Mukthikka V

Mukthikka V

Auto-generated profile, based on publicly available information

Mukthikka V

School of Aeronautical Engineering; Bharath Institute of Higher Education and Technology, India

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 May 2026 in International Journal of Innovative Research in Engineering and Management
Verified
Journal Article
Closed access
MVMukthikka V
Mukthikka V
GSGurpreet Singh
Gurpreet Singh
Multimodal large language models (MLLMs) are currently dominated by visual instruction tuning (VIT), where frozen vision and language backbones are bridged by lightweight trainable modules. We curated and downloaded 30 recent papers (2022–2025) in this direction, with a cumulative 4,555 citations as of April 28, 2026, and analyzed architecture and training trends. The strongest recurring pattern is a shift from full-model tuning to parameter-efficient adaptation, token compression, and data selection. Motivated by this, we propose TinyBridge-TriFuse, a frozen-backbone connector family that combines linear pooled-feature alignment, a small MLP expert, and a token-aware bridge expert. We provide a full IEEE-style formulation, a data-efficient training recipe, and real measured experiments. On frozen OpenCLIP ViT-B/32 features for CIFAR-10, TinyBridge-TriFuse reaches 0.9390 test accuracy, exceeding LinearAlign (0.9350) by +0.0040 and improving over zero-shot CLIP by +0.0728. We also report TinyBridge-DynaFuse, a tiny-gate variant (1,035 extra parameters) that improves calibration to 0.0326 ECE.

1 June 2026 in International Journal of Innovative Research in Engineering and Management
Verified
Journal Article
Open access
PDF file
PTPiyush Thapliyal
Piyush Thapliyal
PMPurva Mundada
Purva Mundada
MVMukthikka V
Mukthikka V
GSGurpreet Singh
Gurpreet Singh
Current systems that generate content across multiple modes, like video and audio, often face issues such as missing objects, weak connections between meanings, and poor coordination between visual and sound elements. These problems often arise because visual and audio parts are created separately, leading to mismatches between what is shown and what is heard. To tackle these issues, this paper introduces SG-WORLD V2, a framework that focuses on planning first for semantic reasoning and coordination in structured world simulations. Instead of directly creating video or audio, SG-WORLD V2 builds a semantic blueprint first. It translates natural language into machine-readable forms using reasoning based on the Universal Scene Graph (USG), adds a Quantification Rule to keep track of requested items; and uses Deterministic Pre-Temporal Synchronization (DPTS) to align audio and visual events during planning. The system also includes a self-correcting mechanism to check for semantic consistency before generating outputs. The framework was tested with 54 different prompts that cover object arrangements, spatial connections, actions, and synchronized audio-video situations. The results showed strong performance in keeping track of objects, maintaining relationships, and creating logically aligned plans. Overall, SG-WORLD V2 offers a clear and dependable semantic planning method that can support future systems for generating video and audio while improving consistency and alignment between different modes.

23 April 2026 in ScienceOpen Preprints
Verified
Preprint
Open access
TBTrina Banerjee
Trina Banerjee
PNPiyush NA
Piyush NA
MVMukthikka V
Mukthikka V
GSGurpreet Singh
Gurpreet Singh
The rapid evolution of cyber threats has necessitated increasingly sophisticated detection methodologies that transcend traditional signature-based approaches. In recent years, Convolutional Neural Networks (CNNs) and multi-modal deep learning architectures have emerged as powerful paradigms for addressing diverse cybersecurity challenges, ranging from malware detection and phishing identification to network intrusion detection and encrypted traffic classification. This paper presents a comprehensive review of 50 recent research works (20222025) that leverage CNN-based and multi-modal learning techniques in the cybersecurity domain. We systematically categorize these approaches into seven distinct groups: (1)~CNN-based malware visualization and detection, (2)~CNN-based phishing and web threat detection, (3)~multi-modal and hybrid intrusion detection systems, (4)~CNN-based IoT and botnet detection, (5)~encrypted and darknet traffic classification, (6)~adversarial and GAN-based cybersecurity methods, and (7)~multi-modal cybersecurity applications. For each category, we analyze core methodologies, architectural innovations, and performance benchmarks. Furthermore, we provide detailed comparisons of datasets, evaluation metrics, and identify critical gaps in current research. Our analysis reveals that while CNN-based visualization achieves remarkable accuracy ($>$98\%) in controlled environments, significant challenges remain in adversarial robustness, cross-domain generalization, real-time deployment, and the effective fusion of heterogeneous data modalities. We conclude by outlining open challenges and future research directions that can advance the field toward more resilient, adaptive, and deployable cybersecurity systems.