Unlocking the Edge Deployment and On-Device Acceleration of Multi-LoRA Enabled One-for-All Foundational LLM
ACL 2026 Findings of the Association for Computational Linguistics
Research
Training- and inference-time efficiency for vision and language models — from GPU-scale inference to on-device compression.
I work on making models cheaper to train and serve without giving up quality — currently inference optimization on GPUs at Nava, and previously on-device compression at Samsung Research. Below are my publications, patents and thesis. Google Scholar →
Publications
TOC-SR: Task-Optimal Compact Diffusion for Image Super Resolution
ECCV 2026 W Oral LoViF Workshop, European Conference on Computer Vision
NanoSD: Edge Efficient Foundation Model for Real Time Image Restoration
CVPR 2026 IEEE/CVF Conference on Computer Vision and Pattern Recognition
EdgeDiT: Hardware-Aware Diffusion Transformers for Efficient On-Device Image Generation
CVPR 2026 W Oral Mobile AI Workshop, CVPR
Quantization with Unified Adaptive Distillation to Enable Multi-LoRA based One-for-All Generative Vision Models on Edge
CVPR 2026 W Oral Mobile AI Workshop, CVPR
Hardware-aware Low Light Image Enhancement on Edge
CVPR 2026 W Spotlight Women in Computer Vision (WiCV) Workshop, CVPR
Mobile-friendly Image De-noising: Hardware Conscious Optimization for Edge Application
ICASSP 2025 IEEE International Conference on Acoustics, Speech and Signal Processing
Patents
Multi-objective Optimization for Balanced Inference-time Latency-efficient and Eviction-aware SWA (Mobile SWA)
Provisional specification filed
A dynamic sliding window for token-eviction that shrinks the KV cache of LLMs with negligible accuracy loss.
System and Method for Accelerating On-Device Large Language Model Inference
Complete specification filed
A self-speculative decoding technique that improves throughput and reduces latency for on-device LLM inference.