arXiv subjects

Uttej Kallakuri

Publications and source records attributed to Uttej Kallakuri.

3 recordsLinked to original sources

arXiv2025

Invited Paper: BitMedViT: Ternary-Quantized Vision Transformer for Medical AI Assistants on the Edge

Vision Transformers (ViTs) have demonstrated strong capabilities in interpreting complex medical imaging data. However, their significant computational and memory demands pose challenges for deployment in real-time, resource-constrained mobile and wearable devices used in clinical environments. We introduce, BiTMedViT, a new class of Edge ViTs serving as medical AI assistants that perform structured analysis of medical images directly on the edge. BiTMedViT utilizes ternary- quantized linear layers tailored for medical imaging and com- bines a training procedure with multi-query attention, preserving stability under ternary weights with low-precision activations. Furthermore, BiTMedViT employs task-aware distillation from a high-capacity teacher to recover accuracy lost due to extreme quantization. Lastly, we also present a pipeline that maps the ternarized ViTs to a custom CUDA kernel for efficient memory bandwidth utilization and latency reduction on the Jetson Orin Nano. Finally, BiTMedViT achieves 86% diagnostic accuracy (89% SOTA) on MedMNIST across 12 datasets, while reducing model size by 43x, memory traffic by 39x, and enabling 16.8 ms inference at an energy efficiency up to 41x that of SOTA models at 183.62 GOPs/J on the Orin Nano. Our results demonstrate a practical and scientifically grounded route for extreme-precision medical imaging ViTs deployable on the edge, narrowing the gap between algorithmic advances and deployable clinical tools.

eess.IV↗

arXiv2025

ATLASv2: LLM-Guided Adaptive Landmark Acquisition and Navigation on the Edge

Autonomous systems deployed on edge devices face significant challenges, including resource constraints, real-time processing demands, and adapting to dynamic environments. This work introduces ATLASv2, a novel system that integrates a fine-tuned TinyLLM, real-time object detection, and efficient path planning to enable hierarchical, multi-task navigation and manipulation all on the edge device, Jetson Nano. ATLASv2 dynamically expands its navigable landmarks by detecting and localizing objects in the environment which are saved to its internal knowledge base to be used for future task execution. We evaluate ATLASv2 in real-world environments, including a handcrafted home and office setting constructed with diverse objects and landmarks. Results show that ATLASv2 effectively interprets natural language instructions, decomposes them into low-level actions, and executes tasks with high success rates. By leveraging generative AI in a fully on-board framework, ATLASv2 achieves optimized resource utilization with minimal prompting latency and power consumption, bridging the gap between simulated environments and real-world applications.

cs.RO↗

arXiv2019

Minimizing Classification Energy of Binarized Neural Network Inference for Wearable Devices

In this paper, we propose a low-power hardware for efficient deployment of binarized neural networks (BNNs) that have been trained for physiological datasets. BNNs constrain weights and feature-map to 1 bit, can pack in as many 1-bit weights as the width of a memory entry provides, and can execute multiple multiply-accumulate (MAC) operations with one fused bit-wise xnor and population-count instruction over aligned packed entries. Our proposed hardware is scalable with the number of processing engines (PEs) and the memory width, both of which adjustable for the most energy efficient configuration given an application. We implement two real case studies including Physical Activity Monitoring and Stress Detection on our platform, and for each case study on the target platform, we seek the optimal PE and memory configurations. Our implementation results indicate that a configuration with a good choice of memory width and number of PEs can be optimized up to 4x and 2.5x in energy consumption respectively on Artix-7 FPGA and on 65nm CMOS ASIC implementation. We also show that, generally, wider memories make more efficient BNN processing hardware. To further reduce the energy, we introduce Pool-Skipping technique that can skip at least 25% of the operations that are accompanied by a Max-Pool layer in BNNs, leading to a total of 22% operation reduction in the Stress Detection case study. Compared to the related works using the same case studies on the same target platform and with the same classification accuracy, our hardware is respectively 4.5x and 250x more energy efficient for the Stress Detection on FPGA and Physical Activity Monitoring on ASIC, respectively.

eess.SP↗