@article{Le_2021_TMC,
author={Le Ngu Nguyen and Stephan Sigg and Jari Lietzen and Rainhard Dieter Findling and Kalle Ruttik},
journal={IEEE Transactions on Mobile Computing},
title={Camouflage learning. Feature value obscuring ambient intelligence for constrained devices},
year={2021},
abstract={Ambient intelligence demands collaboration schemes for distributed constrained devices which are not only highly energy efficient in distributed sensing, processing and communication, but which also respect data privacy. Traditional algorithms for distributed processing suffer in Ambient intelligence domains either from limited data privacy, or from their excessive processing demands for constrained distributed devices. 
In this paper, we present Camouflage learning, a distributed machine learning scheme that obscures the trained model via probabilistic collaboration using physical-layer computation offloading and demonstrate the feasibility of the approach on backscatter communication prototypes and in comparison with Federated learning. We show that Camouflage learning is more energy efficient than traditional schemes and that it requires less communication overhead while reducing the computation load through physical-layer computation offloading. The scheme is synchronization-agnostic and thus appropriate for sharply constrained, synchronization-incapable devices. We demonstrate model training and inference on four distinct datasets and investigate the performance of the scheme with respect to communication range, impact of challenging communication environments, power consumption, and the backscatter hardware prototype.
},
issue_date = {July 2021},
publisher = {IEEE},
volume = { },
number = { },
pages = {1-17},
group = {ambience},
project = {abacus}
}

Ambient intelligence demands collaboration schemes for distributed constrained devices which are not only highly energy efficient in distributed sensing, processing and communication, but which also respect data privacy. Traditional algorithms for distributed processing suffer in Ambient intelligence domains either from limited data privacy, or from their excessive processing demands for constrained distributed devices. In this paper, we present Camouflage learning, a distributed machine learning scheme that obscures the trained model via probabilistic collaboration using physical-layer computation offloading and demonstrate the feasibility of the approach on backscatter communication prototypes and in comparison with Federated learning. We show that Camouflage learning is more energy efficient than traditional schemes and that it requires less communication overhead while reducing the computation load through physical-layer computation offloading. The scheme is synchronization-agnostic and thus appropriate for sharply constrained, synchronization-incapable devices. We demonstrate model training and inference on four distinct datasets and investigate the performance of the scheme with respect to communication range, impact of challenging communication environments, power consumption, and the backscatter hardware prototype.

@inproceedings{Beck2020BCGECG,
  title={BCG and ECG-based secure communication for medical devices in Body Area Networks},
  author={Nils Beck and Si Zuo and Stephan Sigg},
  booktitle={The 19th International Conference on Pervasive Computing and Communications (PerCom 2021), adjunct},
  year={2021},
  abstract={An increasing amount of medical devices, such as pace makers or insulin pumps, is able to communicate in wireless Body Area Networks (BANs). While this facilitates interaction between users and medical devices, something that was previously more complicated or - in the case of implanted devices - often impossible, it also raises security and privacy questions. We exploit the wide availability of ballistocardiographs (BCG) and electocardiographs (ECG) in consumer wearables and propose MEDISCOM, an ad-hoc, implicit and secure communication protocol for medical devices in local BANs. Deriving common secret keys from a body’s BCG or ECG signal. MEDISCOM ensures confidentiality and integrity of sensitive medical data and also continuously authenticates devices, requiring no explicit user interaction and maintaining a low computational overhead. We consider relevant attack vectors and show how MEDISCOM is resilient towards them. Furthermore, we validate the security of the secret keys that our protocol derives on BCG and ECG data from 29 subjects.
},
  group = {ambience},
  project = {abacus}
}

An increasing amount of medical devices, such as pace makers or insulin pumps, is able to communicate in wireless Body Area Networks (BANs). While this facilitates interaction between users and medical devices, something that was previously more complicated or - in the case of implanted devices - often impossible, it also raises security and privacy questions. We exploit the wide availability of ballistocardiographs (BCG) and electocardiographs (ECG) in consumer wearables and propose MEDISCOM, an ad-hoc, implicit and secure communication protocol for medical devices in local BANs. Deriving common secret keys from a body’s BCG or ECG signal. MEDISCOM ensures confidentiality and integrity of sensitive medical data and also continuously authenticates devices, requiring no explicit user interaction and maintaining a low computational overhead. We consider relevant attack vectors and show how MEDISCOM is resilient towards them. Furthermore, we validate the security of the secret keys that our protocol derives on BCG and ECG data from 29 subjects.

@inproceedings{Sigg2020Camouflage,
  title={Camouflage Learning},
  author={Stephan Sigg and Le Ngu Nguyen and Jing Ma},
  booktitle={The 19th International Conference on Pervasive Computing and Communications (PerCom 2021), adjunct},
  year={2021},
  abstract={Federated learning has been proposed as a concept for distributed machine learning which enforces privacy by avoiding sharing private data with a coordinator or distributed nodes. Instead of gathering datasets to a central server for model training in traditional machine learning, in federated learning, model updates are computed locally at distributed devices and merged at a coordinator. However, information on local data might be leaked through the model updates. We propose Camouflage learning, a distributed machine learning scheme that distributes both the data and the model. Neither the distributed devices nor the coordinator is at any point in time in possession of the complete model. Furthermore, data and model are obfuscated during distributed model inference and distributed model training. Camouflage learning can be implemented with various Machine learning schemes.
},
  group = {ambience},
  project = {radiosense, abacus}
}

Federated learning has been proposed as a concept for distributed machine learning which enforces privacy by avoiding sharing private data with a coordinator or distributed nodes. Instead of gathering datasets to a central server for model training in traditional machine learning, in federated learning, model updates are computed locally at distributed devices and merged at a coordinator. However, information on local data might be leaked through the model updates. We propose Camouflage learning, a distributed machine learning scheme that distributes both the data and the model. Neither the distributed devices nor the coordinator is at any point in time in possession of the complete model. Furthermore, data and model are obfuscated during distributed model inference and distributed model training. Camouflage learning can be implemented with various Machine learning schemes.

@inproceedings{Manila2020BatteryLess,
  title={Towards battery-less RF sensing},
  author={Manila Kodali and Le Ngu Nguyen and Stephan Sigg},
  booktitle={The 19th International Conference on Pervasive Computing and Communications (PerCom 2021), WiP},
  year={2021},
  abstract={Recent work has demonstrated the use of the radio interface as a sensing modality for gestures, activities and situational perception. The field generally moves towards larger bandwidths, multiple antennas, and higher, mmWave frequency domains, which allow for the recognition of minute movements. We envision another set of applications for RF sensing: battery-less autonomous sensing devices. In this work, we investigate transceiver-less passive RF-sensors which are excited by the fluctuation of the received power over the wireless channel. In particular, we demonstrate the use of battery-less RF-sensing for applications of on-body gesture recognition integrated into smart garment, as well as the integration of such sensing capabilities into smart surfaces.
},
  group = {ambience},
  project = {radiosense,abacus}
}

Recent work has demonstrated the use of the radio interface as a sensing modality for gestures, activities and situational perception. The field generally moves towards larger bandwidths, multiple antennas, and higher, mmWave frequency domains, which allow for the recognition of minute movements. We envision another set of applications for RF sensing: battery-less autonomous sensing devices. In this work, we investigate transceiver-less passive RF-sensors which are excited by the fluctuation of the received power over the wireless channel. In particular, we demonstrate the use of battery-less RF-sensing for applications of on-body gesture recognition integrated into smart garment, as well as the integration of such sensing capabilities into smart surfaces.

@PhDThesis{LeThesis2020,
	author = "Le Ngu Nguyen",
	title = "Security from Implicit Information",
	school = "Aalto University",
	year = "2020",
	month = "September",
	isbn = "978-952-64-0013-6",
    url_Paper ={https://aaltodoc.aalto.fi/handle/123456789/46392},
    abstract = {We present novel security mechanisms using implicit information extracted from physiological, behavioural, and ambient data. These mechanisms are implemented with reference to device-to-user and inter-device relationships, including: user authentication with transient image-based passwords, device-to-device secure connection initialization based on vocal commands, collaborative inference over the communication channel, and continuous on-body device pairing.
 
Authentication methods based on passwords require users to explicitly set their passwords and change them regularly. We introduce a method to generate always-fresh authentication challenges from videos collected by wearable cameras. We implement two password formats that expect users to arrange or select images according to their chronological information.
 
Radio waves are mainly used for data transmission. We implement function computation over the wireless signals to perform collaborative inference. We encode information into burst sequences in such a way that arithmetic functions can be computed using the interference. Hence, data is hidden inside the wireless signals and implicitly aggregated. Our algorithms allow us to train and deploy a classifier efficiently with the support of minimal backscatter devices.
 
To initialize a connection between a personal device (e.g. smart-phone) and shared appliances (e.g. smart-screens), users are required to explicitly ask for connection information including device identities and PIN codes. We propose to leverage natural vocal commands to select shared appliance types and generate secure communication keys from the audio implicitly. We perform experiments to verify that device proximity defined by audio fingerprints can restrict the range of device-to-device communication.
 
PIN codes in device pairing must be manually entered or verified by users. This is inconvenient in scenarios when pairing is performed frequently or devices have limited user interfaces. Our methods generate secure pairing keys for on-body devices continuously from sensor data. Our mechanisms automatically disconnect the devices when they leave the user's body. To cover all human activities, we leverage gait in human ambulatory actions and heartbeat in resting postures.},
group = {ambience}, 
project = {abacus}}

We present novel security mechanisms using implicit information extracted from physiological, behavioural, and ambient data. These mechanisms are implemented with reference to device-to-user and inter-device relationships, including: user authentication with transient image-based passwords, device-to-device secure connection initialization based on vocal commands, collaborative inference over the communication channel, and continuous on-body device pairing. Authentication methods based on passwords require users to explicitly set their passwords and change them regularly. We introduce a method to generate always-fresh authentication challenges from videos collected by wearable cameras. We implement two password formats that expect users to arrange or select images according to their chronological information. Radio waves are mainly used for data transmission. We implement function computation over the wireless signals to perform collaborative inference. We encode information into burst sequences in such a way that arithmetic functions can be computed using the interference. Hence, data is hidden inside the wireless signals and implicitly aggregated. Our algorithms allow us to train and deploy a classifier efficiently with the support of minimal backscatter devices. To initialize a connection between a personal device (e.g. smart-phone) and shared appliances (e.g. smart-screens), users are required to explicitly ask for connection information including device identities and PIN codes. We propose to leverage natural vocal commands to select shared appliance types and generate secure communication keys from the audio implicitly. We perform experiments to verify that device proximity defined by audio fingerprints can restrict the range of device-to-device communication. PIN codes in device pairing must be manually entered or verified by users. This is inconvenient in scenarios when pairing is performed frequently or devices have limited user interfaces. Our methods generate secure pairing keys for on-body devices continuously from sensor data. Our mechanisms automatically disconnect the devices when they leave the user's body. To cover all human activities, we leverage gait in human ambulatory actions and heartbeat in resting postures.

@inproceedings{salami2020MQTT,
  title={A FAIR Extension for the MQTT Protocol},
  author={Dariush Salami and Olga Streibel and Stephan Sigg},
  booktitle={16th International Conference on Mobility, Sensing and Networking (MSN 2020) },
  year={2020},
  group = {ambience},
  project={abacus}
  
}

@article{Le_2019_multimedia,
author={Le Ngu Nguyen and Stephan Sigg},
journal={IEEE Multimedia Communications -- Frontiers, SI on Social and Mobile Connected Smart Objects},
title={Learning a Classification Model over Vertically-Partitioned Healthcare Data},
year={2019},
  project = {abacus},
group = {ambience}}

@InProceedings{Le_2019_hotsalsa,
author={Le Ngu Nguyen and Stephan Sigg},
title={Learning with Vertically Partitioned Data, Binary Feedback and Random Parameter update},
booktitle={Workshop on Hot Topics in Social and Mobile Connected Smart Objects, in conjuction with IEEE International Conference on Computer Communications (INFOCOM)},
year={2019},
  project = {abacus},
group = {ambience}}