arXiv:2407.03331v2 Announce Type: replace 
Abstract: Emerging Artificial Intelligence of Things (AIoT) applications desire online prediction using deep neural network (DNN) models on mobile devices. However, due to the movement of devices, unfamiliar test samples constantly appear, significantly affecting the prediction accuracy of a pre-trained DNN. In addition, unstable network connection calls for local model inference. In this paper, we propose a light-weight scheme, called Anole, to cope with the local DNN model inference on mobile devices. The core idea of Anole is to first establish an army of compact DNN models, and then adaptively select the model fitting the current test sample best for online inference. The key is to automatically identify model-friendly scenes for training scene-specific DNN models. To this end, we design a weakly-supervised scene representation learning algorithm by combining both human heuristics and feature similarity in separating scenes. Moreover, we further train a model classifier to predict the best-fit scene-specific DNN model for each test sample. We implement Anole on different types of mobile devices and conduct extensive trace-driven and real-world experiments based on unmanned aerial vehicles (UAVs). The results demonstrate that Anole outwits the method of using a versatile large DNN in terms of prediction accuracy (4.5% higher), response time (33.1% faster) and power consumption (45.1% lower).

تم اقتراح مخطط خفيف الوزن جديد يسمى Anole لتحسين الاستدلال عبر الإنترنت لنماذج الشبكات العصبية العميقة (DNN) على الأجهزة المحمولة، خاصة في سياق الذكاء الاصطناعي للأشياء (AIoT). يتناول هذا النهج التحديات التي تطرحها حركة الأجهزة والعينات الاختبارية غير المألوفة التي يمكن أن تؤثر سلبًا على دقة التنبؤ. يهدف Anole إلى اختيار نموذج DNN الأنسب للظروف الحالية للاختبار من خلال تحديد المشاهد المناسبة لتدريب النماذج.

Se ha propuesto un nuevo esquema ligero llamado Anole para mejorar la inferencia en línea de modelos de redes neuronales profundas (DNN) en dispositivos móviles, especialmente en el contexto de la Inteligencia Artificial de las Cosas (AIoT). Este enfoque aborda los desafíos planteados por el movimiento de los dispositivos y las muestras de prueba desconocidas que pueden degradar la precisión de las predicciones. Anole tiene como objetivo seleccionar de manera adaptativa el modelo DNN más adecuado para las condiciones de prueba actuales al identificar escenas favorables para el entrenamiento de…

Un nouveau schéma léger nommé Anole a été proposé pour améliorer l'inférence en ligne des modèles de réseaux de neurones profonds (DNN) sur les appareils mobiles, en particulier dans le contexte de l'Intelligence Artificielle des Objets (AIoT). Cette approche répond aux défis posés par le mouvement des appareils et les échantillons de test inconnus qui peuvent dégrader la précision des prédictions. Anole vise à sélectionner de manière adaptative le modèle DNN le plus adapté aux conditions de test actuelles en identifiant les scènes favorables aux modèles pour l'entraînement.

A new lightweight scheme named Anole has been proposed to enhance online inference of deep neural network (DNN) models on mobile devices, particularly in the context of the Artificial Intelligence of Things (AIoT). This approach addresses challenges posed by device movement and unfamiliar test samples that can degrade prediction accuracy. Anole aims to adaptively select the most suitable DNN model for current test conditions by identifying model-friendly scenes for training.

A Scene-aware Models Adaptation Scheme for Cross-scene Online Inference on Mobile Devices

arXiv:2512.08639v1 Announce Type: new 
Abstract: Aerial Vision-and-Language Navigation (VLN) aims to enable unmanned aerial vehicles (UAVs) to interpret natural language instructions and navigate complex urban environments using onboard visual observation. This task holds promise for real-world applications such as low-altitude inspection, search-and-rescue, and autonomous aerial delivery. Existing methods often rely on panoramic images, depth inputs, or odometry to support spatial reasoning and action planning. These requirements increase system cost and integration complexity, thus hindering practical deployment for lightweight UAVs. We present a unified aerial VLN framework that operates solely on egocentric monocular RGB observations and natural language instructions. The model formulates navigation as a next-token prediction problem, jointly optimizing spatial perception, trajectory reasoning, and action prediction through prompt-guided multi-task learning. Moreover, we propose a keyframe selection strategy to reduce visual redundancy by retaining semantically informative frames, along with an action merging and label reweighting mechanism that mitigates long-tailed supervision imbalance and facilitates stable multi-task co-training. Extensive experiments on the Aerial VLN benchmark validate the effectiveness of our method. Under the challenging monocular RGB-only setting, our model achieves strong results across both seen and unseen environments. It significantly outperforms existing RGB-only baselines and narrows the performance gap with state-of-the-art panoramic RGB-D counterparts. Comprehensive ablation studies further demonstrate the contribution of our task design and architectural choices.

تم تقديم إطار عمل جديد للتنقل الجوي باستخدام الرؤية واللغة (VLN) يمكّن الطائرات بدون طيار (UAV) من تفسير التعليمات باللغة الطبيعية والتنقل في البيئات الحضرية باستخدام ملاحظات RGB أحادية العين فقط. تبسط هذه الطريقة عملية التنقل من خلال تحسين الإدراك المكاني، والتفكير في المسار، وتوقع الإجراءات من خلال التعلم متعدد المهام الموجه بالمحفزات.

Se ha introducido un nuevo marco para la navegación aérea mediante visión y lenguaje (VLN), que permite a los vehículos aéreos no tripulados (UAV) interpretar instrucciones en lenguaje natural y navegar por entornos urbanos utilizando únicamente observaciones RGB monoculares egocéntricas. Este enfoque simplifica el proceso de navegación al optimizar la percepción espacial, el razonamiento de trayectorias y la predicción de acciones a través de un aprendizaje multitarea guiado por indicaciones.

Un nouveau cadre pour la navigation aérienne par vision et langage (VLN) a été introduit, permettant aux véhicules aériens sans pilote (UAV) d'interpréter des instructions en langage naturel et de naviguer dans des environnements urbains en utilisant uniquement des observations RGB monoculaires égocentriques. Cette approche simplifie le processus de navigation en optimisant la perception spatiale, le raisonnement de trajectoire et la prédiction d'action grâce à un apprentissage multitâche guidé par des invites.

A new framework for Aerial Vision-and-Language Navigation (VLN) has been introduced, enabling unmanned aerial vehicles (UAVs) to interpret natural language instructions and navigate urban environments using only egocentric monocular RGB observations. This approach simplifies the navigation process by optimizing spatial perception, trajectory reasoning, and action prediction through prompt-guided multi-task learning.

A Scene-aware Models Adaptation Scheme for Cross-scene Online Inference on Mobile Devices

Was this article worth reading? Share it

LucidQuery AI

Chattermate

Dyad