Brain-Computer Interfaces (BCIs), e.g., Neuralink, are evolving to enable seamless interaction and control across various domains, such as assistive systems and virtual environments. While conventional BCI applications treat users as control units that transmit commands to devices, our vision introduces a paradigm shift toward the novel concept of Human-as-a-Sensor (HaaS), where users themselves function as intelligent multimodal sensing agents. HaaS leverages BCIs to extract contextual spatial information from brain signals as individuals naturally interact with their environment. By decoding neural activity, HaaS complements the limitations of traditional sensors and offers unique advantages for spatially-aware systems. We explore a range of opportunities enabled by HaaS for enhancing spatial awareness and outline multi-disciplinary research challenges in realizing our vision. We also present a feasibility study on using HaaS to track human mobility, showcasing the promise of HaaS for powering future spatially-aware systems.

The need for accurate and scalable modeling of 5G and beyond networks has recently emerged, driven by the reliance on virtual network functions (VNFs) and network slicing. Unfortunately, traditional network modeling approaches fail to capture 5G network behavior since they disregard the domainspecific challenges of modeling 5G networks. We propose vChainNet, the first end-to-end network modeling framework that provides accurate, scalable, and generalizable per-VNF and slice-level modeling for 5G and beyond networks. vChainNet introduces a modular, sequence-to-sequence deep learning architecture that models each VNF independently and composes the per-VNF models for end-to-end slice delay prediction. Our system design overcomes key domain-specific challenges in modeling 5G networks, including the functional variability of VNFs, VNFs' stochastic behavior, and the scale introduced by network densification. To address these challenges, vChainNet tunes a lightweight model composed of over$\mathbf{9 5 \%}$fewer parameters than state-of-the-art models, fuses domain-specific manually-engineered features with automatic feature extraction, and optimizes a distribution-based loss function during model training. Our results show that vChainNet achieves an accuracy improvement up to 10.86 % compared to state-of-the-art traditional network models, while providing a speedup of up to 53.33 times over common packet-level simulators. Furthermore, vChainNet's reliance on tuning a lightweight model allows it to generalize to unseen VNF types without extra hyperparameter tuning efforts. These results demonstrate the accuracy, scalability, and generalization ability of vChainNet for modeling 5G and beyond networks.

Wastewater treatment plants (WWTPs) are among the most energy-intensive components of urban infrastructure and bear strict regulatory responsibilities for wastewater quality. These dual challenges, minimizing energy consumption and maintaining environmental compliance, are deeply interrelated and must be managed simultaneously to achieve sustainable plant operation. This study proposes a framework that comprises two customized components. The first component employs a voting ensemble model based on transformer architecture to predict energy consumption. It processes heterogeneous feature domains — including hydraulic, wastewater, and climatic variables — through parallel attention-driven streams. The outputs from these streams are then aggregated using a weighted voting mechanism to produce the final prediction. Second, a multitask Bidirectional Gated Recurrent Unit (Bi-GRU) forecasts wastewater quality indicators concurrently (ammonia, Biochemical Oxygen Demand (BOD), and Chemical Oxygen Demand (COD)), capturing shared temporal dependencies and reducing model complexity. A hybrid preprocessing strategy is applied, incorporating domain-aware outlier detection (z-score and Interquartile Range (IQR)), K-Nearest Neighbors (KNN) Imputation, and feature selection using Extreme Gradient Boosting (XGBoost). Experimental results showed that. The voting ensemble model achieved the best results for energy consumption prediction with 31.61 of Root Mean Squared Error (RMSE). The multitask Bi-GRU achieved the best results for wastewater quality indicators with RMSE at 6.1689, 48.0323, and 88.2214 for ammonia, BOD, and COD, respectively. This work is among the first to integrate transformer ensembles and multitask learning in a unified WWTP forecasting system. Simultaneously addressing energy efficiency and water quality assurance, this offers a practical, scalable, and intelligent decision-support tool for sustainable wastewater management.

Recent transportation mode detection systems propose leveraging signals from only the serving cell tower to ensure ubiquity and practical deployability across all phones. However, existing solutions employ limited statistical hand-engineered features and traditional machine learning classifiers, leading to low estimation accuracy.

Designing a globally ubiquitous positioning system that works seamlessly in all environments remains a critical area of ongoing research. While Global Navigation Satellite Systems (GNSS), such as GPS, are the predominant technology for global outdoor positioning, they fail in areas with physical obstructions (e.g., dense urban regions and indoors) and are prone to jamming. These limitations significantly restrict their global accessibility and dependability, highlighting the need for supplementary positioning technologies.

All available cellular-based floor identification systems require information from multiple cell towers simultaneously, a feature absent in almost all phones, thus constraining their practicality. To address this limitation, we propose CellFloor, the first floor identification system to achieve high accuracy using only the serving cell tower while being regulatory-compliant. Based on recent advances in NLP, CellFloor builds a domain-specific edge-deployed large language model to identify the floor given a sequence of serving tower signal measurements. Our novel NLP-inspired approach allows CellFloor to extract rich contextual patterns from the signals, overcoming the limited information available when only the serving tower is used. Moreover, CellFloor employs recent advances in deep generative models to improve robustness against serving tower signal variations, enhancing floor identification accuracy. Our extensive evaluation of CellFloor shows consistently re-markable accuracy on multiple real testbeds, where it accurately estimates the exact floor at least 99.49% of the time using only the serving tower. This accuracy is superior to state-of-the-art (SOTA) systems, even when they use all available towers. Furthermore, unlike CellFloor, most SOTA systems fail to meet regulatory requirements when restricted to using just the serving tower, which is the only information available from the majority of current phones in the market. CellFloor also maintains regulatory compliance under different challenging conditions, including using only 20% of the available training data.

Floor identification has gained much attention due to the increasing demand for indoor location-based services, especially prompt emergency response services. Leveraging Cellular signals for floor identification has been of recent interest due to the pervasiveness of cellular technology. However, all current systems rely on information from multiple cell towers concurrently, which is inaccessible in most phones and thus severely limits their deployability.

The need for a ubiquitous and accurate transportation mode estimation system has recently risen. Unfortunately, GPS-based and inertial sensor-based solutions lack this needed ubiquity and large-scale deployability, especially in developing countries. Thus, novel systems have proposed leveraging the more ubiquitous cellular technology. However, these systems either require cell tower locations or rely on information from multiple towers, which limits their deployability.We propose AutoSense, a ubiquitous and easily deployable transportation mode estimation system that works on all phones by relying on handover and received signal strength (RSS) information from only the serving cell tower. AutoSense offers a novel domain-specific deep learning-based system to perform automatic feature extraction and time-series processing. Our system handles several challenges, including limitations in cellular data, lack of location information, overfitting, and information decay in long-term dependencies. We extensively evaluate AutoSense using a real-world public dataset composed of 395 hours of data collected over seven months. Our results show that, compared to state-of-the-art systems, AutoSense can achieve enhancements in average precision and recall of 12.36% and 14.93%, respectively, while providing a highly ubiquitous and deployable solution using only the serving cell tower information.