This paper introduces EdgeProfiler, a fast profiling framework designed for evaluating lightweight Large Language Models (LLMs) on edge systems. While LLMs offer remarkable capabilities in natural language understanding and generation, their high computational, memory, and power requirements often confine them to cloud environments. EdgeProfiler addresses these challenges by providing a systematic methodology for assessing LLM performance in resource-constrained edge settings. The framework profiles compact LLMs, including TinyLLaMA, Gemma3.1B, Llama3.2-1B, and DeepSeek-r1-1.5B, using aggressive quantization techniques and strict memory constraints. Analytical modeling is used to estimate latency, FLOPs, and energy consumption. The profiling reveals that 4-bit quantization reduces model memory usage by approximately 60-70%, while maintaining accuracy within 2-5% of full-precision baselines. Inference speeds are observed to improve by 2-3x compared to FP16 baselines across various edge devices. Power modeling estimates a 35-50% reduction in energy consumption for INT4 configurations, enabling practical deployment on hardware such as Raspberry Pi 4/5 and Jetson Orin Nano Super. Our findings emphasize the importance of efficient profiling tailored to lightweight LLMs in edge environments, balancing accuracy, energy efficiency, and computational feasibility.
Designing efficient neural networks for embedded devices is a critical challenge, particularly in applications requiring real-time performance, such as aerial imaging with drones and UAVs for emergency responses. In this work, we introduce TakuNet, a novel light-weight architecture which employs techniques such as depth-wise convolutions and an early downsampling stem to reduce computational complexity while maintaining high accuracy. It leverages dense connections for fast convergence during training and uses 16-bit floating-point precision for optimization on embedded hardware accelerators. Experimental evaluation on two public datasets shows that TakuNet achieves near-state-of-the-art accuracy in classifying aerial images of emergency situations, despite its minimal parameter count. Real-world tests on embedded devices, namely Jetson Orin Nano and Raspberry Pi, confirm TakuNet's efficiency, achieving more than 650 fps on the 15W Jetson board, making it suitable for real-time AI processing on resource-constrained platforms and advancing the applicability of drones in emergency scenarios. The code and implementation details are publicly released.
Since the creation of the Web, recommender systems (RSs) have been an indispensable mechanism in information filtering. State-of-the-art RSs primarily depend on categorical features, which ecoded by embedding vectors, resulting in excessively large embedding tables. To prevent over-parameterized embedding tables from harming scalability, both academia and industry have seen increasing efforts in compressing RS embeddings. However, despite the prosperity of lightweight embedding-based RSs (LERSs), a wide diversity is seen in evaluation protocols, resulting in obstacles when relating LERS performance to real-world usability. Moreover, despite the common goal of lightweight embeddings, LERSs are evaluated with a single choice between the two main recommendation tasks -- collaborative filtering and content-based recommendation. This lack of discussions on cross-task transferability hinders the development of unified, more scalable solutions. Motivated by these issues, this study investigates various LERSs' performance, efficiency, and cross-task transferability via a thorough benchmarking process. Additionally, we propose an efficient embedding compression method using magnitude pruning, which is an easy-to-deploy yet highly competitive baseline that outperforms various complex LERSs. Our study reveals the distinct performance of LERSs across the two tasks, shedding light on their effectiveness and generalizability. To support edge-based recommendations, we tested all LERSs on a Raspberry Pi 4, where the efficiency bottleneck is exposed. Finally, we conclude this paper with critical summaries of LERS performance, model selection suggestions, and underexplored challenges around LERSs for future research. To encourage future research, we publish source codes and artifacts at \href{this link}{https://github.com/chenxing1999/recsys-benchmark}.
Computer vision has become increasingly prevalent in solving real-world problems across diverse domains, including smart agriculture, fishery, and livestock management. These applications may not require processing many image frames per second, leading practitioners to use single board computers (SBCs). Although many lightweight networks have been developed for mobile/edge devices, they primarily target smartphones with more powerful processors and not SBCs with the low-end CPUs. This paper introduces a CNN-ViT hybrid network called SBCFormer, which achieves high accuracy and fast computation on such low-end CPUs. The hardware constraints of these CPUs make the Transformer's attention mechanism preferable to convolution. However, using attention on low-end CPUs presents a challenge: high-resolution internal feature maps demand excessive computational resources, but reducing their resolution results in the loss of local image details. SBCFormer introduces an architectural design to address this issue. As a result, SBCFormer achieves the highest trade-off between accuracy and speed on a Raspberry Pi 4 Model B with an ARM-Cortex A72 CPU. For the first time, it achieves an ImageNet-1K top-1 accuracy of around 80% at a speed of 1.0 frame/sec on the SBC. Code is available at https://github.com/xyongLu/SBCFormer.
Medical image segmentation relies heavily on large-scale deep learning models, such as UNet-based architectures. However, the real-world utility of such models is limited by their high computational requirements, which makes them impractical for resource-constrained environments such as primary care facilities and conflict zones. Furthermore, shifts in the imaging domain can render these models ineffective and even compromise patient safety if such errors go undetected. To address these challenges, we propose M3D-NCA, a novel methodology that leverages Neural Cellular Automata (NCA) segmentation for 3D medical images using n-level patchification. Moreover, we exploit the variance in M3D-NCA to develop a novel quality metric which can automatically detect errors in the segmentation process of NCAs. M3D-NCA outperforms the two magnitudes larger UNet models in hippocampus and prostate segmentation by 2% Dice and can be run on a Raspberry Pi 4 Model B (2GB RAM). This highlights the potential of M3D-NCA as an effective and efficient alternative for medical image segmentation in resource-constrained environments.
In this research work, we have proposed a thermal tiny-YOLO multi-class object detection (TTYMOD) system as a smart forward sensing system that should remain effective in all weather and harsh environmental conditions using an end-to-end YOLO deep learning framework. It provides enhanced safety and improved awareness features for driver assistance. The system is trained on large-scale thermal public datasets as well as newly gathered novel open-sourced dataset comprising of more than 35,000 distinct thermal frames. For optimal training and convergence of YOLO-v5 tiny network variant on thermal data, we have employed different optimizers which include stochastic decent gradient (SGD), Adam, and its variant AdamW which has an improved implementation of weight decay. The performance of thermally tuned tiny architecture is further evaluated on the public as well as locally gathered test data in diversified and challenging weather and environmental conditions. The efficacy of a thermally tuned nano network is quantified using various qualitative metrics which include mean average precision, frames per second rate, and average inference time. Experimental outcomes show that the network achieved the best mAP of 56.4% with an average inference time/ frame of 4 milliseconds. The study further incorporates optimization of tiny network variant using the TensorFlow Lite quantization tool this is beneficial for the deployment of deep learning architectures on the edge and mobile devices. For this study, we have used a raspberry pi 4 computing board for evaluating the real-time feasibility performance of an optimized version of the thermal object detection network for the automotive sensor suite. The source code, trained and optimized models and complete validation/ testing results are publicly available at https://github.com/MAli-Farooq/Thermal-YOLO-And-Model-Optimization-Using-TensorFlowLite.
Fine-tuning pre-trained language models (PLMs) achieves impressive performance on a range of downstream tasks, and their sizes have consequently been getting bigger. Since a different copy of the model is required for each task, this paradigm is infeasible for storage-constrained edge devices like mobile phones. In this paper, we propose SPARTAN, a parameter efficient (PE) and computationally fast architecture for edge devices that adds hierarchically organized sparse memory after each Transformer layer. SPARTAN freezes the PLM parameters and fine-tunes only its memory, thus significantly reducing storage costs by re-using the PLM backbone for different tasks. SPARTAN contains two levels of memory, with only a sparse subset of parents being chosen in the first level for each input, and children cells corresponding to those parents being used to compute an output representation. This sparsity combined with other architecture optimizations improves SPARTAN's throughput by over 90% during inference on a Raspberry Pi 4 when compared to PE baselines (adapters) while also outperforming the latter by 0.1 points on the GLUE benchmark. Further, it can be trained 34% faster in a few-shot setting, while performing within 0.9 points of adapters. Qualitative analysis shows that different parent cells in SPARTAN specialize in different topics, thus dividing responsibility efficiently.
Monocular depth estimation is an essential task in the computer vision community. While tremendous successful methods have obtained excellent results, most of them are computationally expensive and not applicable for real-time on-device inference. In this paper, we aim to address more practical applications of monocular depth estimation, where the solution should consider not only the precision but also the inference time on mobile devices. To this end, we first develop an end-to-end learning-based model with a tiny weight size (1.4MB) and a short inference time (27FPS on Raspberry Pi 4). Then, we propose a simple yet effective data augmentation strategy, called R2 crop, to boost the model performance. Moreover, we observe that the simple lightweight model trained with only one single loss term will suffer from performance bottleneck. To alleviate this issue, we adopt multiple loss terms to provide sufficient constraints during the training stage. Furthermore, with a simple dynamic re-weight strategy, we can avoid the time-consuming hyper-parameter choice of loss terms. Finally, we adopt the structure-aware distillation to further improve the model performance. Notably, our solution named LiteDepth ranks 2nd in the MAI&AIM2022 Monocular Depth Estimation Challenge}, with a si-RMSE of 0.311, an RMSE of 3.79, and the inference time is 37$ms$ tested on the Raspberry Pi 4. Notably, we provide the fastest solution to the challenge. Codes and models will be released at \url{https://github.com/zhyever/LiteDepth}.
With the continuous development of neural networks for computer vision tasks, more and more network architectures have achieved outstanding success. As one of the most advanced neural network architectures, DenseNet shortcuts all feature maps to solve the model depth problem. Although this network architecture has excellent accuracy with low parameters, it requires an excessive inference time. To solve this problem, HarDNet reduces the connections between the feature maps, making the remaining connections resemble harmonic waves. However, this compression method may result in a decrease in the model accuracy and an increase in the parameters and model size. This network architecture may reduce the memory access time, but its overall performance can still be improved. Therefore, we propose a new network architecture, ThreshNet, using a threshold mechanism to further optimize the connection method. Different numbers of connections for different convolution layers are discarded to accelerate the inference of the network. The proposed network has been evaluated with image classification using CIFAR 10 and SVHN datasets under platforms of NVIDIA RTX 3050 and Raspberry Pi 4. The experimental results show that, compared with HarDNet68, GhostNet, MobileNetV2, ShuffleNet, and EfficientNet, the inference time of the proposed ThreshNet79 is 5%, 9%, 10%, 18%, and 20% faster, respectively. The number of parameters of ThreshNet95 is 55% less than that of HarDNet85. The new model compression and model acceleration methods can speed up the inference time, enabling network models to operate on mobile devices.
Monitoring behaviour in smart homes using sensors can offer insights into changes in the independent ability and long-term health of residents. Passive Infrared motion sensors (PIRs) are standard, however may not accurately track the full duration of movement. They also require line-of-sight to detect motion which can restrict performance and ensures they must be visible to residents. Channel State Information (CSI) is a low cost, unintrusive form of radio sensing which can monitor movement but also offers opportunities to generate rich data. We have developed a novel, self-calibrating motion detection system which uses CSI data collected and processed on a stock Raspberry Pi 4. This system exploits the correlation between CSI frames, on which we perform variance analysis using our algorithm to accurately measure the full period of a resident's movement. We demonstrate the effectiveness of this approach in several real-world environments. Experiments conducted demonstrate that activity start and end time can be accurately detected for motion examples of different intensities at different locations.
Transformers are ubiquitous in Natural Language Processing (NLP) tasks, but they are difficult to be deployed on hardware due to the intensive computation. To enable low-latency inference on resource-constrained hardware platforms, we propose to design Hardware-Aware Transformers (HAT) with neural architecture search. We first construct a large design space with $\textit{arbitrary encoder-decoder attention}$ and $\textit{heterogeneous layers}$. Then we train a $\textit{SuperTransformer}$ that covers all candidates in the design space, and efficiently produces many $\textit{SubTransformers}$ with weight sharing. Finally, we perform an evolutionary search with a hardware latency constraint to find a specialized $\textit{SubTransformer}$ dedicated to run fast on the target hardware. Extensive experiments on four machine translation tasks demonstrate that HAT can discover efficient models for different hardware (CPU, GPU, IoT device). When running WMT'14 translation task on Raspberry Pi-4, HAT can achieve $\textbf{3}\times$ speedup, $\textbf{3.7}\times$ smaller size over baseline Transformer; $\textbf{2.7}\times$ speedup, $\textbf{3.6}\times$ smaller size over Evolved Transformer with $\textbf{12,041}\times$ less search cost and no performance loss. HAT code is https://github.com/mit-han-lab/hardware-aware-transformers.git