Unseen Devices Research Articles

Universal Fingerprint Generation: Controllable Diffusion Model with Multimodal Conditions.

The utilization of synthetic data for fingerprint recognition has garnered increased attention due to its potential to alleviate privacy concerns surrounding sensitive biometric data. However, current methods for generating fingerprints have limitations in creating impressions of the same finger with useful intra-class variations. To tackle this challenge, we present GenPrint, a framework to produce fingerprint images of various types while maintaining identity and offering humanly understandable control over different appearance factors such as fingerprint class, acquisition type, sensor device, and quality level. Unlike previous fingerprint generation approaches, GenPrint is not confined to replicating style characteristics from the training dataset alone: it enables the generation of novel styles from unseen devices without requiring additional fine-tuning. To accomplish these objectives, we developed GenPrint using latent diffusion models with multimodal conditions (text and image) for consistent generation of style and identity. Our experiments leverage a variety of publicly available datasets for training and evaluation. Results demonstrate the benefits of GenPrint in terms of identity preservation, explainable control, and universality of generated images. Importantly, the GenPrint-generated images yield comparable or even superior accuracy to models trained solely on real data and further enhances performance when augmenting the diversity of existing real fingerprint datasets.

Semi-physical simulation-driven contrastive decoupling net for intelligent fault diagnosis of unseen machines under varying speed

Multi-source domain generalization provides a good idea for intelligent fault diagnosis of unseen machines. However, the acquisition of multi-source training sets would not only affect the generalization effect, but also cause huge economic loss. Too much difference between source training sets would lead to negative transfer, while too much similarity between the sets would cause zero transfer. Simultaneously, the current methods based on the assumption of stable speed cannot realize intelligent diagnosis under varying speed. Therefore, a semi-physical simulation-driven contrastive decoupling network is presented. First, a phenomenological model is used to generate large amount of simulation data with different modulation frequencies and signal-to-noise ratios to provide support for the training of the net. Next, with the joint help of simulation data and physical data, a contrastive framework is introduced to extract domain-independent features. Finally, a decoupling framework is proposed to minimize speed-related information while maximizing health-related features, thereby enabling intelligent diagnosis of unseen devices at varying speed. The proposed method is tested on two cases operating under sharp speed variation and compared with some state-of-the-art methods. The results show the superiority of the proposed net in intelligent diagnosis of unseen equipment under drastically varying speed.

DGL: Device Generic Latency Model for Neural Architecture Search on Mobile Devices

The low-cost Neural Architecture Search (NAS) for lightweight networks working on massive mobile devices is essential for fast-developing ICT technology. Current NAS work can not search on unseen devices without latency sampling, which is a big obstacle to the implementation of NAS on mobile devices. In this paper, we overcome this challenge by proposing the Device Generic Latency (DGL) model. By absorbing processor modeling technology, the proposed DGL formula maps the parameters in the interval theory to the seven static configuration parameters of the device. And to make the formula more practical, we refine it to low-cost form by decreasing the number of configuration parameters to four. Then based on this formula, the DGL model is proposed which introduces the network parameters predictor and accuracy predictor to work with the DGL formula to predict the network latency. We propose the DGL-based NAS framework to enable fast searches without latency sampling. Extensive experiments results validate that the DGL model can achieve more accurate latency predictions than existing NAS latency predictors on unseen mobile devices. When configured with current state-of-the-art predictors, DGL-based NAS can search for architectures with higher accuracy that meet the latency limit than other NAS implementations, while using less training time and prediction time. Our work shed light on how to adopt domain knowledge into NAS topic and play important role in low-cost NAS on mobile devices.

Assessing Performance and Limitations of a Device‐Level Machine Learning Approach for Perovskite Solar Cells with an Application to Hole Transport Materials

Machine learning models have become widespread in materials science research. An open‐access and community‐driven database containing over 40 000 perovskite photovoltaic devices has been recently published. This resource enables the application of predictive data‐driven models to correlate device structure with photovoltaic performance, whereas the literature usually focuses on specific device layers. Herein, the concept of device‐level performance prediction is explored using gradient‐boosted regression trees as the core algorithm and Shapley values analysis to interpret and rationalize the results. The main pitfalls and conceptual limitations of the approach are discussed and correlated with the database structure and dimension, by comparing the performance of different choices of descriptors and dataset size. Evidence suggests that the additional features introduced herein, in particular chemical descriptors of perovskite additives, can boost regression performance at a device level. A specific model is finally trained to predict the performance of unseen devices and tested on experimental data from the literature. This task is found to be particularly challenging, as the ability of the model to generalize to a new chemical space is limited by several factors, including the amount and the quality of available data.

Multi-level distance embedding learning for robust acoustic scene classification with unseen devices

Multi-level distance embedding learning for robust acoustic scene classification with unseen devices

Frequency-based CNN and attention module for acoustic scene classification

Acoustic scene classification (ASC) is an audio classification task that identifies the environment in which sounds are recorded. Audio-related machine learning algorithms suffer from the device mismatch problem; that is, when trained from audio data recorded from one device, the algorithms cannot generalize to audio samples recorded using another device. In this study, a novel convolutional neural network, called a frequency-aware convolutional neural network (FACNN), is introduced to solve the device mismatch problem by focusing on the frequency information of the audio samples. Furthermore, an attention module, called the frequency attention network (FANet), is introduced to generate an attention map based on the frequency information of the input feature maps. FANet helps the FACNN to focus on the important frequency information, thus improving performance. The proposed method is trained on the TAU Urban Acoustic Scenes 2019 Mobile development dataset and TAU Urban Acoustic Scenes 2020 Mobile development dataset. The proposed method achieves a state-of-the-art accuracy of 75.99% in the TAU Urban Acoustic Scenes 2019 Mobile development dataset and a competitive result of 72.6% in the TAU Urban Acoustic Scenes 2020 Mobile development dataset. In addition, a comparison of FANet with the convolutional block attention module (CBAM) and the squeeze-and-excitation network (SENet) was performed. The results show that FANet can mitigate the device mismatch problem by improving the performance of the unseen devices.

Analyzing the Feasibility and Generalizability of Fingerprinting Internet of Things Devices

Abstract In recent years, we have seen rapid growth in the use and adoption of Internet of Things (IoT) devices. However, some loT devices are sensitive in nature, and simply knowing what devices a user owns can have security and privacy implications. Researchers have, therefore, looked at fingerprinting loT devices and their activities from encrypted network traffic. In this paper, we analyze the feasibility of fingerprinting IoT devices and evaluate the robustness of such fingerprinting approach across multiple independent datasets — collected under different settings. We show that not only is it possible to effectively fingerprint 188 loT devices (with over 97% accuracy), but also to do so even with multiple instances of the same make-and-model device. We also analyze the extent to which temporal, spatial and data-collection-methodology differences impact fingerprinting accuracy. Our analysis sheds light on features that are more robust against varying conditions. Lastly, we comprehensively analyze the performance of our approach under an open-world setting and propose ways in which an adversary can enhance their odds of inferring additional information about unseen devices (e.g., similar devices manufactured by the same company).

Towards Scalable and Channel-Robust Radio Frequency Fingerprint Identification for LoRa

Radio frequency fingerprint identification (RFFI) is a promising device authentication technique based on transmitter hardware impairments. The device-specific hardware features can be extracted at the receiver by analyzing the received signal and used for authentication. In this paper, we propose a scalable and channel-robust RFFI framework achieved by deep learning powered radio frequency fingerprint (RFF) extractor and channel independent features. Specifically, we leverage deep metric learning to train an RFF extractor, which has excellent generalization ability and can extract RFFs from previously unseen devices. Any devices can be enrolled via the pre-trained RFF extractor and the RFF database can be maintained efficiently for allowing devices to join and leave. Wireless channel impacts the RFF extraction and is tackled by exploiting channel independent features and data augmentation. We carried out extensive experimental evaluation involving 60 commercial off-the-shelf LoRa devices and a USRP N210 software defined radio platform. The results have successfully demonstrated that our framework can achieve excellent generalization abilities for rogue device detection and device classification as well as effective channel mitigation.

B-PO04-036 INNOVATIVE ARTIFICIAL INTELLIGENCE APPLICATION REDUCES FALSE PAUSE ALERTS WHILE MAINTAINING PERFECT TRUE PAUSE ALERT SENSITIVITY FOR INSERTABLE CARDIAC MONITORS

Insertable Cardiac Monitors (ICMs) have a high sensitivity for detecting Pause episodes but also produce false alerts and unnecessary clinic burden. Computation and longevity constrain in-the-can algorithm improvement. A cloud-based artificial intelligence (AI) algorithm may dramatically enhance performance. To train, test and validate a new AI designed specifically for ICM detected Pause episodes, including sinus asystole and AV block longer than preprogramed intervals (e.g., ≥ 3 or 5 sec). A novel 5-member ensemble Neural Network (NN) was developed in TensorFlow to classify multiple types of arrhythmias with a focus on Pause. The ensemble consists of 167,295 trainable parameters. The custom ensemble NN was trained and tested on 24,713 Pause episodes from 4,804 LINQ patients (pt). The data was curated, debiased and cleaned by expert scientists. This NN development resulted in low generalization errors on rare true Pauses. The overall AI bypassed the NN for episodes that were infrequent (≤1/month) or correlated to patient symptoms to ensure preservation. A validation set was created using episodes from 3,416 LINQ II pts implanted from Jul - Nov 2020. All ICM episodes were independently adjudicated by up to 3 reviewers. The endpoints were episode and daily relative sensitivity and specificity, because alerts are triggered on all episodes in a day. The validation set had 382 pts (mean age: 71, history of syncope: 43%) with 2,766 Pause episodes and 1,713 Pause alerted days; adjudicated True positives: 122 pts with 674 Pause episodes (24%) and 392 alerted days (∼23%). AI episode relative specificity and sensitivity were 98.3% (2057/2092) and 99.7% (672/674), respectively. Episode PPV was improved to 95.0% (672/707). False alerts were reduced by 97.4% (1286/1321) while maintaining 100% (392/392) of True alerts. The pt diagnostic yield was 100% (122/122). The episode sensitivity and specificity increased from test to the validation set (+0.5% and +0.4%, respectively), which means the AI generalized across unseen devices extremely well. The Pause AI significantly reduced false alerts while maintaining a perfect alert sensitivity and therefore implementation should ease clinic episode review burden.

Towards label-free 3D segmentation of optical coherence tomography images of the optic nerve head using deep learning.

Recently proposed deep learning (DL) algorithms for the segmentation of optical coherence tomography (OCT) images to quantify the morphological changes to the optic nerve head (ONH) tissues during glaucoma have limited clinical adoption due to their device specific nature and the difficulty in preparing manual segmentations (training data). We propose a DL-based 3D segmentation framework that is easily translatable across OCT devices in a label-free manner (i.e. without the need to manually re-segment data for each device). Specifically, we developed 2 sets of DL networks: the 'enhancer' (enhance OCT image quality and harmonize image characteristics from 3 devices) and the 'ONH-Net' (3D segmentation of 6 ONH tissues). We found that only when the 'enhancer' was used to preprocess the OCT images, the 'ONH-Net' trained on any of the 3 devices successfully segmented ONH tissues from the other two unseen devices with high performance (Dice coefficients > 0.92). We demonstrate that is possible to automatically segment OCT images from new devices without ever needing manual segmentation data from them.

RemNet: remnant convolutional neural network for camera model identification

Camera model identification (CMI) has gained significant importance in image forensics as digitally altered images are becoming increasingly commonplace. In this paper, a novel convolutional neural network (CNN) architecture is proposed for CMI with emphasis given on the preprocessing task considered to be inevitable for removing the scene content that heavily obscures the camera model fingerprints. Unlike the conventional approaches where fixed filters are used for preprocessing, the proposed remnant blocks, when coupled with a classification block and trained end-to-end minimizing the classification loss, learn to suppress the unnecessary image contents dynamically. This helps the classification block extract more robust camera model-specific features for CMI from the remnant of the image. The whole network, called RemNet, consisting of a preprocessing block and a shallow classification block, when trained on 18 models from the Dresden database, shows 100% accuracy for 16 camera models with an overall accuracy of 97.59% on test images from unseen devices, outperforming the state-of-the-art deep CNNs used in CMI. Furthermore, the proposed remnant blocks, when cascaded with the existing deep CNNs, e.g., ResNet, DenseNet, boost their performances by a large margin. The proposed approach proves to be very robust in identifying the source camera models, even if the original images are post-processed. It also achieves an overall accuracy of 95.11% on the IEEE Signal Processing Cup 2018 dataset, which indicates its generalizability.

The invisible robots of global finance

One of the barriers for doing ethics of technology in the domain of finance is that financial technologies usually remain invisible. These hidden and unseen devices, machines, and infrastructures have to be revealed. This paper shows how the "robots" of finance, which function as distance technologies, are not only themselves invisible, but also hide people and places, which is ethically and politically problematic. Furthermore, "the market" appears as a ghostly artificial agent, again rendering humans invisible and making it difficult to ascribe responsibility. Epistemic invisibilities thus become moral invisibilities. Finally, if we want to render finance more socially and ethically responsible, we also have to reveal the hidden efforts of many individuals and communities to re-invent finance by means of alternative financial practices and technologies. Research on responsible innovation should also consider less visible innovation that happens outside academia and industry.

GTID: A Technique for Physical Device <italic>and</italic> Device Type Fingerprinting

In this paper, we introduce GTID, a technique that can actively and passively fingerprint wireless devices and their types using wire-side observations in a local network. GTID exploits information that is leaked as a result of heterogeneity in devices, which is a function of different device hardware compositions and variations in devices’ clock skew. We apply statistical techniques on network traffic to create unique, reproducible device and device type signatures, and use artificial neural networks (ANNs) for classification. We demonstrate the efficacy of our technique on both an isolated testbed and a live campus network (during peak hours) using a corpus of 37 devices representing a wide range of device classes (e.g., iPads, iPhones, Google Phones, etc.) and traffic types (e.g., Skype, SCP, ICMP, etc.). Our experiments provided more than 300 GB of traffic captures which we used for ANN training and performance evaluation. In order for any fingerprinting technique to be practical, it must be able to detect previously unseen devices (i.e., devices for which no stored signature is available) and must be able to withstand various attacks. GTID is a fingerprinting technique to detect previously unseen devices and to illustrate its resilience under various attacker models. We measure the performance of GTID by considering accuracy, recall, and processing time and also illustrate how it can be used to complement existing security mechanisms (e.g., authentication systems) and to detect counterfeit devices.