arXiv subjects

Kapil Dev

Publications and source records attributed to Kapil Dev.

8 recordsLinked to original sources

arXiv2026

Normalisation and Initialisation Strategies for Graph Neural Networks in Blockchain Anomaly Detection

Graph neural networks (GNNs) offer a principled approach to financial fraud detection by jointly learning from node features and transaction graph topology. However, their effectiveness on real-world anti-money laundering (AML) benchmarks depends critically on training practices such as specifically weight initialisation and normalisation that remain underexplored. We present a systematic ablation of initialisation and normalisation strategies across three GNN architectures (GCN, GAT, and GraphSAGE) on the Elliptic Bitcoin dataset. Our experiments reveal that initialisation and normalisation are architecture-dependent: GraphSAGE achieves the strongest performance with Xavier initialisation alone, GAT benefits most from combining GraphNorm with Xavier initialisation, while GCN shows limited sensitivity to these modifications. These findings offer practical, architecture-specific guidance for deploying GNNs in AML pipelines for datasets with severe class imbalance. We release a reproducible experimental framework with temporal data splits, seeded runs, and full ablation results.

cs.LG↗

arXiv2025

Hybridizing Expressive Rendering: Stroke-Based Rendering with Classic and Neural Methods

Non-Photorealistic Rendering (NPR) has long been used to create artistic visualizations that prioritize style over realism, enabling the depiction of a wide range of aesthetic effects, from hand-drawn sketches to painterly renderings. While classical NPR methods, such as edge detection, toon shading, and geometric abstraction, have been well-established in both research and practice, with a particular focus on stroke-based rendering, the recent rise of deep learning represents a paradigm shift. We analyze the similarities and differences between classical and neural network based NPR techniques, focusing on stroke-based rendering (SBR), highlighting their strengths and limitations. We discuss trade offs in quality and artistic control between these paradigms, propose a framework where these approaches can be combined for new possibilities in expressive rendering.

cs.GR↗

arXiv2025

Modeling Aesthetic Preferences in 3D Shapes: A Large-Scale Paired Comparison Study Across Object Categories

Human aesthetic preferences for 3D shapes are central to industrial design, virtual reality, and consumer product development. However, most computational models of 3D aesthetics lack empirical grounding in large-scale human judgments, limiting their practical relevance. We present a large-scale study of human preferences. We collected 22,301 pairwise comparisons across five object categories (chairs, tables, mugs, lamps, and dining chairs) via Amazon Mechanical Turk. Building on a previously published dataset~\cite{dev2020learning}, we introduce new non-linear modeling and cross-category analysis to uncover the geometric drivers of aesthetic preference. We apply the Bradley-Terry model to infer latent aesthetic scores and use Random Forests with SHAP analysis to identify and interpret the most influential geometric features (e.g., symmetry, curvature, compactness). Our cross-category analysis reveals both universal principles and domain-specific trends in aesthetic preferences. We focus on human interpretable geometric features to ensure model transparency and actionable design insights, rather than relying on black-box deep learning approaches. Our findings bridge computational aesthetics and cognitive science, providing practical guidance for designers and a publicly available dataset to support reproducibility. This work advances the understanding of 3D shape aesthetics through a human-centric, data-driven framework.

cs.GR↗

arXiv2020

Design of a Scalable DNA Shearing System Using Phased-Array Fresnel Lens Transducers

In this paper, we present the design of a Deoxyribonucleic Acid (DNA) shearing system based on unique acoustic waves generated using a phased-array Fresnel Lens transducer. Four 90-degree sector-transducers are used to build a circular array transducer. Acoustic simulation results for particle displacement are provided for cases when one, two, three, and all four transducers in the array are excited with RF signals. Each 90degree-transducer is excited with separate RF signal of same or different phase. The proposed transducer structure generates bulk lateral ultrasonic waves in the DNA sample; the lateral waves produce both convergence and vortexing effects in the sample. The converged lateral acoustic waves are required to break the DNA sample-meniscus inside the tube and the rotational component of acoustic field is used to recirculate the DNA sample to get homogeneous shearing with tight fragment distribution. Finally, we present the experimental results of DNA sheared to different mean fragment sizes using the proposed system and validate the shearing-capability of the proposed system.

physics.ins-det↗

arXiv2019

Analytical Modeling and Design of Fresnel Lens Transducers

In this paper, we present an analytical modeling technique for circularly symmetric piezoelectric transducers, also called as Fresnel Lens. We also present the design of a flat/piston transducer that can generate unique acoustic wave patterns, having both converging and vortexing effects. The converging effect is generated by designing the transducer electrodes in the shapes of circular rings using Fresnel formula and exciting it with an RF signal of resonant frequency. The vortexing effect is achieved by cutting the rings to different sector angles: 90, 120, 180 and 270 degrees. We use the analytical model to simulate the performance of these transducers.

physics.app-ph↗

arXiv2018

Implications of Integrated CPU-GPU Processors on Thermal and Power Management Techniques

Heterogeneous processors with architecturally different cores (CPU and GPU) integrated on the same die lead to new challenges and opportunities for thermal and power management techniques because of shared thermal/power budgets between these cores. In this paper, we show that new parallel programming paradigms (e.g., OpenCL) for CPU-GPU processors create a tighter coupling between the workload, the thermal/power management unit and the operating system. Using detailed thermal and power maps of the die from infrared imaging, we demonstrate that in contrast to traditional multi-core CPUs, heterogeneous processors exhibit higher coupled behavior for dynamic voltage and frequency scaling and workload scheduling, in terms of their effect on performance, power, and temperature. Further, we show that by taking the differences in core architectures and relative proximity of different computing cores on the die into consideration, better scheduling schemes could be implemented to reduce both the power density and peak temperature of the die. The findings presented in the paper can be used to improve thermal and power efficiency of heterogeneous CPU-GPU processors.

cs.AR↗

arXiv2016

A Perceptual Aesthetics Measure for 3D Shapes

While the problem of image aesthetics has been well explored, the study of 3D shape aesthetics has focused on specific manually defined features. In this paper, we learn an aesthetics measure for 3D shapes autonomously from raw voxel data and without manually-crafted features by leveraging the strength of deep learning. We collect data from humans on their aesthetics preferences for various 3D shape classes. We take a deep convolutional 3D shape ranking approach to compute a measure that gives an aesthetics score for a 3D shape. We demonstrate our approach with various types of shapes and for applications such as aesthetics-based visualization, search, and scene composition.

cs.GR↗

arXiv2015

Improving Style Similarity Metrics of 3D Shapes

The idea of style similarity metrics has been recently developed for various media types such as 2D clip art and 3D shapes. We explore this style metric problem and improve existing style similarity metrics of 3D shapes in four novel ways. First, we consider the color and texture of 3D shapes which are important properties that have not been previously considered. Second, we explore the effect of clustering a dataset of 3D models by comparing between style metrics for a single object type and style metrics that combine clusters of object types. Third, we explore the idea of user-guided learning for this problem. Fourth, we introduce an iterative approach that can learn a metric from a general set of 3D models. We demonstrate these contributions with various classes of 3D shapes and with applications such as style-based similarity search and scene composition.

cs.GR↗