulties when the training set is contaminated with non-target objects. This work addresses this aspect of the OCC problem and formulates an effective method that leverages the advantages of kernel-based methods to achieve robustness against training label noise while enabling direct deep learning of features from the data to optimise a Fisher-based loss function in the Hilbert space. As such, the proposed OCC approach can be trained in an end-to-end fashion while, by virtue of a Tikhonov regularisation in the Hilbert space, it provides high robustness against the training set contamination. Extensive experiments conducted on multiple datasets in different application scenarios demonstrate that the proposed methodology is robust and performs better than the state-of-the-art algorithms for OCC when the training set is corrupted by contamination.Item Open Access ECOLE: Learning to call copy number variants on whole exome sequencing data(NATURE PORTFOLIO, 2024-01-02) Mandıracıoğlu, Berke; Özden, Furkan; Kaynar, Gün; Yılmaz, Mehmet Alper; Alkan, Can; Çiçek, A.ErcümentCopy number variants (CNV) are shown to contribute to the etiology of several genetic disorders. Accurate detection of CNVs on whole exome sequencing (WES) data has been a long sought-after goal for use in clinics. This was not possible despite recent improvements in performance because algorithms mostly suffer from low precision and even lower recall on expert-curated gold standard call sets. Here, we present a deep learning-based somatic and germline CNV caller for WES data, named ECOLE. Based on a variant of the transformer architecture, the model learns to call CNVs per exon, using high-confidence calls made on matched WGS samples. We further train and fine-tune the model with a small set of expert calls via transfer learning. We show that ECOLE achieves high performance on human expert labelled data for the first time with 68.7% precision and 49.6% recall. This corresponds to precision and recall improvements of 18.7% and 30.8% over the next best-performing methods, respectively. We also show that the same fine-tuning strategy using tumor samples enables ECOLE to detect RT-qPCR-validated variations in bladder cancer samples without the need for a control sample. ECOLE is available at https://github.com/ciceklab/ECOLE. Copy number variants (CNV) are shown to contribute to the etiology of various genetic disorders. Here, authors present ECOLE, a deep learning-based somatic and germline CNV caller for WES data. Utilising a variant of the transformer architecture, the model is trained to call CNVs per exon.Item Open Access Human movement personality detection parameters(IEEE, 2024-06-23) Sonlu, Sinan; Dogan, Yalım; Ergüzen, Arçin Ülkü; Ünalan, Musa Ege; Demirci, Serkan; Durupinar, Funda; Güdükbay, UğurIn this study, we develop a system that detects apparent personality traits from animation data containing human movements. Since the datasets that can be used for this purpose lack sufficient variance, we determined labels for the samples in two datasets containing human animations, in terms of the Five Factor Personality Theory, with the help of a user study. Using these labels, we identified movement parameters highly dependent on personality traits and based on Laban Movement Analysis categories. The artificial neural networks we trained for personality analysis from animation data show that models that take the motion parameters determined in the study as input have a higher accuracy rate than models that take raw animation data as input. Therefore, using the parameters determined in this study to evaluate human movements in terms of their personality traits will increase the systems' success.Item Open Access Contact energy based hindsight experience prioritization(IEEE, 2024-08-08) Sayar, Erdi; Bing, Zhenshan; D'Eramo, Carlo; Öğüz, Salih Özgür; Knoll, AloisMulti-goal robot manipulation tasks with sparse rewards are difficult for reinforcement learning (RL) algorithms due to the inefficiency in collecting successful experiences. Recent algorithms such as Hindsight Experience Replay (HER) expedite learning by taking advantage of failed trajectories and replacing the desired goal with one of the achieved states so that any failed trajectory can be utilized as a contribution to learning. However, HER uniformly chooses failed trajectories, without taking into account which ones might be the most valuable for learning. In this paper, we address this problem and propose a novel approach Contact Energy Based Prioritization (CEBP) to select the samples from the replay buffer based on rich information due to contact, leveraging the touch sensors in the gripper of the robot and object displacement. Our prioritization scheme favors sampling of contact-rich experiences, which are arguably the ones providing the largest amount of information. We evaluate our proposed approach on various sparse reward robotic tasks and compare it with the state-of-the-art methods. We show that our method surpasses or performs on par with those methods on robot manipulation tasks. Finally, we deploy the trained policy from our method to a real Franka robot for a pick-and-place task. We observe that the robot can solve the task successfully. The videos and code are publicly available at: https://erdiphd.github.io/HER force/.Item Open Access Ai-assisted text composition for automated content authoring using transformer-based language models(IEEE, 2024-06-23) Alpdemir, Yusuf; Alpdemir, Mahmut NedimIn this paper, we introduce a hybrid method that combines the use of Controllable Text Generation (CTG) approach via Large Language Models (LLMs), fine-tuned language models and sentence transformers in a single framework to generate real-author styled articles in Turkish language. As such, we seek to exemplify hybrid solutions that produce real-human styled high-quality contents, given limited resources and relatively short text prompts as inputs. To achieve this, we introduce a novel method to assemble an author-specific article in different coherence and fluency levels, based on phrasal control of the CTG process. Control phrases are automatically assembled based on a semantic correlation measure calculated using sentence embed dings corresponding to author articles, that are obtained from pre-trained sentence transformers.Item Embargo A Process Model for AI-Enabled Software Development: A Synthesis From Validation Studies in White Literature(John Wiley & Sons Ltd., 2024-11-18) Erdogan, Tugba G.; Altunel, Haluk; Tarhan, Ayça K.Context: With the fast advancement of techniques in artificial intelligence (AI) and of the target infrastructures in the last decades, AI software is becoming an undeniable part of software system projects. As in most cases in history, however, development methods and guides follow the advancements in technology with phase differences. Purpose: With an aim to elicit and integrate available evidence from AI software development practices into a process model, this study synthesizes the contributions of the validation studies reported in scientific literature. Method: We applied a systematic literature review to retrieve, select, and analyze the primary studies. After a comprehensive and rigorous search and scoping review, we identified 82 studies that make various contributions in relation to AI software development practices. To increase the effectiveness of the synthesis and the usefulness of the outcome, for detailed analysis, we selected 14 primary studies (out of 82) that empirically validated their contributions. Results: We carefully reviewed the selected studies that validate proposals on approaches/models, methods/techniques, tasks/phases, lessons learned/best practices, or workflows. We mapped the steps/activities in these proposals with the knowledge areas in SWEBOK, and using the evidence in this mapping and the primary studies, we synthesized a process model that integrates activities, artifacts, and roles for AI-enabled software system development. Cunclusion: To the best of our knowledge, this is the first study that proposes such a process model by eliciting and gathering the contributions of the validation studies in a bottom-up manner. We expect that the output of this synthesis will be input for further research to validate or improve the process model.Item Embargo A serious game approach to introduce the code review practice(John Wiley & Sons Ltd., 2024-12-22) Ardic, Barış; Tüzün, ErayCode review is a widely utilized practice that focuses on improving code via manual inspections. However, this practice is not addressed adequately in a typical software engineering curriculum. We aim to help address the code review practice knowledge gap between the software engineering curricula and the industry with a serious game approach. We determine our learning objectives around the introduction of the code review process. To realize these objectives, we design, build, and test the serious game. We then conduct three case studies with a total of 280 students. We evaluated the results by comparing the student's knowledge and confidence about code review before and after case studies, as well as evaluating how they performed in code review quizzes and game levels themselves. Our analysis indicates that students had a positive experience during gameplay, and an in-depth examination suggests that playing the game also enhanced their knowledge. We conclude that the game had a positive impact on introducing the code review process. This study represents a step taken toward moving code review education from industry starting positions to higher education. The game and its auxiliary materials are available online.Item Open Access Learning visual similarity for image retrieval with global descriptors and capsule networks(Springer New York LLC, 2024-02) Durmuş, Duygu; Güdükbay, Uğur; Ulusoy, ÖzgürFinding matching images across large and unstructured datasets is vital in many computer vision applications. With the emergence of deep learning-based solutions, various visual tasks, such as image retrieval, have been successfully addressed. Learning visual similarity is crucial for image matching and retrieval tasks. Capsule Networks enable learning richer information that describes the object without losing the essential spatial relationship between the object and its parts. Besides, global descriptors are widely used for representing images. We propose a framework that combines the power of global descriptors and Capsule Networks by benefiting from the information of multiple views of images to enhance the image retrieval performance. The Spatial Grouping Enhance strategy, which enhances sub-features parallelly, and self-attention layers, which explore global dependencies within internal representations of images, are utilized to empower the image representations. The approach captures resemblances between similar images and differences between non-similar images using triplet loss and cost-sensitive regularized cross-entropy loss. The results are superior to the state-of-the-art approaches for the Stanford Online Products Database with Recall@K of 85.0, 94.4, 97.8, and 99.3, where K is 1, 10, 100, and 1000, respectively.Item Open Access A quantitative style analysis of four Turkish authors: changes over time, and differences(Routledge, 2024-10-14) Yıldırım, Onur; Can, FazlıWe present a stylometric analysis of the writings of four famous Turkish authors: Abdülhak Şinasi Hisar, Refik Halid Karay, Ahmet Hamdi Tanpınar, and Halit Ziya Uşaklıgil. Our aim is to internally analyse the shifts in their writing styles and examine the differences between them. First, we evaluate the changes in word lengths in the writers’ novels over time and observe that they do not necessarily follow the pattern of writing with longer words as time passes, which was common for 20th-century Turkish literature. We then employ a sliding text window approach to capture the shift in writing styles in novels, by focusing changes in word lengths throughout the entire text. Based on this analysis, we hypothesize a relationship between changes in word lengths and meaning shift within a novel. Next, we investigate the stylochronometry and authorship attribution problems for these four authors and show that their styles change with time and that their works are distinguishable from each other. Finally, we analyse differences in their vocabulary richness within close contexts and demonstrate a strong relationship between poetic writing and lower vocabulary richness in the running text.Item Open Access Towards understanding personality expression via body motion(2024-05-29) Sonlu, Sinan; Dogan, Yalim; Erguzen, Arcin Ulku; Unalan, Musa Ege; Demirci, Serkan; Durupinar, Funda; Gudukbay, UgurThis work addresses the challenge of data scarcity in personality-labeled datasets by introducing personality labels to clips from two open datasets, ZeroEGGS and Bandai, which provide diverse full-body animations. To this end, we present a user study to annotate short clips from both sets with labels based on the Five-Factor Model (FFM) of personality. We chose features informed by Laban Movement Analysis (LMA) to represent each animation. These features then guided us to select the samples of distinct motion styles to be included in the user study, obtaining high personality variance and keeping the study duration and cost viable. Using the labeled data, we then ran a correlation analysis to find features that indicate high correlation with each personality dimension. Our regression analysis results indicate that highly correlated features are promising in accurate personality estimation. We share our early findings, code, and data publicly.Item Open Access Towards unmasking LGTM smells in code reviews: a comparative study of comment-free and commented reviews(Institute of Electrical and Electronics Engineers Inc., 2024-10-11) Gon, Mahmut Furkan; Tuzun, Eray; Yetistiren, BurakCode review is a crucial component of the software development life cycle and is adopted as a best practice in the industry. However, like any process, counterproductive practices and pitfalls exist within code review, such as the occurrence of the 'Looks Good to Me' (LGTM) smell. LGTM smell occurs when a superficial review is conducted. LGTM review smells can potentially result in the accidental inclusion of low-quality changesets in the codebase, leading to severe bugs, possibly many reopens of the associated issues, and additional time wasted on changes to the changeset. In this study, we aim to explore LGTM smells and examine their potential impacts on the related repository. Given the inherent challenges of automatically detecting LGTM smells in code reviews, this study introduces an alternative approach by categorizing code reviews into two distinct types: comment-free and commented reviews. The primary hypothesis is that comment-free reviews are more prone to LGTM smells due to their lack of detailed examination and discourse. To test this hypothesis, we conduct an empirical analysis on a subset of pull requests (PRs) comprising code reviews from five large-scale software projects. We further investigate the impact of comment-free and commented reviews on key development metrics, specifically focusing on the number of reopens and late commits in PRs. According to the results, 64.7% of the PRs in these projects exhibited comment-free reviews. Our manual analysis reveals that comment-free reviews exhibit the LGTM smell 3.5 times more frequently than the commented reviews. We also observed a statistically significant difference, indicating that comment-free review PRs tend to include more late commits (i.e., commits made after the reviewer's approval) than the commented PRs. However, no statistically significant difference was observed in the reopening ratio of associated issues between comment-free reviews and commented reviews. Our approach provides a novel method for detecting and exploring the impacts of LGTM smell, emphasizing the significance of comprehensive code reviews and setting the stage for future research aimed at automatically identifying LGTM smell occurrences. © 2024 IEEE.Item Open Access Smolbot-vs: a soft modular robot capable of modulating backbone stiffness(2024-08-15) Uygun, Muhammed; Ozcan, OnurHandling complex terrains is often problematic for miniature mobile robots due to their small size and low weight. Soft robots can utilize their compliance to overcome obstacles in the terrain, but their performances decline for tasks that require load-bearing capabilities. In this study, we present a miniature modular robot with tendon-driven variable stiffness backbones and investigate the effectiveness of the rigid and soft configurations for climbing obstacles and crossing gaps. The mechanism utilizes the design of a 3D-printed soft backbone with a layered structure that becomes rigid under compression applied by the linear actuator in the modules. In rigid mode, the robot can climb a 20mm step obstacle and cross a 105mm gap. In contrast, the soft mode obstacle height threshold jumps to 30mm, and the length of the gap that the robot can cross decreases to 55mm, showing that backbone stiffness modulation allows better adaptability for complex terrains.Item Embargo Diversity-aware strategies for static index pruning(Elsevier Ltd, 2024-05-30) Yiğit-Sert, Sevgi; Altıngövde, İsmail Şengör; Ulusoy, ÖzgürStatic index pruning aims to remove redundant parts of an index to reduce the file size and query processing time. In this paper, we focus on the impact of index pruning on the topical diversity of query results obtained over these pruned indexes, due to the emergence of diversity as an important metric of quality in modern search systems. We hypothesize that typical index pruning strategies are likely to harm result diversity, as the latter dimension has been vastly overlooked while designing and evaluating such methods. As a remedy, we introduce three novel diversity-aware pruning strategies aimed at maintaining the diversity effectiveness of query results. In addition to other widely used features, our strategies exploit document clustering methods and word-embeddings to assess the possible impact of index elements on the topical diversity, and to guide the pruning process accordingly. Our thorough experimental evaluations verify that typical index pruning strategies lead to a substantial decline (i.e., up to 50% for some metrics) in the diversity of the results obtained over the pruned indexes. Our diversity-aware approaches remedy such losses to a great extent, and yield more diverse query results, for which scores of the various diversity metrics are closer to those obtained over the full index. Specifically, our best-performing strategy provides gains in result diversity reaching up to 2.9%, 3.0%, 7.5%, and 3.9% wrt. the strongest baseline, in terms of the ERR-IA, alpha-nDCG, P-IA, and ST-Recall metrics (at the cut-off value of 20), respectively. The proposed strategies also yield better scores in terms of an entropy-based fairness metric, confirming the correlation between topical diversity and fairness in this setup.Item Open Access Skydatanet: an object detection algorithm with 2d gaussian loss for uav-based aerial ımages(2025-01-09) Ozkanoglu, Mehmet Akif; Begen, Ali C.; Ozer, SedatIn this paper, we introduce a novel object detection algorithm based on the center-point detection. In our architecture, we introduce using two HourGlass architecture as the backbone, and we introduce using a new module to unify the predictions made after each backbone. Furthermore, since bounding boxes are in varying aspect ratios, as opposed to using a scalar Gaussian variance, we introduce using 2D variance in the Gaussian loss function to predict center-points in our network. We present the performance of our proposed improvements on three aerial datasets by comparing them to center-point based detection algorithms.Item Embargo lp-norm constrained one-class classifier combination(Elsevier BV, 2024-10-16) Nourmohammadi, Sepehr; Rahimzadeh Arashloo, Shervin; Kittler, JosefClassifier fusion is established as an effective methodology for boosting performance in different classification settings and one-class classification is no exception. In this study, we consider the one-class classifier fusion problem by modelling the sparsity/uniformity of the ensemble. To this end, we formulate a convex objective function to learn the weights in a linear ensemble model and impose a variable l(p >= 1)-norm constraint on the weight vector. The vector-norm constraint enables the model to adapt to the intrinsic uniformity/sparsity of the ensemble in the space of base learners and acts as a (soft) classifier selection mechanism by shaping the relative magnitudes of fusion weights. Drawing on the Frank-Wolfe algorithm, we then present an effective approach to solve the proposed convex constrained optimisation problem efficiently. We evaluate the proposed one-class classifier combination approach on multiple data sets from diverse application domains and illustrate its merits in comparison to the existing approaches.Item Open Access LLMs and prompting for unit test generation: a large-scale evaluation(Association for Computing Machinery, Inc, 2024-11-01) Koyuncu, Anıl; Ouedraogo, Wendkuuni C.; Kabore, Kader; Tian, Haoye; Song, Yewei; Klein, Jacques; Lo, David; Bissyande, Tegawende F.Unit testing, essential for identifying bugs, is often neglected due to time constraints. Automated test generation tools exist but typically lack readability and require developer intervention. Large Language Models (LLMs) like GPT and Mistral show potential in test generation, but their effectiveness remains unclear.This study evaluates four LLMs and five prompt engineering techniques, analyzing 216 300 tests for 690 Java classes from diverse datasets. We assess correctness, readability, coverage, and bug detection, comparing LLM-generated tests to EvoSuite. While LLMs show promise, improvements in correctness are needed. The study highlights both the strengths and limitations of LLMs, offering insights for future research. © 2024 Copyright is held by the owner/author(s). Publication rights licensed to ACM.Item Open Access Systematic analysis of speech transcription modeling for reliable assessment of depression severity(Sakarya University, 2024-04-27) Kaynak, Ergün Batuhan; Dibeklioğlu, HamdiIn evaluating the severity of depression, we rigorously investigate a segmented deep learning framework that employs speech transcriptions for predicting levels of depression. Within this framework, we examine the effectiveness of well-known deep learning models for generating useful features for gauging depression. We validate the chosen models using the openly accessible Extended Distress Analysis Interview Corpus (EDAIC) as a dataset. Through our findings and analytical commentary, we demonstrate that valuable features for depression severity estimation can be achieved without leveraging the sequential relationships among textual descriptors. Specifically, temporal aggregation of latent representations surpasses the current best performing methods that utilize recurrent models, exhibiting an 8.8% improvement in Concordance Correlation Coefficient (CCC).Item Open Access Deep Convolutional Networks for PET Super-Resolution(SPIE; Philips Res; Merck & Co Inc; Guerbet Grp; GE Res, 2024) Özaltan, Cemhan Kaan; Türkölmez, Emir; Aksoy, Selim; Çiçek, A. Ercüment; Namer,I. Jacques; Colliot, OlivierPositron emission tomography (PET) provides valuable functional information that is widely used in clinical domains such as oncology and neurology. However, the structural quality of PET images may not be sufficient to effectively evaluate small regions of interest. Image super-resolution techniques aim to recover a high-resolution image from an input low-resolution version. We study adaptations of deep convolutional neural network architectures for improving the spatial resolution of PET images. The proposed super-resolution model involves a deep architecture that uses convolutional blocks together with various residual connections for more effective and efficient training. We use the supervised setting where the downscaled versions of the original PET images are given as the low-resolution input to the deep networks and the original images are used as the high-resolution target data to be recovered. Experiments show that the proposed model performs better than a multi-scale convolutional architecture according to both quantitative performance metrics and visual qualitative evaluation.Item Embargo Personality perception in human videos altered by motion transfer networks(Elsevier Ltd, 2024-02-01) Yurtoğlu, Ayda; Sonlu, Sinan; Doğan, Yalım; Güdükbay, UğurThe successful portrayal of personality in digital characters improves communication and immersion. Current research focuses on expressing personality through modifying animations using heuristic rules or data-driven models. While studies suggest motion style highly influences the apparent personality, the role of appearance can be similarly essential. This work analyzes the influence of movement and appearance on the perceived personality of short videos altered by motion transfer networks. We label the personalities in conference video clips with a user study to determine the samples that best represent the Five-Factor model’s high, neutral, and low traits. We alter these videos using the Thin-Plate Spline Motion Model, utilizing the selected samples as the source and driving inputs. We follow five different cases to study the influence of motion and appearance on personality perception. Our comparative study reveals that motion and appearance influence different factors: motion strongly affects perceived extraversion, and appearance helps convey agreeableness and neuroticism.