- Personality Transfer in Human Animation
Arçin Ülkü Ergüzen
Ph.D. Student
(Supervisor:Prof. Dr. Uğur Güdükbay) Computer Engineering Department
Bilkent University
Abstract: Personality is the individual’s interrelated behavioral and emotional patterns that form the unique self. Personality-enriched animations benefit digital characters, thereby improving realism and communication. Body movements, among other modalities, can include strong cues for personality expression. We focus on altering the body movements of human animation to express the desired personality traits following two approaches: (i) a traditional approach that utilizes handcrafted motion adjustments following heuristic rules, and (ii) a data-driven approach that separates content and personality into different latent spaces to reconstruct the same motion with altered personality. While the sample size does not affect the traditional approach, the scarcity of personality-labeled animation datasets prevents the use of sophisticated data-driven models; to this end, we employ Neural Motion Fields (NeMF) in our data-driven personality transfer architecture. We evaluate the performance of the two approaches through a three-part user study; different models stand out for altering specific personality factors.
DATE: April 20, Monday @ 15:30 Place: EA 502
2. Agentic Evaluation of Software Engineering Artifacts via a Platform Testbed
Hesam Matinpouya
Ph.D. Student
(Supervisor:Asst.Prof.Anıl Koyuncu) Computer Engineering Department
Bilkent University
Abstract: Software engineering research often depends on people evaluating artifacts in tasks such as code summarization, commit annotation, semantic similarity of functions, code review, and UML/design evaluation, yet this process remains a critical bottleneck. These studies take time to run, and human-based annotation is expensive, slow, and difficult to scale; even expert annotators often produce inconsistent judgments. These challenges limit the reliability and reproducibility of empirical software engineering studies. Meanwhile, large language models (LLMs) have recently shown strong performance on many software engineering tasks, which raises an important question: can they also be used to evaluate software artifacts, either by acting like human annotators or by helping them during the evaluation process? In this work, I propose a platform-based agentic framework that provides a means of systematically evaluating software artifacts through the use of LLMs. Unlike previous approaches that just consider LLMs to produce a single direct answer, this approach views evaluation as an interactive process where agents are assigned roles and permits experimentation with a range of evaluation methods. It supports three modes of evaluation: (1) LLMs working as independent evaluators, (2) human evaluators assisted by LLMs, and (3) multi-agent systems with predefined roles that engage in deliberation before making their decision. Reproducibility and fine-grained analysis are guaranteed by means of standardization of tasks, conditioning of roles, and traceability of interactions throughout the framework. This research presents an innovative experiment that analyzes the LLMs’ abilities to act as evaluators across various software engineering tasks. It contributes empirical evidence about the applicability of agentic LLM-based systems to emulate or even improve upon human judgment.
DATE: April 20, Monday @ 15:50 Place: EA 502
3. GPU-Accelerated Topology-Preserving Loss Functions for Biomedical Image Segmentation
Ahmet Caner Akar
Master Student
(Supervisor:Asst.Prof.Doruk Öner) Computer Engineering Department
Bilkent University
Abstract: Preserving the topological integrity of curvilinear anatomical structures, such as blood vessels and neurons, is critical for biomedical image segmentation. While topology-aware loss functions such as CAPE, MALIS, and Persistent Homology effectively address this challenge, their high computational cost severely limits their practical adoption. In this work, we analyze the computational bottlenecks of these loss functions and develop custom CUDA kernels to accelerate their execution on GPUs. Specifically, we implement a parallel delta-stepping Dijkstra algorithm for shortest-path computation, which forms the core of the CAPE loss, integrate it with PyTorch’s autograd framework, and present comprehensive performance benchmarks against CPU baselines. Our preliminary results demonstrate that shortest path computation accounts for over 60% of total CPU execution time, and GPU parallelization has significant potential to eliminate this bottleneck, enabling topology-preserving models to be trained on large-scale 3D biomedical datasets within practical timeframes.
DATE: April 20, Monday @ 16:30 Place: EA 502