A5034311510- Adversarial Robustness in Machine Learning
- Indoor and Outdoor Localization Technologies
- Anomaly Detection Techniques and Applications
- Speech Recognition and Synthesis
- Explainable Artificial Intelligence (XAI)
- Topic Modeling
- Advanced Neural Network Applications
- Gait Recognition and Analysis
- Infrastructure Maintenance and Monitoring
- Speech and dialogue systems
- Advanced Steganography and Watermarking Techniques
- Natural Language Processing Techniques
- Cryptography and Data Security
- Chaos-based Image/Signal Encryption
- Speech and Audio Processing
- Advanced Malware Detection Techniques
- Music and Audio Processing
- Underwater Vehicles and Communication Systems
- IoT-based Smart Home Systems
- Recommender Systems and Techniques
- ICT in Developing Communities
- Text and Document Classification Technologies
- Mobile Crowdsensing and Crowdsourcing
- Mobile Learning in Education
- Tactile and Sensory Interactions
Google (United States)
2021-2024
University of California, Los Angeles
2015-2019
UCLA Health
2017-2019
Embedded Systems (United States)
2018
University of South Carolina
2012-2013
Egypt-Japan University of Science and Technology
2012
Deep neural networks (DNNs) are vulnerable to adversarial examples, perturbations correctly classified examples which can cause the model misclassify. In image domain, these often be made virtually indistinguishable human perception, causing humans and state-of-the-art models disagree. However, in natural language small clearly perceptible, replacement of a single word drastically alter semantics document. Given challenges, we use black-box population-based optimization algorithm generate...
Deep neural networks have achieved near-human accuracy levels in various types of classification and prediction tasks including images, text, speech, video data. However, the continue to be treated mostly as black-box function approximators, mapping a given input output. The next step this human-machine evolutionary process - incorporating these into mission critical processes such medical diagnosis, planning control requires level trust association with machine Typically, statistical...
The existence of a worldwide indoor floorplans database can lead to significant growth in location-based applications, especially for environments. In this paper, we present CrowdInside: crowdsourcing-based system the automatic construction buildings floorplans. CrowdInside leverages smart phones sensors that are ubiquitously available with humans who use building automatically and transparently construct accurate motion traces. These traces generated based on novel technique reducing errors...
Speech is a common and effective way of communication between humans, modern consumer devices such as smartphones home hubs are equipped with deep learning based accurate automatic speech recognition to enable natural interaction humans machines. Recently, researchers have demonstrated powerful attacks against machine models that can fool them produceincorrect results. However, nearly all previous research in adversarial has focused on image object detection models. In this short paper, we...
Deep neural networks are vulnerable to adversarial examples, even in the black-box setting, where attacker is restricted solely query access. Existing approaches generating examples typically require a significant number of queries, either for training substitute network or performing gradient estimation. We introduce GenAttack, gradient-free optimization technique that uses genetic algorithms synthesizing setting. Our experiments on different datasets (MNIST, CIFAR-10, and ImageNet) show...
The state-of-art models for speech synthesis and voice conversion are capable of generating synthetic that is perceptually indistinguishable from bonafide human speech. These methods represent a threat to the automatic speaker verification (ASV) systems. Additionally, replay attacks where attacker uses previously recorded genuine also possible. We present our solution ASVSpoof2019 competition, which aims develop countermeasure systems distinguish between spoofing speeches. Our model inspired...
The mission of tracking a pedestrian is valuable for many applications including walking distance estimation the purpose pervasive healthcare, museum and shopping mall guides, locating emergency responders. In this paper, we show how accurate ubiquitous can be performed using only inertial sensors embedded in his/her mobile phone. Our work depends on performing dead reckoning to track user's movement. main challenge that needs addressed handling noise low cost quality cell phones. proposed...
Indoor localization using mobile sensors has gained momentum lately. Most of the current systems rely on an extensive calibration step to achieve high accuracy. We propose SemanticSLAM, a novel unsupervised indoor scheme that bypasses need for war-driving. SemanticSLAM leverages idea certain locations in environment have unique signature one or more phone sensors. Climbing stairs, example, distinct pattern phone's accelerometer; specific spot may experience unusual magnetic interference...
Our ability to synthesize sensory data that preserves specific statistical properties of the real has had tremendous implications on privacy and big analytics. The synthetic can be used as a substitute for selective segments - are sensitive user thus protecting resulting in improved However, increasingly adversarial roles taken by recipients such mobile apps, or other cloud-based analytics services, mandate data, addition preserving properties, should also "difficult" distinguish from data....
Deep neural networks (DNNs) are vulnerable to adversarial examples, perturbations correctly classified examples which can cause the model misclassify. In image domain, these often virtually indistinguishable human perception, causing humans and state-of-the-art models disagree. However, in natural language small clearly perceptible, replacement of a single word drastically alter semantics document. Given challenges, we use black-box population-based optimization algorithm generate...
Deep neural networks have achieved state-of-the-art performance on various tasks. However, lack of interpretability and transparency makes it easier for malicious attackers to inject trojan backdoor into the networks, which will make model behave abnormally when a sample with specific trigger is input. In this paper, we propose NeuronInspect, framework detect backdoors in deep via output explanation techniques. NeuronInspect first identifies existence attack targets by generating heatmap...
Mobile devices have become an essential part of our daily lives. By virtue both their increasing computing power and the recent progress made in AI, mobile evolved to act as intelligent assistants many tasks rather than a mere way making phone calls. However, popular commonly used tools frameworks for machine intelligence are still lacking ability make proper use available heterogeneous resources on devices. In this paper, we study benefits utilizing (CPU GPU) commodity android while running...
Cloud computing services provide a scalable solution for the storage and processing of images multimedia files. However, concerns about privacy risks prevent users from sharing their personal with third-party services. In this paper, we describe design implementation CryptoImg, library modular preserving image operations over encrypted images. By using homomorphic encryption, CryptoImg allows to delegate remote servers without any concerns. Currently, supports subset most frequently used...
Deep neural networks are vulnerable to adversarial examples, even in the black-box setting, where attacker is restricted solely query access. Existing approaches generating examples typically require a significant number of queries, either for training substitute network or performing gradient estimation. We introduce GenAttack, gradient-free optimization technique that uses genetic algorithms synthesizing setting. Our experiments on different datasets (MNIST, CIFAR-10, and ImageNet) show...
The existence of a worldwide indoor floorplans database can lead to significant growth in location-based applications, especially for environments. In this paper, we present CrowdInside: crowdsourcing-based system the automatic construction buildings floorplans. CrowdInside leverages smart phones sensors that are ubiquitously available with humans who use building automatically and transparently construct accurate motion traces. These traces generated based on novel technique reducing errors...
Although different interaction modalities have been proposed in the field of human-computer interface (HCI), only a few these techniques could reach end users because scalability and usability issues. Given popularity growing number IoT devices, selecting one out many devices becomes hurdle typical smarthome environment. Therefore, an easy-to-learn, scalable, non-intrusive modality has to be explored. In this paper, we propose pointing approach interact with as is arguably natural way for...
Traditional recommender systems such as matrix factorization methods have primarily focused on learning a shared dense embedding space to represent both items and user preferences. Subsequently, sequence models RNN, GRUs, and, recently, Transformers emerged excelled in the task of sequential recommendation. This requires understanding structure present users' historical interactions predict next item they may like. Building upon success Large Language Models (LLMs) variety tasks, researchers...
Mapping and navigation applications are now considered popular services for mobile phones users. However, despite the fact that people spend most of their time indoors, these still limited to indoor spaces due lack large-scale floorplan databases. In this paper, we present JustWalk: a crowd intelligence-based system automatic construction buildings floorplans. JustWalk employs participatory sensing approach using smartphones ubiquitously available with users who visit building automatically...
Situational understanding (SU) requires a combination of insight - the ability to accurately perceive an existing situation and foresight anticipate how may develop in future. SU involves information fusion as well model representation inference. Commonly, heterogenous data sources must be exploited process: often including both hard soft products. In coalition context, processing resources will also distributed subjected restrictions on sharing. It necessary for human loop processes,...
Our particular research in the Distributed Analytics and Information Science International Technology Alliance (DAIS ITA) is focused on "Anticipatory Situational Understanding for Coalitions". This paper takes concrete example of detecting predicting traffic congestion UK road transport network from existing generic sensing sources, such as real-time CCTV imagery video, which are publicly available this purpose. scenario has been chosen carefully we believe that a typical city, all data...
Personalized IoT adapt their behavior based on contextual information, such as user and location. Unfortunately, the fact that personalized to context opens a side-channel leaks private information about user. To end, we start by studying extent which malicious eavesdropper can monitor actions taken an system extract user's information. In particular, show two concrete instantiations (in of mobile phones smart homes) new category spyware refer Context-Aware Adaptation Based Spyware (SpyCon)....
As more and online search queries come from voice, automatic speech recognition becomes a key component to deliver relevant results. Errors introduced by (ASR) lead irrelevant results returned the user, thus causing user dissatisfaction. In this paper, we introduce an approach, "Mondegreen", correct voice in text space without depending on audio signals, which may not always be available due system constraints or privacy bandwidth (for example, some ASR systems run on-device) considerations....